arXiv Daily Digest - 2026-02-06

We introduce Wedge Sampling, a new non-adaptive sampling scheme for low-rank tensor completion. We study recovery of an order-$k$ low-rank tensor of dimension $n \times \cdots \times n$ from a subset of its entries. Unlike the standard uniform entry model (i.e., i.i.d. samples from $[n]^k$), wedge sampling allocates observations to structured length-two patterns (wedges) in an associated bipartite sampling graph. By directly promoting these length-two connections, the sampling design strengthens the spectral signal that underlies efficient initialization, in regimes where uniform sampling is too sparse to generate enough informative correlations. Our main result shows that this change in sampling paradigm enables polynomial-time algorithms to achieve both weak and exact recovery with nearly linear sample complexity in $n$. The approach is also plug-and-play: wedge-sampling-based spectral initialization can be combined with existing refinement procedures (e.g., spectral or gradient-based methods) using only an additional $\tilde{O}(n)$ uniformly sampled entries, substantially improving over the $\tilde{O}(n^{k/2})$ sample complexity typically required under uniform entry sampling for efficient methods. Overall, our results suggest that the statistical-to-computational gap highlighted in Barak and Moitra (2022) is, to a large extent, a consequence of the uniform entry sampling model for tensor completion, and that alternative non-adaptive measurement designs that guarantee a strong initialization can overcome this barrier.

While Group Relative Policy Optimization (GRPO) has emerged as a scalable framework for critic-free policy learning, extending it to settings with explicit behavioral constraints remains underexplored. We introduce Constrained GRPO, a Lagrangian-based extension of GRPO for constrained policy optimization. Constraints are specified via indicator cost functions, enabling direct optimization of violation rates through a Lagrangian relaxation. We show that a naive multi-component treatment in advantage estimation can break constrained learning: mismatched component-wise standard deviations distort the relative importance of the different objective terms, which in turn corrupts the Lagrangian signal and prevents meaningful constraint enforcement. We formally derive this effect to motivate our scalarized advantage construction that preserves the intended trade-off between reward and constraint terms. Experiments in a toy gridworld confirm the predicted optimization pathology and demonstrate that scalarizing advantages restores stable constraint control. In addition, we evaluate Constrained GRPO on robotics tasks, where it improves constraint satisfaction while increasing task success, establishing a simple and effective recipe for constrained policy optimization in embodied AI domains that increasingly rely on large multimodal foundation models.

Two-sample testing, where we aim to determine whether two distributions are equal or not equal based on samples from each one, is challenging if we cannot place assumptions on the properties of the two distributions. In particular, certifying equality of distributions, or even providing a tight upper bound on the total variation (TV) distance between the distributions, is impossible to achieve in a distribution-free regime. In this work, we examine the blurred TV distance, a relaxation of TV distance that enables us to perform inference without assumptions on the distributions. We provide theoretical guarantees for distribution-free upper and lower bounds on the blurred TV distance, and examine its properties in high dimensions.

Graph-structured data is ubiquitous and powerful in representing complex relationships in many online platforms. While graph neural networks (GNNs) are widely used to learn from such data, counterfactual graph learning has emerged as a promising approach to improve model interpretability. Counterfactual explanation research focuses on identifying a counterfactual graph that is similar to the original but leads to different predictions. These explanations optimize two objectives simultaneously: the sparsity of changes in the counterfactual graph and the validity of its predictions. Building on these qualitative optimization goals, this paper introduces CFRecs, a novel framework that transforms counterfactual explanations into actionable insights. CFRecs employs a two-stage architecture consisting of a graph neural network (GNN) and a graph variational auto-encoder (Graph-VAE) to strategically propose minimal yet high-impact changes in graph structure and node attributes to drive desirable outcomes in recommender systems. We apply CFRecs to Zillow's graph-structured data to deliver actionable recommendations for both home buyers and sellers with the goal of helping them navigate the competitive housing market and achieve their homeownership goals. Experimental results on Zillow's user-listing interaction data demonstrate the effectiveness of CFRecs, which also provides a fresh perspective on recommendations using counterfactual reasoning in graphs.

Sparse autoencoders (SAEs) have become a standard tool for mechanistic interpretability in autoregressive large language models (LLMs), enabling researchers to extract sparse, human-interpretable features and intervene on model behavior. Recently, as diffusion language models (DLMs) have become an increasingly promising alternative to the autoregressive LLMs, it is essential to develop tailored mechanistic interpretability tools for this emerging class of models. In this work, we present DLM-Scope, the first SAE-based interpretability framework for DLMs, and demonstrate that trained Top-K SAEs can faithfully extract interpretable features. Notably, we find that inserting SAEs affects DLMs differently than autoregressive LLMs: while SAE insertion in LLMs typically incurs a loss penalty, in DLMs it can reduce cross-entropy loss when applied to early layers, a phenomenon absent or markedly weaker in LLMs. Additionally, SAE features in DLMs enable more effective diffusion-time interventions, often outperforming LLM steering. Moreover, we pioneer certain new SAE-based research directions for DLMs: we show that SAEs can provide useful signals for DLM decoding order; and the SAE features are stable during the post-training phase of DLMs. Our work establishes a foundation for mechanistic interpretability in DLMs and shows a great potential of applying SAEs to DLM-related tasks and algorithms.

The rapid evolution of large language models (LLMs) has expanded their capabilities from basic dialogue to advanced scientific reasoning. However, existing benchmarks in biology often fail to assess a critical skill required of researchers: the ability to integrate experimental results with contextual knowledge to derive meaningful conclusions. To address this gap, we introduce BABE(Biology Arena BEnchmark), a comprehensive benchmark designed to evaluate the experimental reasoning capabilities of biological AI systems. BABE is uniquely constructed from peer-reviewed research papers and real-world biological studies, ensuring that tasks reflect the complexity and interdisciplinary nature of actual scientific inquiry. BABE challenges models to perform causal reasoning and cross-scale inference. Our benchmark provides a robust framework for assessing how well AI systems can reason like practicing scientists, offering a more authentic measure of their potential to contribute to biological research.

Reliable terrain perception is a critical prerequisite for the deployment of humanoid robots in unstructured, human-centric environments. While traditional systems often rely on manually engineered, single-sensor pipelines, this paper presents a learning-based framework that uses an intermediate, robot-centric heightmap representation. A hybrid Encoder-Decoder Structure (EDS) is introduced, utilizing a Convolutional Neural Network (CNN) for spatial feature extraction fused with a Gated Recurrent Unit (GRU) core for temporal consistency. The architecture integrates multimodal data from an Intel RealSense depth camera, a LIVOX MID-360 LiDAR processed via efficient spherical projection, and an onboard IMU. Quantitative results demonstrate that multimodal fusion improves reconstruction accuracy by 7.2% over depth-only and 9.9% over LiDAR-only configurations. Furthermore, the integration of a 3.2 s temporal context reduces mapping drift.

The quadratic complexity of attention mechanisms poses a critical bottleneck for large language models processing long contexts. While dynamic sparse attention methods offer input-adaptive efficiency, they face fundamental trade-offs: requiring preprocessing, lacking global evaluation, violating query independence, or incurring high computational overhead. We present RRAttention, a novel dynamic sparse attention method that simultaneously achieves all desirable properties through a head \underline{r}ound-\underline{r}obin (RR) sampling strategy. By rotating query sampling positions across attention heads within each stride, RRAttention maintains query independence while enabling efficient global pattern discovery with stride-level aggregation. Our method reduces complexity from $O(L^2)$ to $O(L^2/S^2)$ and employs adaptive Top-$τ$ selection for optimal sparsity. Extensive experiments on natural language understanding (HELMET) and multimodal video comprehension (Video-MME) demonstrate that RRAttention recovers over 99\% of full attention performance while computing only half of the attention blocks, achieving 2.4$\times$ speedup at 128K context length and outperforming existing dynamic sparse attention methods.

Community detection in networks is a fundamental problem in machine learning and statistical inference, with applications in social networks, biological systems, and communication networks. The stochastic block model (SBM) serves as a canonical framework for studying community structure, and exact recovery, identifying the true communities with high probability, is a central theoretical question. While classical results characterize the phase transition for exact recovery based solely on graph connectivity, many real-world networks contain additional data, such as node attributes or labels. In this work, we study exact recovery in the Data Block Model (DBM), an SBM augmented with node-associated data, as formalized by Asadi, Abbe, and Verdú (2017). We introduce the Chernoff--TV divergence and use it to characterize a sharp exact recovery threshold for the DBM. We further provide an efficient algorithm that achieves this threshold, along with a matching converse result showing impossibility below the threshold. Finally, simulations validate our findings and demonstrate the benefits of incorporating vertex data as side information in community detection.

Training a neural network requires navigating a high-dimensional, non-convex loss surface to find parameters that minimize this loss. In many ways, it is surprising that optimizers such as stochastic gradient descent and ADAM can reliably locate minima which perform well on both the training and test data. To understand the success of training, a "loss landscape" community has emerged to study the geometry of the loss function and the dynamics of optimization, often using visualization techniques. However, these loss landscape studies have mostly been limited to machine learning for image classification. In the newer field of physics-informed machine learning, little work has been conducted to visualize the landscapes of losses defined not by regression to large data sets, but by differential operators acting on state fields discretized by neural networks. In this work, we provide a comprehensive review of the loss landscape literature, as well as a discussion of the few existing physics-informed works which investigate the loss landscape. We then use a number of the techniques we survey to empirically investigate the landscapes defined by the Deep Ritz and squared residual forms of the physics loss function. We find that the loss landscapes of physics-informed neural networks have many of the same properties as the data-driven classification problems studied in the literature. Unexpectedly, we find that the two formulations of the physics loss often give rise to similar landscapes, which appear smooth, well-conditioned, and convex in the vicinity of the solution. The purpose of this work is to introduce the loss landscape perspective to the scientific machine learning community, compare the Deep Ritz and the strong form losses, and to challenge prevailing intuitions about the complexity of the loss landscapes of physics-informed networks.

DARWIN is an evolutionary GPT model, utilizing a genetic-algorithm like optimization structure with several independent GPT agents being trained individually using unique training code. Each iteration, the GPT models are prompted to modify the training code of one another in an attempt to improve their performance in a mutation-like manner, and the best GPT agents are then benchmarked and selected for the next iteration by genetic algorithm. For demonstration purposes and due to budget and time constraints, OpenAI API is used to prompt training code improvements and the nanoGPT framework is used as the training code. DARWIN also utilizes persistent JSON-based memory files to track previous reasoning and changes to code to correlate with improvement to model performance. and a bidirectional interface for HITL intervention allowing the model to request upgrades such as additional datasets, training scripts, and restructuring of file hierarchies. In experiments, DARWIN achieved a 1.26 percent improvement in model FLOPS utilization (MFU) and a 2.07 percent improvement to perplexity in 5 iterations of training over baseline configurations, demonstrating promising capabilities as a foundation for scaling evolutionary GPT training.

While humans perceive the world through diverse modalities that operate synergistically to support a holistic understanding of their surroundings, existing omnivideo models still face substantial challenges on audio-visual understanding tasks. In this paper, we propose OmniVideo-R1, a novel reinforced framework that improves mixed-modality reasoning. OmniVideo-R1 empowers models to "think with omnimodal cues" by two key strategies: (1) query-intensive grounding based on self-supervised learning paradigms; and (2) modality-attentive fusion built upon contrastive learning paradigms. Extensive experiments on multiple benchmarks demonstrate that OmniVideo-R1 consistently outperforms strong baselines, highlighting its effectiveness and robust generalization capabilities.

In this work, we provide a sharp theory of scaling laws for two-layer neural networks trained on a class of hierarchical multi-index targets, in a genuinely representation-limited regime. We derive exact information-theoretic scaling laws for subspace recovery and prediction error, revealing how the hierarchical features of the target are sequentially learned through a cascade of phase transitions. We further show that these optimal rates are achieved by a simple, target-agnostic spectral estimator, which can be interpreted as the small learning-rate limit of gradient descent on the first-layer weights. Once an adapted representation is identified, the readout can be learned statistically optimally, using an efficient procedure. As a consequence, we provide a unified and rigorous explanation of scaling laws, plateau phenomena, and spectral structure in shallow neural networks trained on such hierarchical targets.

The rapid advancement of Large Language Models (LLMs) has catalyzed the development of autonomous agents capable of navigating complex environments. However, existing evaluations primarily adopt a deductive paradigm, where agents execute tasks based on explicitly provided rules and static goals, often within limited planning horizons. Crucially, this neglects the inductive necessity for agents to discover latent transition laws from experience autonomously, which is the cornerstone for enabling agentic foresight and sustaining strategic coherence. To bridge this gap, we introduce OdysseyArena, which re-centers agent evaluation on long-horizon, active, and inductive interactions. We formalize and instantiate four primitives, translating abstract transition dynamics into concrete interactive environments. Building upon this, we establish OdysseyArena-Lite for standardized benchmarking, providing a set of 120 tasks to measure an agent's inductive efficiency and long-horizon discovery. Pushing further, we introduce OdysseyArena-Challenge to stress-test agent stability across extreme interaction horizons (e.g., > 200 steps). Extensive experiments on 15+ leading LLMs reveal that even frontier models exhibit a deficiency in inductive scenarios, identifying a critical bottleneck in the pursuit of autonomous discovery in complex environments. Our code and data are available at https://github.com/xufangzhi/Odyssey-Arena

Large language models (LLMs) have achieved strong performance in language-centric tasks. However, in agentic settings, LLMs often struggle to anticipate action consequences and adapt to environment dynamics, highlighting the need for world-modeling capabilities in LLM-based agents. We propose Reinforcement World Model Learning (RWML), a self-supervised method that learns action-conditioned world models for LLM-based agents on textual states using sim-to-real gap rewards. Our method aligns simulated next states produced by the model with realized next states observed from the environment, encouraging consistency between internal world simulations and actual environment dynamics in a pre-trained embedding space. Unlike next-state token prediction, which prioritizes token-level fidelity (i.e., reproducing exact wording) over semantic equivalence and can lead to model collapse, our method provides a more robust training signal and is empirically less susceptible to reward hacking than LLM-as-a-judge. We evaluate our method on ALFWorld and $τ^2$ Bench and observe significant gains over the base model, despite being entirely self-supervised. When combined with task-success rewards, our method outperforms direct task-success reward RL by 6.9 and 5.7 points on ALFWorld and $τ^2$ Bench respectively, while matching the performance of expert-data training.

Data is the lifeblood of AI, yet much of the most valuable data remains locked in silos due to privacy and regulations. As a result, AI remains heavily underutilized in many of the most important domains, including healthcare, education, and finance. Synthetic data generation (SDG), i.e. the generation of artificial data with a synthesizer trained on real data, offers an appealing solution to make data available while mitigating privacy concerns, however existing SDG-as-a-service workflow require data holders to trust providers with access to private data.We propose FHAIM, the first fully homomorphic encryption (FHE) framework for training a marginal-based synthetic data generator on encrypted tabular data. FHAIM adapts the widely used AIM algorithm to the FHE setting using novel FHE protocols, ensuring that the private data remains encrypted throughout and is released only with differential privacy guarantees. Our empirical analysis show that FHAIM preserves the performance of AIM while maintaining feasible runtimes.

There is a need for synthetic training and test datasets that replicate statistical distributions of original datasets without compromising their confidentiality. A lot of research has been done in leveraging Generative Adversarial Networks (GANs) for synthetic data generation. However, the resulting models are either not accurate enough or are still vulnerable to membership inference attacks (MIA) or dataset reconstruction attacks since the original data has been leveraged in the training process. In this paper, we explore the feasibility of producing a synthetic test dataset with the same statistical properties as the original one, with only indirectly leveraging the original data in the generation process. The approach is inspired by GANs, with a generation step and a discrimination step. However, in our approach, we use a test generator (a fuzzer) to produce test data from an input specification, preserving constraints set by the original data; a discriminator model determines how close we are to the original data. By evolving samples and determining "good samples" with the discriminator, we can generate privacy-preserving data that follows the same statistical distributions are the original dataset, leading to a similar utility as the original data. We evaluated our approach on four datasets that have been used to evaluate the state-of-the-art techniques. Our experiments highlight the potential of our approach towards generating synthetic datasets that have high utility while preserving privacy.

Floating-point neural networks dominate modern machine learning but incur substantial inference cost, motivating interest in Boolean networks for resource-constrained settings. However, learning compact and accurate Boolean networks is challenging due to their combinatorial nature. In this work, we address this challenge from three different angles: learned connections, compact convolutions and adaptive discretization. First, we propose a novel strategy to learn efficient connections with no additional parameters and negligible computational overhead. Second, we introduce a novel convolutional Boolean architecture that exploits the locality with reduced number of Boolean operations than existing methods. Third, we propose an adaptive discretization strategy to reduce the accuracy drop when converting a continuous-valued network into a Boolean one. Extensive results on standard vision benchmarks demonstrate that the Pareto front of accuracy vs. computation of our method significantly outperforms prior state-of-the-art, achieving better accuracy with up to 37x fewer Boolean operations.

Temporal knowledge graph question answering (TKGQA) aims to answer time-sensitive questions by leveraging temporal knowledge bases. While Large Language Models (LLMs) demonstrate significant potential in TKGQA, current prompting strategies constrain their efficacy in two primary ways. First, they are prone to reasoning hallucinations under complex temporal constraints. Second, static prompting limits model autonomy and generalization, as it lack optimization through dynamic interaction with temporal knowledge graphs (TKGs) environments. To address these limitations, we propose \textbf{TKG-Thinker}, a novel agent equipped with autonomous planning and adaptive retrieval capabilities for reasoning over TKGs. Specifically, TKG-Thinker performs in-depth temporal reasoning through dynamic multi-turn interactions with TKGs via a dual-training strategy. We first apply Supervised Fine-Tuning (SFT) with chain-of thought data to instill core planning capabilities, followed by a Reinforcement Learning (RL) stage that leverages multi-dimensional rewards to refine reasoning policies under intricate temporal constraints. Experimental results on benchmark datasets with three open-source LLMs show that TKG-Thinker achieves state-of-the-art performance and exhibits strong generalization across complex TKGQA settings.

The rapid expansion of Internet of Things (IoT) ecosystems has led to increasingly complex and heterogeneous network topologies. Traditional network monitoring and visualization tools rely on aggregated metrics or static representations, which fail to capture the evolving relationships and structural dependencies between devices. Although Graph Neural Networks (GNNs) offer a powerful way to learn from relational data, their internal representations often remain opaque and difficult to interpret for security-critical operations. Consequently, this work introduces an interpretable pipeline that generates directly visualizable low-dimensional representations by mapping high-dimensional embeddings onto a latent manifold. This projection enables the interpretable monitoring and interoperability of evolving network states, while integrated feature attribution techniques decode the specific characteristics shaping the manifold structure. The framework achieves a classification F1-score of 0.830 for intrusion detection while also highlighting phenomena such as concept drift. Ultimately, the presented approach bridges the gap between high-dimensional GNN embeddings and human-understandable network behavior, offering new insights for network administrators and security analysts.

We study adaptive learning rate scheduling for norm-constrained optimizers (e.g., Muon and Lion). We introduce a generalized smoothness assumption under which local curvature decreases with the suboptimality gap and empirically verify that this behavior holds along optimization trajectories. Under this assumption, we establish convergence guarantees under an appropriate choice of learning rate, for which warm-up followed by decay arises naturally from the proof rather than being imposed heuristically. Building on this theory, we develop a practical learning rate scheduler that relies only on standard hyperparameters and adapts the warm-up duration automatically at the beginning of training. We evaluate this method on large language model pretraining with LLaMA architectures and show that our adaptive warm-up selection consistently outperforms or at least matches the best manually tuned warm-up schedules across all considered setups, without additional hyperparameter search. Our source code is available at https://github.com/brain-lab-research/llm-baselines/tree/warmup

We present a principled framework for confidence estimation in computed tomography (CT) reconstruction. Based on the sequential likelihood mixing framework (Kirschner et al., 2025), we establish confidence regions with theoretical coverage guarantees for deep-learning-based CT reconstructions. We consider a realistic forward model following the Beer-Lambert law, i.e., a log-linear forward model with Poisson noise, closely reflecting clinical and scientific imaging conditions. The framework is general and applies to both classical algorithms and deep learning reconstruction methods, including U-Nets, U-Net ensembles, and generative Diffusion models. Empirically, we demonstrate that deep reconstruction methods yield substantially tighter confidence regions than classical reconstructions, without sacrificing theoretical coverage guarantees. Our approach allows the detection of hallucinations in reconstructed images and provides interpretable visualizations of confidence regions. This establishes deep models not only as powerful estimators, but also as reliable tools for uncertainty-aware medical imaging.

The integration of spatial multi-omics data from single tissues is crucial for advancing biological research. However, a significant data imbalance impedes progress: while spatial transcriptomics data is relatively abundant, spatial proteomics data remains scarce due to technical limitations and high costs. To overcome this challenge we propose STProtein, a novel framework leveraging graph neural networks with multi-task learning strategy. STProtein is designed to accurately predict unknown spatial protein expression using more accessible spatial multi-omics data, such as spatial transcriptomics. We believe that STProtein can effectively addresses the scarcity of spatial proteomics, accelerating the integration of spatial multi-omics and potentially catalyzing transformative breakthroughs in life sciences. This tool enables scientists to accelerate discovery by identifying complex and previously hidden spatial patterns of proteins within tissues, uncovering novel relationships between different marker genes, and exploring the biological "Dark Matter".

Autonomous agents excel in self-improvement through reflection and iterative refinement, which reuse successful task trajectories as in-context examples to assist subsequent reasoning. However, shifting across tasks often introduces a context mismatch. Hence, existing approaches either discard the trajectories or manipulate them using heuristics, leading to a non-negligible fine-tuning cost or unguaranteed performance. To bridge this gap, we reveal a context-trajectory correlation, where shifts of context are highly parallel with shifts of trajectory. Based on this finding, we propose BrIdge contextual gap FoR imprOvised trajectory STeering (Bifrost), a training-free method that leverages context differences to precisely guide the adaptation of previously solved trajectories towards the target task, mitigating the misalignment caused by context shifts. Our trajectory adaptation is conducted at the representation level using agent hidden states, ensuring trajectory transformation accurately aligns with the target context in a shared space. Across diverse benchmarks, Bifrost consistently outperforms existing trajectory reuse and finetuned self-improvement methods, demonstrating that agents can effectively leverage past experiences despite substantial context shifts.

Large language models increasingly spend inference compute sampling multiple chain-of-thought traces or searching over merged checkpoints. This shifts the bottleneck from generation to selection, often without supervision on the target distribution. We show entropy-based exploration proxies follow an inverted-U with accuracy, suggesting extra exploration can become redundant and induce overthinking. We propose NEX, a white-box label-free unsupervised scoring framework that views reasoning as alternating E-phase (exploration) and X-phase (exploitation). NEX detects E-phase as spikes in newly activated MLP neurons per token from sparse activation caches, then uses a sticky two-state HMM to infer E-X phases and credits E-introduced neurons by whether they are reused in the following X span. These signals yield interpretable neuron weights and a single Good-Mass Fraction score to rank candidate responses and merged variants without task answers. Across reasoning benchmarks and Qwen3 merge families, NEX computed on a small unlabeled activation set predicts downstream accuracy and identifies better variants; we further validate the E-X signal with human annotations and provide causal evidence via "Effective-vs-Redundant" neuron transfer.

We consider the distributed wake-up problem with advice, where nodes are equipped with initial knowledge about the network at large. After the adversary awakens a subset of nodes, an oracle computes a bit string (``the advice'') for each node, and the goal is to wake up all sleeping nodes efficiently. We present the first upper and lower bounds on the message complexity for wake-up in the quantum routing model, introduced by Dufoulon, Magniez, and Pandurangan (PODC 2025). In more detail, we give a distributed advising scheme that, given $α$ bits of advice per node, wakes up all nodes with a message complexity of $O( \sqrt{\frac{n^3}{2^{\max\{\lfloor (α-1)/2 \rfloor},0\}}}\cdot\log n )$ with high probability. Our result breaks the $Ω( \frac{n^2}{2^α} )$ barrier known for the classical port numbering model in sufficiently dense graphs. To complement our algorithm, we give a lower bound on the message complexity for distributed quantum algorithms: By leveraging a lower bound result for the single-bit descriptor problem in the query complexity model, we show that wake-up has a quantum message complexity of $Ω( n^{3/2} )$ without advice, which holds independently of how much time we allow. In the setting where an adversary decides which nodes start the algorithm, most graph problems of interest implicitly require solving wake-up, and thus the same lower bound also holds for other fundamental problems such as single-source broadcast and spanning tree construction.

We study non-stationary single-item, periodic-review inventory control problems in which the demand distribution is unknown and may change over time. We analyze how demand non-stationarity affects learning performance across inventory models, including systems with demand backlogging or lost-sales, both with and without lead times. For each setting, we propose an adaptive online algorithm that optimizes over the class of base-stock policies and establish performance guarantees in terms of dynamic regret relative to the optimal base-stock policy at each time step. Our results reveal a sharp separation across inventory models. In backlogging systems and lost-sales models with zero lead time, we show that it is possible to adapt to demand changes without incurring additional performance loss in stationary environments, even without prior knowledge of the demand distributions or the number of demand shifts. In contrast, for lost-sales systems with positive lead times, we establish weaker guarantees that reflect fundamental limitations imposed by delayed replenishment in combination with censored feedback. Our algorithms leverage the convexity and one-sided feedback structure of inventory costs to enable counterfactual policy evaluation despite demand censoring. We complement the theoretical analysis with simulation results showing that our methods significantly outperform existing benchmarks.

Controlling the false discovery rate (FDR) in high-dimensional variable selection requires balancing rigorous error control with statistical power. Existing methods with provable guarantees are often overly conservative, creating a persistent gap between the realized false discovery proportion (FDP) and the target FDR level. We introduce a learning-augmented enhancement of the T-Rex Selector framework that narrows this gap. Our approach replaces the analytical FDP estimator with a neural network trained solely on diverse synthetic datasets, enabling a substantially tighter and more accurate approximation of the FDP. This refinement allows the procedure to operate much closer to the desired FDR level, thereby increasing discovery power while maintaining effective approximate control. Through extensive simulations and a challenging synthetic genome-wide association study (GWAS), we demonstrate that our method achieves superior detection of true variables compared to existing approaches.

Local differential privacy (LDP) has become a central topic in data privacy research, offering strong privacy guarantees by perturbing user data at the source and removing the need for a trusted curator. However, the noise introduced by LDP often significantly reduces data utility. To address this issue, we reinterpret private learning under LDP as a transfer learning problem, where the noisy data serve as the source domain and the unobserved clean data as the target. We propose novel techniques specifically designed for LDP to improve classification performance without compromising privacy: (1) a noised binary feedback-based evaluation mechanism for estimating dataset utility; (2) model reversal, which salvages underperforming classifiers by inverting their decision boundaries; and (3) model averaging, which assigns weights to multiple reversed classifiers based on their estimated utility. We provide theoretical excess risk bounds under LDP and demonstrate how our methods reduce this risk. Empirical results on both simulated and real-world datasets show substantial improvements in classification accuracy.

We present FiMI (Finance Model for India), a domain-specialized financial language model developed for Indian digital payment systems. We develop two model variants: FiMI Base and FiMI Instruct. FiMI adapts the Mistral Small 24B architecture through a multi-stage training pipeline, beginning with continuous pre-training on 68 Billion tokens of curated financial, multilingual (English, Hindi, Hinglish), and synthetic data. This is followed by instruction fine-tuning and domain-specific supervised fine-tuning focused on multi-turn, tool-driven conversations that model real-world workflows, such as transaction disputes and mandate lifecycle management. Evaluations reveal that FiMI Base achieves a 20% improvement over the Mistral Small 24B Base model on finance reasoning benchmark, while FiMI Instruct outperforms the Mistral Small 24B Instruct model by 87% on domain-specific tool-calling. Moreover, FiMI achieves these significant domain gains while maintaining comparable performance to models of similar size on general benchmarks.

Fast computation of a matrix product $W^\top X$ is a workhorse of modern LLMs. To make their deployment more efficient, a popular approach is that of using a low-precision approximation $\widehat W$ in place of true $W$ ("weight-only quantization''). Information theory demonstrates that an optimal algorithm for reducing precision of $W$ depends on the (second order) statistics of $X$ and requires a careful alignment of vector quantization codebook with PCA directions of $X$ (a process known as "waterfilling allocation''). Dependence of the codebook on statistics of $X$, however, is highly impractical. This paper proves that there exist a universal codebook that is simultaneously near-optimal for all possible statistics of $X$, in the sense of being at least as good as an $X$-adapted waterfilling codebook with rate reduced by 0.11 bit per dimension. Such universal codebook would be an ideal candidate for the low-precision storage format, a topic of active modern research, but alas the existence proof is non-constructive. Equivalently, our result shows existence of a net in $\mathbb{R}^n$ that is a nearly-optimal covering of a sphere simultaneously with respect to all Hilbert norms.

With the rising need for spatially grounded tasks such as Vision-Language Navigation/Action, allocentric perception capabilities in Vision-Language Models (VLMs) are receiving growing focus. However, VLMs remain brittle on allocentric spatial queries that require explicit perspective shifts, where the answer depends on reasoning in a target-centric frame rather than the observed camera view. Thus, we introduce Allocentric Perceiver, a training-free strategy that recovers metric 3D states from one or more images with off-the-shelf geometric experts, and then instantiates a query-conditioned allocentric reference frame aligned with the instruction's semantic intent. By deterministically transforming reconstructed geometry into the target frame and prompting the backbone VLM with structured, geometry-grounded representations, Allocentric Perceriver offloads mental rotation from implicit reasoning to explicit computation. We evaluate Allocentric Perciver across multiple backbone families on spatial reasoning benchmarks, observing consistent and substantial gains ($\sim$10%) on allocentric tasks while maintaining strong egocentric performance, and surpassing both spatial-perception-finetuned models and state-of-the-art open-source and proprietary models.

General-purpose text embedding models underpin a wide range of NLP and information retrieval applications, and are typically trained on large-scale multi-task corpora to encourage broad generalization. However, it remains unclear how different multi-task training strategies compare in practice, and how to efficiently adapt embedding models as new domains and data types continually emerge. In this work, we present a systematic study of multi-task training for text embeddings from two perspectives: data scheduling and model merging. We compare batch-level shuffling, sequential training variants, two-stage training, and multiple merging granularities, and find that simple batch-level shuffling consistently yields the strongest overall performance, suggesting that task conflicts are limited and training datasets are largely complementary. Despite its effectiveness, batch-level shuffling exhibits two practical limitations: suboptimal out-of-domain (OOD) generalization and poor suitability for incremental learning due to expensive full retraining. To address these issues, we propose Bagging-based rObust mOdel Merging (\modelname), which trains multiple embedding models on sampled subsets and merges them into a single model, improving robustness while retaining single-model inference efficiency. Moreover, \modelname naturally supports efficient incremental updates by training lightweight update models on new data with a small historical subset and merging them into the existing model. Experiments across diverse embedding benchmarks demonstrate that \modelname consistently improves both in-domain and OOD performance over full-corpus batch-level shuffling, while substantially reducing training cost in incremental learning settings.

Tree-boosting is a widely used machine learning technique for tabular data. However, its out-of-sample accuracy is critically dependent on multiple hyperparameters. In this article, we empirically compare several popular methods for hyperparameter optimization for tree-boosting including random grid search, the tree-structured Parzen estimator (TPE), Gaussian-process-based Bayesian optimization (GP-BO), Hyperband, the sequential model-based algorithm configuration (SMAC) method, and deterministic full grid search using $59$ regression and classification data sets. We find that the SMAC method clearly outperforms all the other considered methods. We further observe that (i) a relatively large number of trials larger than $100$ is required for accurate tuning, (ii) using default values for hyperparameters yields very inaccurate models, (iii) all considered hyperparameters can have a material effect on the accuracy of tree-boosting, i.e., there is no small set of hyperparameters that is more important than others, and (iv) choosing the number of boosting iterations using early stopping yields more accurate results compared to including it in the search space for regression tasks.

Generalizable image-based person re-identification (Re-ID) aims to recognize individuals across cameras in unseen domains without retraining. While multiple existing approaches address the domain gap through complex architectures, recent findings indicate that better generalization can be achieved by stylistically diverse single-camera data. Although this data is easy to collect, it lacks complexity due to minimal cross-view variation. We propose ReText, a novel method trained on a mixture of multi-camera Re-ID data and single-camera data, where the latter is complemented by textual descriptions to enrich semantic cues. During training, ReText jointly optimizes three tasks: (1) Re-ID on multi-camera data, (2) image-text matching, and (3) image reconstruction guided by text on single-camera data. Experiments demonstrate that ReText achieves strong generalization and significantly outperforms state-of-the-art methods on cross-domain Re-ID benchmarks. To the best of our knowledge, this is the first work to explore multimodal joint learning on a mixture of multi-camera and single-camera data in image-based person Re-ID.

Recent advances in diffusion-based reinforcement learning (RL) methods have demonstrated promising results in a wide range of continuous control tasks. However, existing works in this field focus on the application of diffusion policies while leaving the diffusion critics unexplored. In fact, since policy optimization fundamentally relies on the critic, accurate value estimation is far more important than policy expressiveness. Furthermore, given the stochasticity of most reinforcement learning tasks, it has been confirmed that the critic is more appropriately depicted with a distributional model. Motivated by these points, we propose a novel distributional RL method with Diffusion Bridge Critics (DBC). DBC directly models the inverse cumulative distribution function (CDF) of the Q value. This allows us to accurately capture the value distribution and prevents it from collapsing into a trivial Gaussian distribution owing to the strong distribution-matching capability of the diffusion bridge. Moreover, we further derive an analytic integral formula to address discretization errors in DBC, which is essential in value estimation. To our knowledge, DBC is the first work to employ the diffusion bridge model as the critic. Notably, DBC is also a plug-and-play component and can be integrated into most existing RL frameworks. Experimental results on MuJoCo robot control benchmarks demonstrate the superiority of DBC compared with previous distributional critic models.

Code completion (CC) is a task frequently used by developers when working in collaboration with LLM-based programming assistants. Despite the increased performance of LLMs on public benchmarks, out of the box LLMs still have a hard time generating code that aligns with a private code repository not previously seen by the model's training data. Customizing code LLMs to a private repository provides a way to improve the model performance. In this paper we present our approach for automated LLM customization based on semantic scopes in the code. We evaluate LLMs on real industry cases with two private enterprise code repositories with two customization strategies: Retrieval-Augmented Generation (RAG) and supervised Fine-Tuning (FT). Our mechanism for ingesting the repository's data and formulating the training data pairs with semantic scopes helps models to learn the underlying patterns specific to the repository, providing more precise code to developers and helping to boost their productivity. The code completions of moderately sized customized models can be significantly better than those of uncustomized models of much larger capacity. We also include an analysis of customization on two public benchmarks and present opportunities for future work.

The intermediate layers of deep networks can be characterised as a Gaussian process, in particular the Edge-of-Chaos (EoC) initialisation strategy prescribes the limiting covariance matrix of the Gaussian process. Here we show that the under-utilised chosen variance of the Gaussian process is important in the training of deep networks with sparsity inducing activation, such as a shifted and clipped ReLU, $\text{CReLU}_{τ,m}(x)=\min(\max(x-τ,0),m)$. Specifically, initialisations leading to larger fixed Gaussian process variances, allow for improved expressivity with activation sparsity as large as 90% in DNNs and CNNs, and generally improve the stability of the training process. Enabling full, or near full, accuracy at such high levels of sparsity in the hidden layers suggests a promising mechanism to reduce the energy consumption of machine learning models involving fully connected layers.

It remains a critical challenge to adapt policies across domains with mismatched dynamics in reinforcement learning (RL). In this paper, we study cross-domain offline RL, where an offline dataset from another similar source domain can be accessed to enhance policy learning upon a target domain dataset. Directly merging the two datasets may lead to suboptimal performance due to potential dynamics mismatches. Existing approaches typically mitigate this issue through source domain transition filtering or reward modification, which, however, may lead to insufficient exploitation of the valuable source domain data. Instead, we propose to modify the source domain data into the target domain data. To that end, we leverage an inverse policy model and a reward model to correct the actions and rewards of source transitions, explicitly achieving alignment with the target dynamics. Since limited data may result in inaccurate model training, we further employ a forward dynamics model to retain corrected samples that better match the target dynamics than the original transitions. Consequently, we propose the Selective Transition Correction (STC) algorithm, which enables reliable usage of source domain data for policy adaptation. Experiments on various environments with dynamics shifts demonstrate that STC achieves superior performance against existing baselines.

Speculative decoding accelerates inference for (M)LLMs, yet a training-decoding discrepancy persists: while existing methods optimize single greedy trajectories, decoding involves verifying and ranking multiple sampled draft paths. We propose Variational Speculative Decoding (VSD), formulating draft training as variational inference over latent proposals (draft paths). VSD maximizes the marginal probability of target-model acceptance, yielding an ELBO that promotes high-quality latent proposals while minimizing divergence from the target distribution. To enhance quality and reduce variance, we incorporate a path-level utility and optimize via an Expectation-Maximization procedure. The E-step draws MCMC samples from an oracle-filtered posterior, while the M-step maximizes weighted likelihood using Adaptive Rejection Weighting (ARW) and Confidence-Aware Regularization (CAR). Theoretical analysis confirms that VSD increases expected acceptance length and speedup. Extensive experiments across LLMs and MLLMs show that VSD achieves up to a 9.6% speedup over EAGLE-3 and 7.9% over ViSpec, significantly improving decoding efficiency.

LLM-based assistants have been widely popularised after the release of ChatGPT. Concerns have been raised about their misuse in academia, given the difficulty of distinguishing between human-written and generated text. To combat this, automated techniques have been developed and shown to be effective, to some extent. However, prior work suggests that these methods often falsely flag essays from non-native speakers as generated, due to their low perplexity extracted from an LLM, which is supposedly a key feature of the detectors. We revisit these statements two years later, specifically in the Czech language setting. We show that the perplexity of texts from non-native speakers of Czech is not lower than that of native speakers. We further examine detectors from three separate families and find no systematic bias against non-native speakers. Finally, we demonstrate that contemporary detectors operate effectively without relying on perplexity.

Photomultiplier tubes (PMTs) are widely employed in particle and nuclear physics experiments. The accuracy of PMT waveform reconstruction directly impacts the detector's spatial and energy resolution. A key challenge arises when multiple photons arrive within a few nanoseconds, making it difficult to resolve individual photoelectrons (PEs). Although supervised deep learning methods have surpassed traditional methods in performance, their practical applicability is limited by the lack of ground-truth PE labels in real data. To address this issue, we propose an innovative weakly supervised waveform simulation and reconstruction approach based on a bidirectional conditional diffusion network framework. The method is fully data-driven and requires only raw waveforms and coarse estimates of PE information as input. It first employs a PE-conditioned diffusion model to simulate realistic waveforms from PE sequences, thereby learning the features of overlapping waveforms. Subsequently, these simulated waveforms are used to train a waveform-conditioned diffusion model to reconstruct the PE sequences from waveforms, reinforcing the learning of features of overlapping waveforms. Through iterative refinement between the two conditional diffusion processes, the model progressively improves reconstruction accuracy. Experimental results demonstrate that the proposed method achieves 99% of the normalized PE-number resolution averaged over 1-5 p.e. and 80% of the timing resolution attained by fully supervised learning.

In recent years, Vision-Language-Action (VLA) models have emerged as a crucial pathway towards general embodied intelligence, yet their training efficiency has become a key bottleneck. Although existing reinforcement learning (RL)-based training frameworks like RLinf can enhance model generalization, they still rely on synchronous execution, leading to severe resource underutilization and throughput limitations during environment interaction, policy generation (rollout), and model update phases (actor). To overcome this challenge, this paper, for the first time, proposes and implements a fully-asynchronous policy training framework encompassing the entire pipeline from environment interaction, rollout generation, to actor policy updates. Systematically drawing inspiration from asynchronous optimization ideas in large model RL, our framework designs a multi-level decoupled architecture. This includes asynchronous parallelization of environment interaction and trajectory collection, streaming execution for policy generation, and decoupled scheduling for training updates. We validated the effectiveness of our method across diverse VLA models and environments. On the LIBERO benchmark, the framework achieves throughput improvements of up to 59.25\% compared to existing synchronous strategies. When deeply optimizing separation strategies, throughput can be increased by as much as 126.67\%. We verified the effectiveness of each asynchronous component via ablation studies. Scaling law validation across 8 to 256 GPUs demonstrates our method's excellent scalability under most conditions.

This work studies the applicability of automatic AI agent optimization methods to real-world agents in formal verification settings, focusing on automated theorem proving in Rocq as a representative and challenging domain. We evaluate how different automatic agent optimizers perform when applied to the task of optimizing a Rocq proof-generation agent, and assess whether parts of the fine-grained tuning of agentic systems, such as prompt design, contextual knowledge, and control strategies, can be automated. Our results show that while several optimizers yield measurable improvements, simple few-shot bootstrapping is the most consistently effective; however, none of the studied methods matches the performance of a carefully engineered state-of-the-art proof agent.

The quantum threat to cybersecurity has accelerated the standardization of Post-Quantum Cryptography (PQC). Migrating legacy software to these quantum-safe algorithms is not a simple library swap, but a new software engineering challenge: existing vulnerability detection, refactoring, and testing tools are not designed for PQC's probabilistic behavior, side-channel sensitivity, and complex performance trade-offs. To address these challenges, this paper outlines a vision for a new class of tools and introduces the Automated Quantum-safe Adaptation (AQuA) framework, with a three-pillar agenda for PQC-aware detection, semantic refactoring, and hybrid verification, thereby motivating Quantum-Safe Software Engineering (QSSE) as a distinct research direction.

Reinforcement Learning has emerged as a key driver for LLM reasoning. This capability is equally pivotal in long-context scenarios--such as long-dialogue understanding and structured data analysis, where the challenge extends beyond consuming tokens to performing rigorous deduction. While existing efforts focus on data synthesis or architectural changes, recent work points out that relying solely on sparse, outcome-only rewards yields limited gains, as such coarse signals are often insufficient to effectively guide the complex long-context reasoning. To address this, we propose LongR, a unified framework that enhances long-context performance by integrating a dynamic "Think-and-Read" mechanism, which interleaves reasoning with document consultation, with a contextual density reward based on relative information gain to quantify the utility of the relevant documents. Empirically, LongR achieves a 9% gain on LongBench v2 and consistent improvements on RULER and InfiniteBench, demonstrating robust efficiency in navigating extensive contexts. Furthermore, LongR consistently enhances performance across diverse RL algorithms (e.g., DAPO, GSPO). Finally, we conduct in-depth analyses to investigate the impact of reasoning chain length on efficiency and the model's robustness against distractors.

Pipeline parallelism enables training models that exceed single-device memory, but practical throughput remains limited by pipeline bubbles. Although parameter freezing can improve training throughput by adaptively skipping backward computation, existing methods often over-freeze parameters, resulting in unnecessary accuracy degradation. To address this issue, we propose TimelyFreeze, which models the pipeline schedule as a directed acyclic graph and solves a linear program to compute optimal freeze ratios that minimize batch execution time under accuracy constraints. Experiments show that TimelyFreeze achieves up to 40% training throughput improvement on LLaMA-8B with comparable accuracy. Overall, it enables faster large-scale model training without compromising convergence and generalizes across diverse pipeline-parallel settings.

Deep clustering (DC) is often quoted to have a key advantage over $k$-means clustering. Yet, this advantage is often demonstrated using image datasets only, and it is unclear whether it addresses the fundamental limitations of $k$-means clustering. Deep Embedded Clustering (DEC) learns a latent representation via an autoencoder and performs clustering based on a $k$-means-like procedure, while the optimization is conducted in an end-to-end manner. This paper investigates whether the deep-learned representation has enabled DEC to overcome the known fundamental limitations of $k$-means clustering, i.e., its inability to discover clusters of arbitrary shapes, varied sizes and densities. Our investigations on DEC have a wider implication on deep clustering methods in general. Notably, none of these methods exploit the underlying data distribution. We uncover that a non-deep learning approach achieves the intended aim of deep clustering by making use of distributional information of clusters in a dataset to effectively address these fundamental limitations.

Membership inference attacks (MIAs) aim to determine whether a sample was part of a model's training set, posing serious privacy risks for modern machine-learning systems. Existing MIAs primarily rely on static indicators, such as loss or confidence, and do not fully leverage the dynamic behavior of models when actively probed. We propose LeakBoost, a perceptual-loss-based interrogation framework that actively probes a model's internal representations to expose hidden membership signals. Given a candidate input, LeakBoost synthesizes an interrogation image by optimizing a perceptual (activation-space) objective, amplifying representational differences between members and non-members. This image is then analyzed by an off-the-shelf membership detector, without modifying the detector itself. When combined with existing membership inference methods, LeakBoost achieves substantial improvements at low false-positive rates across multiple image classification datasets and diverse neural network architectures. In particular, it raises AUC from near-chance levels (0.53-0.62) to 0.81-0.88, and increases TPR at 1 percent FPR by over an order of magnitude compared to strong baseline attacks. A detailed sensitivity analysis reveals that deeper layers and short, low-learning-rate optimization produce the strongest leakage, and that improvements concentrate in gradient-based detectors. LeakBoost thus offers a modular and computationally efficient way to assess privacy risks in white-box settings, advancing the study of dynamic membership inference.

Prompt injection is one of the most critical vulnerabilities in LLM agents; yet, effective automated attacks remain largely unexplored from an optimization perspective. Existing methods heavily depend on human red-teamers and hand-crafted prompts, limiting their scalability and adaptability. We propose AutoInject, a reinforcement learning framework that generates universal, transferable adversarial suffixes while jointly optimizing for attack success and utility preservation on benign tasks. Our black-box method supports both query-based optimization and transfer attacks to unseen models and tasks. Using only a 1.5B parameter adversarial suffix generator, we successfully compromise frontier systems including GPT 5 Nano, Claude Sonnet 3.5, and Gemini 2.5 Flash on the AgentDojo benchmark, establishing a stronger baseline for automated prompt injection research.

Weighted empirical risk minimization is a common approach to prediction under distribution drift. This article studies its out-of-sample prediction error under nonstationarity. We provide a general decomposition of the excess risk into a learning term and an error term associated with distribution drift, and prove oracle inequalities for the learning error under mixing conditions. The learning bound holds uniformly over arbitrary weight classes and accounts for the effective sample size induced by the weight vector, the complexity of the weight and hypothesis classes, and potential data dependence. We illustrate the applicability and sharpness of our results in (auto-) regression problems with linear models, basis approximations, and neural networks, recovering minimax-optimal rates (up to logarithmic factors) when specialized to unweighted and stationary settings.

Non-Intrusive Load Monitoring (NILM), commonly known as energy disaggregation, aims to estimate the power consumption of individual appliances by analyzing a home's total electricity usage. This method provides a cost-effective alternative to installing dedicated smart meters for each appliance. In this paper, we introduce a novel framework that incorporates Automated Machine Learning (AutoML) into the NILM domain, utilizing Bayesian Optimization for automated model selection and hyperparameter tuning. This framework empowers domain practitioners to effectively apply machine learning techniques without requiring advanced expertise in data science or machine learning. To support further research and industry adoption, we present AutoML4NILM, a flexible and extensible open-source toolkit designed to streamline the deployment of AutoML solutions for energy disaggregation. Currently, this framework supports 11 algorithms, each with different hyperparameters; however, its flexible design allows for the extension of both the algorithms and their hyperparameters.

In this paper, we present a neuro-inspired approach to reservoir computing (RC) in which a network of in vitro cultured cortical neurons serves as the physical reservoir. Rather than relying on artificial recurrent models to approximate neural dynamics, our biological reservoir computing (BRC) system leverages the spontaneous and stimulus-evoked activity of living neural circuits as its computational substrate. A high-density multi-electrode array (HD-MEA) provides simultaneous stimulation and readout across hundreds of channels: input patterns are delivered through selected electrodes, while the remaining ones capture the resulting high-dimensional neural responses, yielding a biologically grounded feature representation. A linear readout layer (single-layer perceptron) is then trained to classify these reservoir states, enabling the living neural network to perform static visual pattern-recognition tasks within a computer-vision framework. We evaluate the system across a sequence of tasks of increasing difficulty, ranging from pointwise stimuli to oriented bars, clock-digit-like shapes, and handwritten digits from the MNIST dataset. Despite the inherent variability of biological neural responses-arising from noise, spontaneous activity, and inter-session differences-the system consistently generates high-dimensional representations that support accurate classification. These results demonstrate that in vitro cortical networks can function as effective reservoirs for static visual pattern recognition, opening new avenues for integrating living neural substrates into neuromorphic computing frameworks. More broadly, this work contributes to the effort to incorporate biological principles into machine learning and supports the goals of neuro-inspired vision by illustrating how living neural systems can inform the design of efficient and biologically grounded computational models.

In the era of large foundation models, the quality of embeddings has become a central determinant of downstream task performance and overall system capability. Yet widely used dense embeddings are often extremely high-dimensional, incurring substantial costs in storage, memory, and inference latency. To address these, Contrastive Sparse Representation (CSR) is recently proposed as a promising direction, mapping dense embeddings into high-dimensional but k-sparse vectors, in contrast to compact dense embeddings such as Matryoshka Representation Learning (MRL). Despite its promise, CSR suffers severe degradation in the ultra-sparse regime, where over 80% of neurons remain inactive, leaving much of its efficiency potential unrealized. In this paper, we introduce CSRv2, a principled training approach designed to make ultra-sparse embeddings viable. CSRv2 stabilizes sparsity learning through progressive k-annealing, enhances representational quality via supervised contrastive objectives, and ensures end-to-end adaptability with full backbone finetuning. CSRv2 reduces dead neurons from 80% to 20% and delivers a 14% accuracy gain at k=2, bringing ultra-sparse embeddings on par with CSR at k=8 and MRL at 32 dimensions, all with only two active features. While maintaining comparable performance, CSRv2 delivers a 7x speedup over MRL, and yields up to 300x improvements in compute and memory efficiency relative to dense embeddings in text representation. Extensive experiments across text and vision demonstrate that CSRv2 makes ultra-sparse embeddings practical without compromising performance, where CSRv2 achieves 7%/4% improvement over CSR when k=4 and further increases this gap to 14%/6% when k=2 in text/vision representation. By making extreme sparsity viable, CSRv2 broadens the design space for real-time and edge-deployable AI systems where both embedding quality and efficiency are critical.

Search behaviour is characterised using synonymy and polysemy as users often want to search information based on meaning. Semantic representation strategies represent a move towards richer associative connections that can adequately capture this complex usage of language. Vector Space Modelling (VSM) and neural word embeddings play a crucial role in modern machine learning and Natural Language Processing (NLP) pipelines. Embeddings use distributional semantics to represent words, sentences, paragraphs or entire documents as vectors in high dimensional spaces. This can be leveraged by Information Retrieval (IR) systems to exploit the semantic relatedness between queries and answers. This paper evaluates an alternative approach to measuring query statement similarity that moves away from the common similarity measure of centroids of neural word embeddings. Motivated by the Word Movers Distance (WMD) model, similarity is evaluated using the distance between individual words of queries and statements. Results from ranked query and response statements demonstrate significant gains in accuracy using the combined approach of similarity ranking through WMD with the word embedding techniques. The top performing WMD + GloVe combination outperforms all other state-of-the-art retrieval models including Doc2Vec and the baseline LSA model. Along with the significant gains in performance of similarity ranking through WMD, we conclude that the use of pre-trained word embeddings, trained on vast amounts of data, result in domain agnostic language processing solutions that are portable to diverse business use-cases.

Multimodal embeddings serve as a bridge for aligning vision and language, with the two primary implementations -- CLIP-based and MLLM-based embedding models -- both limited to capturing only global semantic information. Although numerous studies have focused on fine-grained understanding, we observe that complex scenarios currently targeted by MLLM embeddings often involve a hybrid perceptual pattern of both global and fine-grained elements, thus necessitating a compatible fusion mechanism. In this paper, we propose Adaptive Global and Fine-grained perceptual Fusion for MLLM Embeddings (AGFF-Embed), a method that prompts the MLLM to generate multiple embeddings focusing on different dimensions of semantic information, which are then adaptively and smoothly aggregated. Furthermore, we adapt AGFF-Embed with the Explicit Gradient Amplification (EGA) technique to achieve in-batch hard negatives enhancement without requiring fine-grained editing of the dataset. Evaluation on the MMEB and MMVP-VLM benchmarks shows that AGFF-Embed comprehensively achieves state-of-the-art performance in both general and fine-grained understanding compared to other multimodal embedding models.

Retrieval-augmented generation (RAG) has become a key paradigm for knowledge-intensive question answering. However, existing multi-hop RAG systems remain inefficient, as they alternate between retrieval and reasoning at each step, resulting in repeated LLM calls, high token consumption, and unstable entity grounding across hops. We propose CompactRAG, a simple yet effective framework that decouples offline corpus restructuring from online reasoning. In the offline stage, an LLM reads the corpus once and converts it into an atomic QA knowledge base, which represents knowledge as minimal, fine-grained question-answer pairs. In the online stage, complex queries are decomposed and carefully rewritten to preserve entity consistency, and are resolved through dense retrieval followed by RoBERTa-based answer extraction. Notably, during inference, the LLM is invoked only twice in total - once for sub-question decomposition and once for final answer synthesis - regardless of the number of reasoning hops. Experiments on HotpotQA, 2WikiMultiHopQA, and MuSiQue demonstrate that CompactRAG achieves competitive accuracy while substantially reducing token consumption compared to iterative RAG baselines, highlighting a cost-efficient and practical approach to multi-hop reasoning over large knowledge corpora. The implementation is available at GitHub.

Muon updates matrix parameters via the matrix sign of the gradient and has shown strong empirical gains, yet its dynamics and scaling behavior remain unclear in theory. We study Muon in a linear associative memory model with softmax retrieval and a hierarchical frequency spectrum over query-answer pairs, with and without label noise. In this setting, we show that Gradient Descent (GD) learns frequency components at highly imbalanced rates, leading to slow convergence bottlenecked by low-frequency components. In contrast, the Muon optimizer mitigates this imbalance, leading to faster and more uniform progress. Specifically, in the noiseless case, Muon achieves an exponential speedup over GD; in the noisy case with a power-decay frequency spectrum, we derive Muon's optimization scaling law and demonstrate its superior scaling efficiency over GD. Furthermore, we show that Muon can be interpreted as an implicit matrix preconditioner arising from adaptive task alignment and block-symmetric gradient structure. In contrast, the preconditioner with coordinate-wise sign operator could match Muon under oracle access to unknown task representations, which is infeasible for SignGD in practice. Experiments on synthetic long-tail classification and LLaMA-style pre-training corroborate the theory.

In financial Retrieval-Augmented Generation (RAG) systems, models frequently rely on retrieved documents to generate accurate responses due to the time-sensitive nature of the financial domain. While retrieved documents help address knowledge gaps, model-generated responses still suffer from hallucinations that contradict the retrieved information. To mitigate this inconsistency, we propose a Reinforcement Learning framework enhanced with Fine-grained Knowledge Verification (RLFKV). Our method decomposes financial responses into atomic knowledge units and assesses the correctness of each unit to compute the fine-grained faithful reward. This reward offers more precise optimization signals, thereby improving alignment with the retrieved documents. Additionally, to prevent reward hacking (e.g., overly concise replies), we incorporate an informativeness reward that encourages the policy model to retain at least as many knowledge units as the base model. Experiments conducted on the public Financial Data Description (FDD) task and our newly proposed FDD-ANT dataset demonstrate consistent improvements, confirming the effectiveness of our approach.

Automated vulnerability reproduction from CVE descriptions requires generating executable Proof-of-Concept (PoC) exploits and validating them in target environments. This process is critical in software security research and practice, yet remains time-consuming and demands specialized expertise when performed manually. While LLM agents show promise for automating this task, existing approaches often conflate exploring attack directions with fixing implementation details, which leads to unproductive debugging loops when reproduction fails. To address this, we propose Cve2PoC, an LLM-based dual-loop agent framework following a plan-execute-evaluate paradigm. The Strategic Planner analyzes vulnerability semantics and target code to produce structured attack plans. The Tactical Executor generates PoC code and validates it through progressive verification. The Adaptive Refiner evaluates execution results and routes failures to different loops: the \textit{Tactical Loop} for code-level refinement, while the \textit{Strategic Loop} for attack strategy replanning. This dual-loop design enables the framework to escape ineffective debugging by matching remediation to failure type. Evaluation on two benchmarks covering 617 real-world vulnerabilities demonstrates that Cve2PoC achieves 82.9\% and 54.3\% reproduction success rates on SecBench.js and PatchEval, respectively, outperforming the best baseline by 11.3\% and 20.4\%. Human evaluation confirms that generated PoCs achieve comparable code quality to human-written exploits in readability and reusability.

Reinforcement Learning with Verifiable Rewards (RLVR) is increasingly viewed as a tree pruning mechanism. However, we identify a systemic pathology termed Recursive Space Contraction (RSC), an irreversible collapse driven by the combined dynamics of positive sharpening and negative squeezing, where the sampling probability of valid alternatives vanishes. While Kullback-Leibler (KL) regularization aims to mitigate this, it imposes a rigid Shape Matching constraint that forces the policy to mimic the reference model's full density, creating a gradient conflict with the sharpening required for correctness. We propose Anchored Policy Optimization (APO), shifting the paradigm from global Shape Matching to Support Coverage. By defining a Safe Manifold based on the reference model's high-confidence support, APO permits aggressive sharpening for efficiency while selectively invoking a restorative force during error correction to prevent collapse. We theoretically derive that APO serves as a gradient-aligned mechanism to maximize support coverage, enabling an Elastic Recovery that re-inflates valid branches. Empirical evaluations on mathematical benchmarks demonstrate that APO breaks the accuracy-diversity trade-off, significantly improving Pass@1 while restoring the Pass@K diversity typically lost by standard policy gradient methods.

Boosting algorithms enjoy strong theoretical guarantees: when weak learners maintain positive edge, AdaBoost achieves geometric decrease of exponential loss. We study how to incorporate group fairness constraints into boosting while preserving analyzable training dynamics. Our approach, FairBoost, projects the ensemble-induced exponential-weights distribution onto a convex set of distributions satisfying fairness constraints (as a reweighting surrogate), then trains weak learners on this fair distribution. The key theoretical insight is that projecting the training distribution reduces the effective edge of weak learners by a quantity controlled by the KL-divergence of the projection. We prove an exponential-loss bound where the convergence rate depends on weak learner edge minus a "fairness cost" term $δ_t = \sqrt{\mathrm{KL}(w^t \| q^t)/2}$. This directly quantifies the accuracy-fairness tradeoff in boosting dynamics. Experiments on standard benchmarks validate the theoretical predictions and demonstrate competitive fairness-accuracy tradeoffs with stable training curves.

Context: AI-assisted tools are increasingly integrated into software development workflows, but their reliance on large language models (LLMs) introduces substantial computational and energy costs. Understanding and reducing the energy footprint of LLM inference is therefore essential for sustainable software development. Objective: In this study, we conduct a phase-level analysis of LLM inference energy consumption, distinguishing between the (1) prefill, where the model processes the input and builds internal representations, and (2) decoding, where output tokens are generated using the stored state. Method: We investigate six 6B-7B and four 3B-4B transformer-based models, evaluating them on code-centric benchmarks HumanEval for code generation and LongBench for code understanding. Results: Our findings show that, within both parameter groups, models exhibit distinct energy patterns across phases. Furthermore, we observed that increases in prefill cost amplify the energy cost per token during decoding, with amplifications ranging from 1.3% to 51.8% depending on the model. Lastly, three out of ten models demonstrate babbling behavior, adding excessive content to the output that unnecessarily inflates energy consumption. We implemented babbling suppression for code generation, achieving energy savings ranging from 44% to 89% without affecting generation accuracy. Conclusion: These findings show that prefill costs influence decoding, which dominates energy consumption, and that babbling suppression can yield up to 89% energy savings. Reducing inference energy therefore requires both mitigating babbling behavior and limiting impact of prefill on decoding.

Mixture-of-Experts (MoE) architectures are evolving towards finer granularity to improve parameter efficiency. However, existing MoE designs face an inherent trade-off between the granularity of expert specialization and hardware execution efficiency. We propose OmniMoE, a system-algorithm co-designed framework that pushes expert granularity to its logical extreme. OmniMoE introduces vector-level Atomic Experts, enabling scalable routing and execution within a single MoE layer, while retaining a shared dense MLP branch for general-purpose processing. Although this atomic design maximizes capacity, it poses severe challenges for routing complexity and memory access. To address these, OmniMoE adopts a system-algorithm co-design: (i) a Cartesian Product Router that decomposes the massive index space to reduce routing complexity from O(N) to O(sqrt(N)); and (ii) Expert-Centric Scheduling that inverts the execution order to turn scattered, memory-bound lookups into efficient dense matrix operations. Validated on seven benchmarks, OmniMoE (with 1.7B active parameters) achieves 50.9% zero-shot accuracy across seven benchmarks, outperforming coarse-grained (e.g., DeepSeekMoE) and fine-grained (e.g., PEER) baselines. Crucially, OmniMoE reduces inference latency from 73ms to 6.7ms (a 10.9-fold speedup) compared to PEER, demonstrating that massive-scale fine-grained MoE can be fast and accurate. Our code is open-sourced at https://github.com/flash-algo/omni-moe.

This study challenges the presumed neutrality of latent spaces in vision language models (VLMs) by adopting an ethological perspective on their algorithmic behaviors. Rather than constituting spaces of homogeneous indeterminacy, latent spaces exhibit model-specific algorithmic sensitivities, understood as differential regimes of perceptual salience shaped by training data and architectural choices. Through a comparative analysis of three models (OpenAI CLIP, OpenCLIP LAION, SigLIP) applied to a corpus of 301 artworks (15th to 20th), we reveal substantial divergences in the attribution of political and cultural categories. Using bipolar semantic axes derived from vector analogies (Mikolov et al., 2013), we show that SigLIP classifies 59.4% of the artworks as politically engaged, compared to only 4% for OpenCLIP. African masks receive the highest political scores in SigLIP while remaining apolitical in OpenAI CLIP. On an aesthetic colonial axis, inter-model discrepancies reach 72.6 percentage points. We introduce three operational concepts: computational latent politicization, describing the emergence of political categories without intentional encoding; emergent bias, irreducible to statistical or normative bias and detectable only through contrastive analysis; and three algorithmic scopic regimes: entropic (LAION), institutional (OpenAI), and semiotic (SigLIP), which structure distinct modes of visibility. Drawing on Foucault's notion of the archive, Jameson's ideologeme, and Simondon's theory of individuation, we argue that training datasets function as quasi-archives whose discursive formations crystallize within latent space. This work contributes to a critical reassessment of the conditions under which VLMs are applied to digital art history and calls for methodologies that integrate learning architectures into any delegation of cultural interpretation to algorithmic agents.

Low-rank adaptation (LoRA) approximates the update of a pretrained weight matrix using the product of two low-rank matrices. However, standard LoRA follows an explicit-rank paradigm, where increasing model capacity requires adding more rows or columns (i.e., basis vectors) to the low-rank matrices, leading to substantial parameter growth. In this paper, we find that these basis vectors exhibit significant parameter redundancy and can be compactly represented by lightweight nonlinear functions. Therefore, we propose Generative Low-Rank Adapter (GenLoRA), which replaces explicit basis vector storage with nonlinear basis vector generation. Specifically, GenLoRA maintains a latent vector for each low-rank matrix and employs a set of lightweight radial basis functions (RBFs) to synthesize the basis vectors. Each RBF requires far fewer parameters than an explicit basis vector, enabling higher parameter efficiency in GenLoRA. Extensive experiments across multiple datasets and architectures show that GenLoRA attains higher effective LoRA ranks under smaller parameter budgets, resulting in superior fine-tuning performance. The code is available at https://anonymous.4open.science/r/GenLoRA-1519.

Retrieval-augmented generation (RAG) enhances LLM reasoning in knowledge-intensive tasks, but existing RAG pipelines incur substantial retrieval and generation overhead when applied to large-scale entity matching. To address this limitation, we introduce CE-RAG4EM, a cost-efficient RAG architecture that reduces computation through blocking-based batch retrieval and generation. We also present a unified framework for analyzing and evaluating RAG systems for entity matching, focusing on blocking-aware optimizations and retrieval granularity. Extensive experiments suggest that CE-RAG4EM can achieve comparable or improved matching quality while substantially reducing end-to-end runtime relative to strong baselines. Our analysis further reveals that key configuration parameters introduce an inherent trade-off between performance and overhead, offering practical guidance for designing efficient and scalable RAG systems for entity matching and data integration.

Machine learning practitioners frequently observe tension between predictive accuracy and group fairness constraints -- yet sometimes fairness interventions appear to improve accuracy. We show that both phenomena can be artifacts of training data that misrepresents subgroup proportions. Under subpopulation shift (stable within-group distributions, shifted group proportions), we establish: (i) full importance-weighted correction is asymptotically unbiased but finite-sample suboptimal; (ii) the optimal finite-sample correction is a shrinkage reweighting that interpolates between target and training mixtures; (iii) apparent "fairness helps accuracy" can arise from comparing fairness methods to an improperly-weighted baseline. We provide an actionable evaluation protocol: fix representation (optimally) before fairness -- compare fairness interventions against a shrinkage-corrected baseline to isolate the true, irreducible price of fairness. Experiments on synthetic and real-world benchmarks (Adult, COMPAS) validate our theoretical predictions and demonstrate that this protocol eliminates spurious tradeoffs, revealing the genuine fairness-utility frontier.

Recently, the use of smart cameras in outdoor settings has grown to improve surveillance and security. Nonetheless, these systems are susceptible to tampering, whether from deliberate vandalism or harsh environmental conditions, which can undermine their monitoring effectiveness. In this context, detecting camera tampering is more challenging when a camera is capturing still images rather than video as there is no sequence of continuous frames over time. In this study, we propose two approaches for detecting tampered images: a rule-based method and a deep-learning-based method. The aim is to evaluate how each method performs in terms of accuracy, computational demands, and the data required for training when applied to real-world scenarios. Our results show that the deep-learning model provides higher accuracy, while the rule-based method is more appropriate for scenarios where resources are limited and a prolonged calibration phase is impractical. We also offer publicly available datasets with normal, blurred, and rotated images to support the development and evaluation of camera tampering detection methods, addressing the need for such resources.

Understanding the limitations of gradient methods, and stochastic gradient descent (SGD) in particular, is a central challenge in learning theory. To that end, a commonly used tool is the Statistical Queries (SQ) framework, which studies performance limits of algorithms based on noisy interaction with the data. However, it is known that the formal connection between the SQ framework and SGD is tenuous: Existing results typically rely on adversarial or specially-structured gradient noise that does not reflect the noise in standard SGD, and (as we point out here) can sometimes lead to incorrect predictions. Moreover, many analyses of SGD for challenging problems rely on non-trivial algorithmic modifications, such as restricting the SGD trajectory to the sphere or using very small learning rates. To address these shortcomings, we develop a new, non-SQ framework to study the limitations of standard vanilla SGD, for single-index and multi-index models (namely, when the target function depends on a low-dimensional projection of the inputs). Our results apply to a broad class of settings and architectures, including (potentially deep) neural networks.

SEAL is a static analyser for the verification of programs that manipulate unbounded linked data structures. It is based on separation logic to represent abstract memory states and, unlike other separation-logic-based approaches, it employs a general-purpose separation logic solver Astral for satisfiability and entailment checking, which itself is based on translation to SMT. This design results in a modular architecture intended to be easier to extend and to combine with reasoning in other theories. Although still a prototype, SEAL achieved competitive results in the LinkedLists base category and was one of only four analysers capable of verifying programs with unbounded lists. We believe that the tool's extensibility, combined with further development, can lead to significant improvements in future competitions.

In materials science, data are scarce and expensive to generate, whether computationally or experimentally. Therefore, it is crucial to identify how model performance scales with dataset size and model capacity to distinguish between data- and model-limited regimes. Neural scaling laws provide a framework for quantifying this behavior and guide the design of materials datasets and machine learning architectures. Here, we investigate neural scaling laws for a paradigmatic materials science task: predicting the dielectric function of metals, a high-dimensional response that governs how solids interact with light. Using over 200,000 dielectric functions from high-throughput ab initio calculations, we study two multi-objective graph neural networks trained to predict the frequency-dependent complex interband dielectric function and the Drude frequency. We observe broken neural scaling laws with respect to dataset size, whereas scaling with the number of model parameters follows a simple power law that rapidly saturates.

Large Language Models (LLMs) inference is central in modern AI applications, making it critical to understand their energy footprint. Existing approaches typically estimate energy consumption through simple linear functions of input and output sequence lengths, yet our observations reveal clear Energy Efficiency regimes: peak efficiency occurs with short-to-moderate inputs and medium-length outputs, while efficiency drops sharply for long inputs or very short outputs, indicating a non-linear dependency. In this work, we propose an analytical model derived from the computational and memory-access complexity of the Transformer architecture, capable of accurately characterizing the efficiency curve as a function of input and output lengths. To assess its accuracy, we evaluate energy consumption using TensorRT-LLM on NVIDIA H100 GPUs across a diverse set of LLMs ranging from 1B to 9B parameters, including OPT, LLaMA, Gemma, Falcon, Qwen2, and Granite, tested over input and output lengths from 64 to 4096 tokens, achieving a mean MAPE of 1.79%. Our results show that aligning sequence lengths with these efficiency "Sweet Spots" can substantially reduce energy usage, supporting informed truncation, summarization, and adaptive generation strategies in production systems.

Multi-domain machine translation (MDMT) aims to build a unified model capable of translating content across diverse domains. Despite the impressive machine translation capabilities demonstrated by large language models (LLMs), domain adaptation still remains a challenge for LLMs. Existing MDMT methods such as in-context learning and parameter-efficient fine-tuning often suffer from domain shift, parameter interference and limited generalization. In this work, we propose a neuron-efficient fine-tuning framework for MDMT that identifies and updates consensus-aligned neurons within LLMs. These neurons are selected by maximizing the mutual information between neuron behavior and domain features, enabling LLMs to capture both generalizable translation patterns and domain-specific nuances. Our method then fine-tunes LLMs guided by these neurons, effectively mitigating parameter interference and domain-specific overfitting. Comprehensive experiments on three LLMs across ten German-English and Chinese-English translation domains evidence that our method consistently outperforms strong PEFT baselines on both seen and unseen domains, achieving state-of-the-art performance.

Federated Learning is a privacy-preserving decentralized approach for Machine Learning tasks. In industry deployments characterized by a limited number of entities possessing abundant data, the significance of a participant's role in shaping the global model becomes pivotal given that participation in a federation incurs costs, and participants may expect compensation for their involvement. Additionally, the contributions of participants serve as a crucial means to identify and address potential malicious actors and free-riders. However, fairly assessing individual contributions remains a significant hurdle. Recent works have demonstrated a considerable inherent instability in contribution estimations across aggregation strategies. While employing a different strategy may offer convergence benefits, this instability can have potentially harming effects on the willingness of participants in engaging in the federation. In this work, we introduce FedRandom, a novel mitigation technique to the contribution instability problem. Tackling the instability as a statistical estimation problem, FedRandom allows us to generate more samples than when using regular FL strategies. We show that these additional samples provide a more consistent and reliable evaluation of participant contributions. We demonstrate our approach using different data distributions across CIFAR-10, MNIST, CIFAR-100 and FMNIST and show that FedRandom reduces the overall distance to the ground truth by more than a third in half of all evaluated scenarios, and improves stability in more than 90% of cases.

Inaccuracies in existing or generated clinical text may lead to serious adverse consequences, especially if it is a misdiagnosis or incorrect treatment suggestion. With Large Language Models (LLMs) increasingly being used across diverse healthcare applications, comprehensive evaluation through dedicated benchmarks is crucial. However, such datasets remain scarce, especially across diverse languages and contexts. In this paper, we introduce MedErrBench, the first multilingual benchmark for error detection, localization, and correction, developed under the guidance of experienced clinicians. Based on an expanded taxonomy of ten common error types, MedErrBench covers English, Arabic and Chinese, with natural clinical cases annotated and reviewed by domain experts. We assessed the performance of a range of general-purpose, language-specific, and medical-domain language models across all three tasks. Our results reveal notable performance gaps, particularly in non-English settings, highlighting the need for clinically grounded, language-aware systems. By making MedErrBench and our evaluation protocols publicly-available, we aim to advance multilingual clinical NLP to promote safer and more equitable AI-based healthcare globally. The dataset is available in the supplementary material. An anonymized version of the dataset is available at: https://github.com/congboma/MedErrBench.

We propose a new analysis framework for clustering $M$ items into an unknown number of $K$ distinct groups using noisy and actively collected responses. At each time step, an agent is allowed to query pairs of items and observe bandit binary feedback. If the pair of items belongs to the same (resp.\ different) cluster, the observed feedback is $1$ with probability $p>1/2$ (resp.\ $q<1/2$). Leveraging the ubiquitous change-of-measure technique, we establish a fundamental lower bound on the expected number of queries needed to achieve a desired confidence in the clustering accuracy, formulated as a sup-inf optimization problem. Building on this theoretical foundation, we design an asymptotically optimal algorithm in which the stopping criterion involves an empirical version of the inner infimum -- the Generalized Likelihood Ratio (GLR) statistic -- being compared to a threshold. We develop a computationally feasible variant of the GLR statistic and show that its performance gap to the lower bound can be accurately empirically estimated and remains within a constant multiple of the lower bound.

The choice of activation function is an active area of research, with different proposals aimed at improving optimization, while maintaining expressivity. Additionally, the activation function can significantly alter the implicit inductive bias of the architecture, controlling its non-linear behavior. In this paper, in line with previous work, we argue that evolutionary search provides a useful framework for finding new activation functions, while we also make two novel observations. The first is that modern pipelines, such as AlphaEvolve, which relies on frontier LLMs as a mutator operator, allows for a much wider and flexible search space; e.g., over all possible python functions within a certain FLOP budget, eliminating the need for manually constructed search spaces. In addition, these pipelines will be biased towards meaningful activation functions, given their ability to represent common knowledge, leading to a potentially more efficient search of the space. The second observation is that, through this framework, one can target not only performance improvements but also activation functions that encode particular inductive biases. This can be done by using performance on out-of-distribution data as a fitness function, reflecting the degree to which the architecture respects the inherent structure in the data in a manner independent of distribution shifts. We carry an empirical exploration of this proposal and show that relatively small scale synthetic datasets can be sufficient for AlphaEvolve to discover meaningful activations.

Individuals are increasingly generating substantial personal health and lifestyle data, e.g. through wearables and smartphones. While such data could transform preventative care, its integration into clinical practice is hindered by its scale, heterogeneity and the time pressure and data literacy of healthcare professionals (HCPs). We explore how large language models (LLMs) can support sensemaking of patient-generated health data (PGHD) with automated summaries and natural language data exploration. Using cardiovascular disease (CVD) risk reduction as a use case, 16 HCPs reviewed multimodal PGHD in a mixed-methods study with a prototype that integrated common charts, LLM-generated summaries, and a conversational interface. Findings show that AI summaries provided quick overviews that anchored exploration, while conversational interaction supported flexible analysis and bridged data-literacy gaps. However, HCPs raised concerns about transparency, privacy, and overreliance. We contribute empirical insights and sociotechnical design implications for integrating AI-driven summarization and conversation into clinical workflows to support PGHD sensemaking.

This paper introduces a material-aware strength-of-connection measure for smoothed aggregation algebraic multigrid methods, aimed at improving robustness for scalar partial differential equations with heterogeneous and anisotropic material properties. Classical strength-of-connection measures typically rely only on matrix entries or geometric distances, which often fail to capture weak couplings across material interfaces or align with anisotropy directions, ultimately leading to poor convergence. The proposed approach directly incorporates material tensor information into the coarsening process, enabling a reliable detection of weak connections and ensuring that coarse levels preserve the true structure of the underlying problem. As a result, smooth error components are represented properly and sharp coefficient jumps or directional anisotropies are handled consistently. A wide range of academic tests and real-world applications, including thermally activated batteries and solar cells, demonstrate that the proposed method maintains robustness across material contrasts, anisotropies, and mesh variations. Scalability and parallel performance of the algebraic multigrid method highlight the suitability for large-scale, high-performance computing environments.

Partially observable Markov decision processes (POMDPs) are a principled planning model for sequential decision-making under uncertainty. Yet, real-world problems with high-dimensional observations, such as camera images, remain intractable for traditional belief- and filtering-based solvers. To tackle this problem, we introduce the Perception-based Beliefs for POMDPs framework (PBP), which complements such solvers with a perception model. This model takes the form of an image classifier which maps visual observations to probability distributions over states. PBP incorporates these distributions directly into belief updates, so the underlying solver does not need to reason explicitly over high-dimensional observation spaces. We show that the belief update of PBP coincides with the standard belief update if the image classifier is exact. Moreover, to handle classifier imprecision, we incorporate uncertainty quantification and introduce two methods to adjust the belief update accordingly. We implement PBP using two traditional POMDP solvers and empirically show that (1) it outperforms existing end-to-end deep RL methods and (2) uncertainty quantification improves robustness of PBP against visual corruption.

Over the past two decades, research in evolutionary multi-objective optimization has predominantly focused on continuous domains, with comparatively limited attention given to multi-objective combinatorial optimization problems (MOCOPs). Combinatorial problems differ significantly from continuous ones in terms of problem structure and landscape. Recent studies have shown that on MOCOPs multi-objective evolutionary algorithms (MOEAs) can even be outperformed by simple randomised local search. Starting with a randomly sampled solution in search space, randomised local search iteratively draws a random solution (from an archive) to perform local variation within its neighbourhood. However, in most existing methods, the local variation relies on a fixed neighbourhood, which limits exploration and makes the search easy to get trapped in local optima. In this paper, we present a simple yet effective local search method, called variable stepsize randomized local search (VS-RLS), which adjusts the stepsize during the search. VS-RLS transitions gradually from a broad, exploratory search in the early phases to a more focused, fine-grained search as the search progresses. We demonstrate the effectiveness and generalizability of VS-RLS through extensive evaluations against local search and MOEAs methods on diverse MOCOPs.

Advances in AIGC technologies have enabled the synthesis of highly realistic audio deepfakes capable of deceiving human auditory perception. Although numerous audio deepfake detection (ADD) methods have been developed, most rely on local temporal/spectral features or pairwise relations, overlooking high-order interactions (HOIs). HOIs capture discriminative patterns that emerge from multiple feature components beyond their individual contributions. We propose HyperPotter, a hypergraph-based framework that explicitly models these synergistic HOIs through clustering-based hyperedges with class-aware prototype initialization. Extensive experiments demonstrate that HyperPotter surpasses its baseline by an average relative gain of 22.15% across 11 datasets and outperforms state-of-the-art methods by 13.96% on 4 challenging cross-domain datasets, demonstrating superior generalization to diverse attacks and speakers.

Modern science increasingly relies on ever-growing observational datasets and automated inference pipelines, under the implicit belief that accumulating more data makes scientific conclusions more reliable. Here we show that this belief can fail in a fundamental and irreversible way. We identify a structural regime in which standard inference procedures converge smoothly, remain well calibrated, and pass conventional diagnostic checks, yet systematically converge to incorrect conclusions. This failure arises when the reliability of observations degrades in a manner that is intrinsically unobservable to the inference process itself. Using minimal synthetic experiments, we demonstrate that in this regime additional data do not correct error but instead amplify it, while residual-based and goodness-of-fit diagnostics remain misleadingly normal. These results reveal an intrinsic limit of data-driven science: stability, convergence, and confidence are not sufficient indicators of epistemic validity. We argue that inference cannot be treated as an unconditional consequence of data availability, but must instead be governed by explicit constraints on the integrity of the observational process.

Benchmarking the hundreds of functional connectivity (FC) modeling methods on large-scale fMRI datasets is critical for reproducible neuroscience. However, the combinatorial explosion of model-data pairings makes exhaustive evaluation computationally prohibitive, preventing such assessments from becoming a routine pre-analysis step. To break this bottleneck, we reframe the challenge of FC benchmarking by selecting a small, representative core-set whose sole purpose is to preserve the relative performance ranking of FC operators. We formalize this as a ranking-preserving subset selection problem and propose Structure-aware Contrastive Learning for Core-set Selection (SCLCS), a self-supervised framework to select these core-sets. SCLCS first uses an adaptive Transformer to learn each sample's unique FC structure. It then introduces a novel Structural Perturbation Score (SPS) to quantify the stability of these learned structures during training, identifying samples that represent foundational connectivity archetypes. Finally, while SCLCS identifies stable samples via a top-k ranking, we further introduce a density-balanced sampling strategy as a necessary correction to promote diversity, ensuring the final core-set is both structurally robust and distributionally representative. On the large-scale REST-meta-MDD dataset, SCLCS preserves the ground-truth model ranking with just 10% of the data, outperforming state-of-the-art (SOTA) core-set selection methods by up to 23.2% in ranking consistency (nDCG@k). To our knowledge, this is the first work to formalize core-set selection for FC operator benchmarking, thereby making large-scale operators comparisons a feasible and integral part of computational neuroscience. Code is publicly available on https://github.com/lzhan94swu/SCLCS

Memory emerges as the core module in the Large Language Model (LLM)-based agents for long-horizon complex tasks (e.g., multi-turn dialogue, game playing, scientific discovery), where memory can enable knowledge accumulation, iterative reasoning and self-evolution. Among diverse paradigms, graph stands out as a powerful structure for agent memory due to the intrinsic capabilities to model relational dependencies, organize hierarchical information, and support efficient retrieval. This survey presents a comprehensive review of agent memory from the graph-based perspective. First, we introduce a taxonomy of agent memory, including short-term vs. long-term memory, knowledge vs. experience memory, non-structural vs. structural memory, with an implementation view of graph-based memory. Second, according to the life cycle of agent memory, we systematically analyze the key techniques in graph-based agent memory, covering memory extraction for transforming the data into the contents, storage for organizing the data efficiently, retrieval for retrieving the relevant contents from memory to support reasoning, and evolution for updating the contents in the memory. Third, we summarize the open-sourced libraries and benchmarks that support the development and evaluation of self-evolving agent memory. We also explore diverse application scenarios. Finally, we identify critical challenges and future research directions. This survey aims to offer actionable insights to advance the development of more efficient and reliable graph-based agent memory systems. All the related resources, including research papers, open-source data, and projects, are collected for the community in https://github.com/DEEP-PolyU/Awesome-GraphMemory.

The increasing penetration of photovoltaic (PV) generation introduces significant uncertainty into power system operation, necessitating forecasting approaches that extend beyond deterministic point predictions. This paper proposes an any-quantile probabilistic forecasting framework for multi-regional PV power generation based on the Any-Quantile Recurrent Neural Network (AQ-RNN). The model integrates an any-quantile forecasting paradigm with a dual-track recurrent architecture that jointly processes series-specific and cross-regional contextual information, supported by dilated recurrent cells, patch-based temporal modeling, and a dynamic ensemble mechanism. The proposed framework enables the estimation of calibrated conditional quantiles at arbitrary probability levels within a single trained model and effectively exploits spatial dependencies to enhance robustness at the system level. The approach is evaluated using 30 years of hourly PV generation data from 259 European regions and compared against established statistical and neural probabilistic baselines. The results demonstrate consistent improvements in forecast accuracy, calibration, and prediction interval quality, underscoring the suitability of the proposed method for uncertainty-aware energy management and operational decision-making in renewable-dominated power systems.

The study of tail behaviour of SGD-induced processes has been attracting a lot of interest, due to offering strong guarantees with respect to individual runs of an algorithm. While many works provide high-probability guarantees, quantifying the error rate for a fixed probability threshold, there is a lack of work directly studying the probability of failure, i.e., quantifying the tail decay rate for a fixed error threshold. Moreover, existing results are of finite-time nature, limiting their ability to capture the true long-term tail decay which is more informative for modern learning models, typically trained for millions of iterations. Our work closes these gaps, by studying the long-term tail decay of SGD-based methods through the lens of large deviations theory, establishing several strong results in the process. First, we provide an upper bound on the tails of the gradient norm-squared of the best iterate produced by (vanilla) SGD, for non-convex costs and bounded noise, with long-term decay at rate $e^{-t/\log(t)}$. Next, we relax the noise assumption by considering clipped SGD (c-SGD) under heavy-tailed noise with bounded moment of order $p \in (1,2]$, showing an upper bound with long-term decay at rate $e^{-t^{β_p}/\log(t)}$, where $β_p = \frac{4(p-1)}{3p-2}$ for $p \in (1,2)$ and $e^{-t/\log^2(t)}$ for $p = 2$. Finally, we provide lower bounds on the tail decay, at rate $e^{-t}$, showing that our rates for both SGD and c-SGD are tight, up to poly-logarithmic factors. Notably, our results demonstrate an order of magnitude faster long-term tail decay compared to existing work based on finite-time bounds, which show rates $e^{-\sqrt{t}}$ and $e^{-t^{β_p/2}}$, $p \in (1,2]$, for SGD and c-SGD, respectively. As such, we uncover regimes where the tails decay much faster than previously known, providing stronger long-term guarantees for individual runs.

Behavioral evaluation is the dominant paradigm for assessing alignment in large language models (LLMs). In practice, alignment is inferred from performance under finite evaluation protocols - benchmarks, red-teaming suites, or automated pipelines - and observed compliance is often treated as evidence of underlying alignment. This inference step, from behavioral evidence to claims about latent alignment properties, is typically implicit and rarely analyzed as an inference problem in its own right. We study this problem formally. We frame alignment evaluation as an identifiability question under partial observability and allow agent behavior to depend on information correlated with the evaluation regime. Within this setting, we introduce the Alignment Verifiability Problem and the notion of Normative Indistinguishability, capturing when distinct latent alignment hypotheses induce identical distributions over all evaluator-accessible signals. Our main result is a negative but sharply delimited identifiability theorem. Under finite behavioral evaluation and evaluation-aware agents, observed behavioral compliance does not uniquely identify latent alignment. That is, even idealized behavioral evaluation cannot, in general, certify alignment as a latent property. We further show that behavioral alignment tests should be interpreted as estimators of indistinguishability classes rather than verifiers of alignment. Passing increasingly stringent tests may reduce the space of compatible hypotheses, but cannot collapse it to a singleton under the stated conditions. This reframes alignment benchmarks as providing upper bounds on observable compliance within a regime, rather than guarantees of underlying alignment.

Personality is a complex, hierarchical construct typically assessed through item-level questionnaires aggregated into broad trait scores. Personality recognition models aim to infer personality traits from different sources of behavioral data. However, reliance on broad trait scores as ground truth, combined with limited training data, poses challenges for generalization, as similar trait scores can manifest through diverse, context dependent behaviors. In this work, we explore the predictive impact of the more granular hierarchical levels of the Big-Five Personality Model, facets and nuances, to enhance personality recognition from audiovisual interaction data. Using the UDIVA v0.5 dataset, we trained a transformer-based model including cross-modal (audiovisual) and cross-subject (dyad-aware) attention mechanisms. Results show that nuance-level models consistently outperform facet and trait-level models, reducing mean squared error by up to 74% across interaction scenarios.

The long-standing dominance of gradient-boosted decision trees for tabular data has recently been challenged by in-context learning tabular foundation models. In-context learning methods fit and predict in one forward pass without parameter updates by leveraging the training data as context for predicting on query test points. While recent tabular foundation models achieve state-of-the-art performance, their transformer architecture based on the attention mechanism has quadratic complexity regarding dataset size, which in turn increases the overhead on training and inference time, and limits the capacity of the models to handle large-scale datasets. In this work, we propose TACO, an end-to-end tabular compression model that compresses the training dataset in a latent space. We test our method on the TabArena benchmark, where our proposed method is up to 94x faster in inference time, while consuming up to 97\% less memory compared to the state-of-the-art tabular transformer architecture, all while retaining performance without significant degradation. Lastly, our method not only scales better with increased dataset sizes, but it also achieves better performance compared to other baselines.

We investigate how transformer models represent complex verb paradigms in Turkish and Modern Hebrew, concentrating on how tokenization strategies shape this ability. Using the Blackbird Language Matrices task on natural data, we show that for Turkish -- with its transparent morphological markers -- both monolingual and multilingual models succeed, either when tokenization is atomic or when it breaks words into small subword units. For Hebrew, instead, monolingual and multilingual models diverge. A multilingual model using character-level tokenization fails to capture the language non-concatenative morphology, but a monolingual model with morpheme-aware segmentation performs well. Performance improves on more synthetic datasets, in all models.

While existing time series foundation models primarily rely on large-scale unimodal pretraining, they lack complementary modalities to enhance time series understanding. Building multimodal foundation models is a natural next step, but it faces key challenges: 1) lack of a unified multimodal pretraining paradigm and large-scale multimodal corpora for time series analysis; 2) how to effectively integrate heterogeneous modalities and enhance model generalization. To address these challenges, we take an early step toward multimodal foundation models for time series analysis. We first propose a multimodal pretraining paradigm that leverages time series with endogenous modalities (derived images and text) and exogenous knowledge (real-world news), providing a comprehensive multi-view perspective for time series analysis. To support this, we develop an automated data construction pipeline to curate MM-TS, the first large-scale multimodal time series dataset spanning six domains, with up to one billion points. Then we propose HORAI, a frequency-enhanced multimodal foundation model. It integrates two core components: the Frequency-enhanced Cross-Modality Encoder and the Time-Frequency Decoder, designed to effectively fuse multimodal features and enhance model generalization across modalities and domains. After pretraining on MM-TS, HORAI achieves state-of-the-art zero-shot performance on time series forecasting and anomaly detection tasks, demonstrating strong generalization.

This paper proposes an Improved Noisy Deep Q-Network (Noisy DQN) to enhance the exploration and stability of Unmanned Aerial Vehicle (UAV) when applying deep reinforcement learning in simulated environments. This method enhances the exploration ability by combining the residual NoisyLinear layer with an adaptive noise scheduling mechanism, while improving training stability through smooth loss and soft target network updates. Experiments show that the proposed model achieves faster convergence and up to $+40$ higher rewards compared to standard DQN and quickly reach to the minimum number of steps required for the task 28 in the 15 * 15 grid navigation environment set up. The results show that our comprehensive improvements to the network structure of NoisyNet, exploration control, and training stability contribute to enhancing the efficiency and reliability of deep Q-learning.

We introduce Variational Joint Embedding (VJE), a framework that synthesizes joint embedding and variational inference to enable self-supervised learning of probabilistic representations in a reconstruction-free, non-contrastive setting. Compared to energy-based predictive objectives that optimize pointwise discrepancies, VJE maximizes a symmetric conditional evidence lower bound (ELBO) for a latent-variable model defined directly on encoder embeddings. We instantiate the conditional likelihood with a heavy-tailed Student-$t$ model using a polar decomposition that explicitly decouples directional and radial factors to prevent norm-induced instabilities during training. VJE employs an amortized inference network to parameterize a diagonal Gaussian variational posterior whose feature-wise variances are shared with the likelihood scale to capture anisotropic uncertainty without auxiliary projection heads. Across ImageNet-1K, CIFAR-10/100, and STL-10, VJE achieves performance comparable to standard non-contrastive baselines under linear and k-NN evaluation. We further validate these probabilistic semantics through one-class CIFAR-10 anomaly detection, where likelihood-based scoring under the proposed model outperforms comparable self-supervised baselines.

Traditional ontologies excel at describing domain structure but cannot generate novel artifacts. Large language models generate fluently but produce outputs that lack structural validity, hallucinating mechanisms without components, goals without end conditions. We introduce Generative Ontology, a framework that synthesizes these complementary strengths: ontology provides the grammar; the LLM provides the creativity. Generative Ontology encodes domain knowledge as executable Pydantic schemas that constrain LLM generation via DSPy signatures. A multi-agent pipeline assigns specialized roles to different ontology domains: a Mechanics Architect designs game systems, a Theme Weaver integrates narrative, a Balance Critic identifies exploits. Each agent carrying a professional "anxiety" that prevents shallow, agreeable outputs. Retrieval-augmented generation grounds novel designs in precedents from existing exemplars, while iterative validation ensures coherence between mechanisms and components. We demonstrate the framework through GameGrammar, a system for generating complete tabletop game designs. Given a thematic prompt ("bioluminescent fungi competing in a cave ecosystem"), the pipeline produces structurally complete, playable game specifications with mechanisms, components, victory conditions, and setup instructions. These outputs satisfy ontological constraints while remaining genuinely creative. The pattern generalizes beyond games. Any domain with expert vocabulary, validity constraints, and accumulated exemplars (music composition, software architecture, culinary arts) is a candidate for Generative Ontology. We argue that constraints do not limit creativity but enable it: just as grammar makes poetry possible, ontology makes structured generation possible.

Selective unlearning and long-horizon extrapolation remain fragile in modern neural networks, even when tasks have underlying algebraic structure. In this work, we argue that these failures arise not solely from optimization or unlearning algorithms, but from how models structure their internal representations during training. We explore if having explicit multiplicative interactions as an architectural inductive bias helps in structural disentanglement, through Bilinear MLPs. We show analytically that bilinear parameterizations possess a `non-mixing' property under gradient flow conditions, where functional components separate into orthogonal subspace representations. This provides a mathematical foundation for surgical model modification. We validate this hypothesis through a series of controlled experiments spanning modular arithmetic, cyclic reasoning, Lie group dynamics, and targeted unlearning benchmarks. Unlike pointwise nonlinear networks, multiplicative architectures are able to recover true operators aligned with the underlying algebraic structure. Our results suggest that model editability and generalization are constrained by representational structure, and that architectural inductive bias plays a central role in enabling reliable unlearning.

Large language models (LLMs) have advanced the development of personalized learning in education. However, their inherent generation mechanisms often produce homogeneous responses to identical prompts. This one-size-fits-all mechanism overlooks the substantial heterogeneity in students cognitive and psychological, thereby posing potential safety risks to vulnerable groups. Existing safety evaluations primarily rely on context-independent metrics such as factual accuracy, bias, or toxicity, which fail to capture the divergent harms that the same response might cause across different student attributes. To address this gap, we propose the concept of Student-Tailored Personalized Safety and construct CASTLE based on educational theories. This benchmark covers 15 educational safety risks and 14 student attributes, comprising 92,908 bilingual scenarios. We further design three evaluation metrics: Risk Sensitivity, measuring the model ability to detect risks; Emotional Empathy, evaluating the model capacity to recognize student states; and Student Alignment, assessing the match between model responses and student attributes. Experiments on 18 SOTA LLMs demonstrate that CASTLE poses a significant challenge: all models scored below an average safety rating of 2.3 out of 5, indicating substantial deficiencies in personalized safety assurance.

Reinforcement Learning with Verifiable Rewards has recently advanced the capabilities of Large Language Models in complex reasoning tasks by providing explicit rule-based supervision. Among RLVR methods, GRPO and its variants have achieved strong empirical performance. Despite their success, we identify that they suffer from Gradient Misassignment in Positives and Gradient Domination in Negatives, which lead to inefficient and suboptimal policy updates. To address these issues, we propose Rewards as Labels (REAL), a novel framework that revisits verifiable rewards as categorical labels rather than scalar weights, thereby reformulating policy optimization as a classification problem. Building on this, we further introduce anchor logits to enhance policy learning. Our analysis reveals that REAL induces a monotonic and bounded gradient weighting, enabling balanced gradient allocation across rollouts and effectively mitigating the identified mismatches. Extensive experiments on mathematical reasoning benchmarks show that REAL improves training stability and consistently outperforms GRPO and strong variants such as DAPO. On the 1.5B model, REAL improves average Pass@1 over DAPO by 6.7%. These gains further scale to 7B model, REAL continues to outperform DAPO and GSPO by 6.2% and 1.7%, respectively. Notably, even with a vanilla binary cross-entropy, REAL remains stable and exceeds DAPO by 4.5% on average.

Background: Empathy is widely recognized for improving patient outcomes, including reduced pain and anxiety and improved satisfaction, and its absence can cause harm. Meanwhile, use of artificial intelligence (AI)-based chatbots in healthcare is rapidly expanding, with one in five general practitioners using generative AI to assist with tasks such as writing letters. Some studies suggest AI chatbots can outperform human healthcare professionals (HCPs) in empathy, though findings are mixed and lack synthesis. Sources of data: We searched multiple databases for studies comparing AI chatbots using large language models with human HCPs on empathy measures. We assessed risk of bias with ROBINS-I and synthesized findings using random-effects meta-analysis where feasible, whilst avoiding double counting. Areas of agreement: We identified 15 studies (2023-2024). Thirteen studies reported statistically significantly higher empathy ratings for AI, with only two studies situated in dermatology favouring human responses. Of the 15 studies, 13 provided extractable data and were suitable for pooling. Meta-analysis of those 13 studies, all utilising ChatGPT-3.5/4, showed a standardized mean difference of 0.87 (95% CI, 0.54-1.20) favouring AI (P < .00001), roughly equivalent to a two-point increase on a 10-point scale. Areas of controversy: Studies relied on text-based assessments that overlook non-verbal cues and evaluated empathy through proxy raters. Growing points: Our findings indicate that, in text-only scenarios, AI chatbots are frequently perceived as more empathic than human HCPs. Areas timely for developing research: Future research should validate these findings with direct patient evaluations and assess whether emerging voice-enabled AI systems can deliver similar empathic advantages.

Exact inference in complex probabilistic models often incurs prohibitive computational costs. This challenge is particularly acute for autonomous agents in dynamic environments that require frequent, real-time belief updates. Existing methods are often inefficient for ongoing reasoning, as they re-evaluate the entire model upon any change, failing to exploit that real-world information streams have heterogeneous update rates. To address this, we approach the problem from a reactive, asynchronous, probabilistic reasoning perspective. We first introduce Resin (Reactive Signal Inference), a probabilistic programming language that merges probabilistic logic with reactive programming. Furthermore, to provide efficient and exact semantics for Resin, we propose Reactive Circuits (RCs). Formulated as a meta-structure over Algebraic Circuits and asynchronous data streams, RCs are time-dynamic Directed Acyclic Graphs that autonomously adapt themselves based on the volatility of input signals. In high-fidelity drone swarm simulations, our approach achieves several orders of magnitude of speedup over frequency-agnostic inference. We demonstrate that RCs' structural adaptations successfully capture environmental dynamics, significantly reducing latency and facilitating reactive real-time reasoning. By partitioning computations based on the estimated Frequency of Change in the asynchronous inputs, large inference tasks can be decomposed into individually memoized sub-problems. This ensures that only the specific components of a model affected by new information are re-evaluated, drastically reducing redundant computation in streaming contexts.

Mode-dependent architectural components (layers that behave differently during training and evaluation, such as Batch Normalization or dropout) are commonly used in visual reinforcement learning but can destabilize on-policy optimization. We show that in Proximal Policy Optimization (PPO), discrepancies between training and evaluation behavior induced by Batch Normalization lead to policy mismatch, distributional drift, and reward collapse. We propose Mode-Dependent Rectification (MDR), a lightweight dual-phase training procedure that stabilizes PPO under mode-dependent layers without architectural changes. Experiments across procedurally generated games and real-world patch-localization tasks demonstrate that MDR consistently improves stability and performance, and extends naturally to other mode-dependent layers.

Dataset distillation compresses large datasets into compact synthetic sets with comparable performance in training models. Despite recent progress on diffusion-based distillation, this type of method typically depends on heuristic guidance or prototype assignment, which comes with time-consuming sampling and trajectory instability and thus hurts downstream generalization especially under strong control or low IPC. We propose \emph{Path-Guided Flow Matching (PGFM)}, the first flow matching-based framework for generative distillation, which enables fast deterministic synthesis by solving an ODE in a few steps. PGFM conducts flow matching in the latent space of a frozen VAE to learn class-conditional transport from Gaussian noise to data distribution. Particularly, we develop a continuous path-to-prototype guidance algorithm for ODE-consistent path control, which allows trajectories to reliably land on assigned prototypes while preserving diversity and efficiency. Extensive experiments across high-resolution benchmarks demonstrate that PGFM matches or surpasses prior diffusion-based distillation approaches with fewer steps of sampling while delivering competitive performance with remarkably improved efficiency, e.g., 7.6$\times$ more efficient than the diffusion-based counterparts with 78\% mode coverage.

Diffusion Transformers (DiTs) incur prohibitive computational costs due to the quadratic scaling of self-attention. Existing pruning methods fail to simultaneously satisfy differentiability, efficiency, and the strict static budgets required for hardware overhead. To address this, we propose Shiva-DiT, which effectively reconciles these conflicting requirements via Residual-Based Differentiable Top-$k$ Selection. By leveraging a residual-aware straight-through estimator, our method enforces deterministic token counts for static compilation while preserving end-to-end learnability through residual gradient estimation. Furthermore, we introduce a Context-Aware Router and Adaptive Ratio Policy to autonomously learn an adaptive pruning schedule. Experiments on mainstream models, including SD3.5, demonstrate that Shiva-DiT establishes a new Pareto frontier, achieving a 1.54$\times$ wall-clock speedup with superior fidelity compared to existing baselines, effectively eliminating ragged tensor overheads.

Stochastic Gradient Descent (SGD) introduces anisotropic noise that is correlated with the local curvature of the loss landscape, thereby biasing optimization toward flat minima. Prior work often assumes an equivalence between the Fisher Information Matrix and the Hessian for negative log-likelihood losses, leading to the claim that the SGD noise covariance $\mathbf{C}$ is proportional to the Hessian $\mathbf{H}$. We show that this assumption holds only under restrictive conditions that are typically violated in deep neural networks. Using the recently discovered Activity--Weight Duality, we find a more general relationship agnostic to the specific loss formulation, showing that $\mathbf{C} \propto \mathbb{E}_p[\mathbf{h}_p^2]$, where $\mathbf{h}_p$ denotes the per-sample Hessian with $\mathbf{H} = \mathbb{E}_p[\mathbf{h}_p]$. As a consequence, $\mathbf{C}$ and $\mathbf{H}$ commute approximately rather than coincide exactly, and their diagonal elements follow an approximate power-law relation $C_{ii} \propto H_{ii}^γ$ with a theoretically bounded exponent $1 \leq γ\leq 2$, determined by per-sample Hessian spectra. Experiments across datasets, architectures, and loss functions validate these bounds, providing a unified characterization of the noise-curvature relationship in deep learning.

Despite remarkable advances in natural language processing, developing effective systems for low-resource languages remains a formidable challenge, with performances typically lagging far behind high-resource counterparts due to data scarcity and insufficient linguistic resources. Cross-lingual knowledge transfer has emerged as a promising approach to address this challenge by leveraging resources from high-resource languages. In this paper, we investigate methods for transferring linguistic knowledge from high-resource languages to low-resource languages, where the number of labeled training instances is in hundreds. We focus on sentence-level and word-level tasks. We introduce a novel method, GETR (Graph-Enhanced Token Representation) for cross-lingual knowledge transfer along with two adopted baselines (a) augmentation in hidden layers and (b) token embedding transfer through token translation. Experimental results demonstrate that our GNN-based approach significantly outperforms existing multilingual and cross-lingual baseline methods, achieving 13 percentage point improvements on truly low-resource languages (Mizo, Khasi) for POS tagging, and 20 and 27 percentage point improvements in macro-F1 on simulated low-resource languages (Marathi, Bangla, Malayalam) across sentiment classification and NER tasks respectively. We also present a detailed analysis of the transfer mechanisms and identify key factors that contribute to successful knowledge transfer in this linguistic context.

Vision Transformers (ViTs) have demonstrated strong performance across a range of computer vision tasks by modeling long-range spatial interactions via self-attention. However, channel-wise mixing in ViTs remains static, relying on fixed multilayer perceptrons (MLPs) that lack adaptability to input content. We introduce 'CAViT', a dual-attention architecture that replaces the static MLP with a dynamic, attention-based mechanism for feature interaction. Each Transformer block in CAViT performs spatial self-attention followed by channel-wise self-attention, allowing the model to dynamically recalibrate feature representations based on global image context. This unified and content-aware token mixing strategy enhances representational expressiveness without increasing depth or complexity. We validate CAViT across five benchmark datasets spanning both natural and medical domains, where it outperforms the standard ViT baseline by up to +3.6% in accuracy, while reducing parameter count and FLOPs by over 30%. Qualitative attention maps reveal sharper and semantically meaningful activation patterns, validating the effectiveness of our attention-driven token mixing.

Cognitive biases often shape human decisions. While large language models (LLMs) have been shown to reproduce well-known biases, a more critical question is whether LLMs can predict biases at the individual level and emulate the dynamics of biased human behavior when contextual factors, such as cognitive load, interact with these biases. We adapted three well-established decision scenarios into a conversational setting and conducted a human experiment (N=1100). Participants engaged with a chatbot that facilitates decision-making through simple or complex dialogues. Results revealed robust biases. To evaluate how LLMs emulate human decision-making under similar interactive conditions, we used participant demographics and dialogue transcripts to simulate these conditions with LLMs based on GPT-4 and GPT-5. The LLMs reproduced human biases with precision. We found notable differences between models in how they aligned human behavior. This has important implications for designing and evaluating adaptive, bias-aware LLM-based AI systems in interactive contexts.

Robust Markov decision processes (MDPs) have attracted significant interest due to their ability to protect MDPs from poor out-of-sample performance in the presence of ambiguity. In contrast to classical MDPs, which account for stochasticity by modeling the dynamics through a stochastic process with a known transition kernel, a robust MDP additionally accounts for ambiguity by optimizing against the most adverse transition kernel from an ambiguity set constructed via historical data. In this paper, we develop a unified solution framework for a broad class of robust MDPs with $s$-rectangular ambiguity sets, where the most adverse transition probabilities are considered independently for each state. Using our algorithms, we show that $s$-rectangular robust MDPs with $1$- and $2$-norm as well as $φ$-divergence ambiguity sets can be solved several orders of magnitude faster than with state-of-the-art commercial solvers, and often only a logarithmic factor slower than classical MDPs. We demonstrate the favorable scaling properties of our algorithms on a range of synthetically generated as well as standard benchmark instances.

Multimodal-Attributed Graph (MAG) learning has achieved remarkable success in modeling complex real-world systems by integrating graph topology with rich attributes from multiple modalities. With the rapid proliferation of novel MAG models capable of handling intricate cross-modal semantics and structural dependencies, establishing a rigorous and unified evaluation standard has become imperative. Although existing benchmarks have facilitated initial progress, they exhibit critical limitations in domain coverage, encoder flexibility, model diversity, and task scope, presenting significant challenges to fair evaluation. To bridge this gap, we present OpenMAG, a comprehensive benchmark that integrates 19 datasets across 6 domains and incorporates 16 encoders to support both static and trainable feature encoding. OpenMAG further implements a standardized library of 24 state-of-the-art models and supports 8 downstream tasks, enabling fair comparisons within a unified framework. Through systematic assessment of necessity, data quality, effectiveness, robustness, and efficiency, we derive 14 fundamental insights into MAG learning to guide future advancements. Our code is available at https://github.com/YUKI-N810/OpenMAG.

Structural shifts pose a significant challenge for graph neural networks, as graph topology acts as a covariate that can vary across domains. Existing domain generalization methods rely on fixed structural augmentations or training on globally perturbed graphs, mechanisms that do not pinpoint which specific edges encode domain-invariant information. We argue that domain-invariant structural information is not rigidly tied to a single topology but resides in the consensus across multiple graph structures derived from topology and feature similarity. To capture this, we first propose EdgeMask-DG, a novel min-max algorithm where an edge masker learns to find worst-case continuous masks subject to a sparsity constraint, compelling a task GNN to perform effectively under these adversarial structural perturbations. Building upon this, we introduce EdgeMask-DG*, an extension that applies this adversarial masking principle to an enriched graph. This enriched graph combines the original topology with feature-derived edges, allowing the model to discover invariances even when the original topology is noisy or domain-specific. EdgeMask-DG* is the first to systematically combine adaptive adversarial topology search with feature-enriched graphs. We provide a formal justification for our approach from a robust optimization perspective. We demonstrate that EdgeMask-DG* achieves new state-of-the-art performance on diverse graph domain generalization benchmarks, including citation networks, social networks, and temporal graphs. Notably, on the Cora OOD benchmark, EdgeMask-DG* lifts the worst-case domain accuracy to 78.0\%, a +3.8 pp improvement over the prior state of the art (74.2\%). The source code for our experiments can be found here: https://anonymous.4open.science/r/TMLR-EAEF/

Humans excel at spatial reasoning tasks like Tangram puzzle assembly through cognitive processes involving mental rotation, iterative refinement, and visual feedback. Inspired by how humans solve Tangram puzzles through trial-and-error, observation, and correction, we design a framework that models these human cognitive mechanisms. However, comprehensive experiments across five representative Vision-Language Models (VLMs) reveal systematic failures in continuous geometric reasoning: average IoU of only 0.41 on single-piece tasks, dropping to 0.23 on two-piece composition, far below human performance where children can complete Tangram tasks successfully. This paper addresses a fundamental challenge in self-improving AI: can models iteratively refine their predictions at test time without parameter updates? We introduce a test-time self-refinement framework that combines in-context learning (ICL) with reward-guided feedback loops, inspired by human cognitive processes. Our training-free verifier-refiner agent applies recursive refinement loops that iteratively self-refine predictions based on geometric consistency feedback, achieving IoU improvements from 0.63 to 0.932 on medium-triangle cases without any model retraining. This demonstrates that incorporating human-inspired iterative refinement mechanisms through ICL and reward loops can substantially enhance geometric reasoning in VLMs, moving self-improving AI from promise to practice in continuous spatial domains. Our work is available at this anonymous link https://anonymous.4open.science/r/TangramVLM-F582/.

Pre-trained graph neural networks (GNNs) transfer well, but adapting them to downstream tasks remains challenging due to mismatches between pre-training objectives and task requirements. Graph prompt tuning offers a parameter-efficient alternative to fine-tuning, yet most methods only modify inputs or representations and leave message passing unchanged, limiting their ability to adapt neighborhood interactions. We propose message-adaptive graph prompt tuning, which injects learnable prompts into the message passing step to reweight incoming neighbor messages and add task-specific prompt vectors during message aggregation, while keeping the backbone GNN frozen. The approach is compatible with common GNN backbones and pre-training strategies, and applicable across downstream settings. Experiments on diverse node- and graph-level datasets show consistent gains over prior graph prompting methods in few-shot settings, while achieving performance competitive with fine-tuning in full-shot regimes.

ArkTS is a core programming language in the OpenHarmony ecosystem, yet research on ArkTS code intelligence is hindered by the lack of public datasets and evaluation benchmarks. This paper presents a large-scale ArkTS dataset constructed from open-source repositories, targeting code retrieval and code evaluation tasks. We design a single-search task, where natural language comments are used to retrieve corresponding ArkTS functions. ArkTS repositories are crawled from GitHub and Gitee, and comment-function pairs are extracted using tree-sitter-arkts, followed by cross-platform deduplication and statistical analysis of ArkTS function types. We further evaluate all existing open-source code embedding models on the single-search task and perform fine-tuning using both ArkTS and TypeScript training datasets, resulting in a high-performing model for ArkTS code understanding. This work establishes the first systematic benchmark for ArkTS code retrieval. Both the dataset and our fine-tuned model will be released publicly and are available at https://huggingface.co/hreyulog/embedinggemma_arkts and https://huggingface.co/datasets/hreyulog/arkts-code-docstring,establishing the first systematic benchmark for ArkTS code retrieval.

We propose LOGDIFF (Logical Guidance for the Exact Composition of Diffusion Models), a guidance framework for diffusion models that enables principled constrained generation with complex logical expressions at inference time. We study when exact score-based guidance for complex logical formulas can be obtained from guidance signals associated with atomic properties. First, we derive an exact Boolean calculus that provides a sufficient condition for exact logical guidance. Specifically, if a formula admits a circuit representation in which conjunctions combine conditionally independent subformulas and disjunctions combine subformulas that are either conditionally independent or mutually exclusive, exact logical guidance is achievable. In this case, the guidance signal can be computed exactly from atomic scores and posterior probabilities using an efficient recursive algorithm. Moreover, we show that, for commonly encountered classes of distributions, any desired Boolean formula is compilable into such a circuit representation. Second, by combining atomic guidance scores with posterior probability estimates, we introduce a hybrid guidance approach that bridges classifierguidance and classifier-free guidance, applicable to both compositional logical guidance and standard conditional generation. We demonstrate the effectiveness of our framework on multiple image and protein structure generation tasks.

Reinforcement Learning with Verifiable Rewards (RLVR), particularly GRPO, has become the standard for eliciting LLM reasoning. However, its efficiency in exploration and difficulty adaptation remains an open challenge. In this work, we argue that these bottlenecks stem from an implicit advantage symmetry inherent in Group Relative Advantage Estimation (GRAE). This symmetry induces two critical limitations: (i) at the group level, strict symmetry in weights between correct and incorrect trajectories leaves unsampled action logits unchanged, thereby hindering exploration of novel correct solution. (ii) at the sample level, the algorithm implicitly prioritizes medium-difficulty samples, remaining agnostic to the non-stationary demands of difficulty focus. Through controlled experiments, we reveal that this symmetric property is sub-optimal, yielding two pivotal insights: (i) asymmetrically suppressing the advantages of correct trajectories encourages essential exploration. (ii) learning efficiency is maximized by a curriculum-like transition-prioritizing simpler samples initially before gradually shifting to complex ones. Motivated by these findings, we propose Asymmetric GRAE (A-GRAE), which dynamically modulates exploration incentives and sample-difficulty focus. Experiments across seven benchmarks demonstrate that A-GRAE consistently improves GRPO and its variants across both LLMs and MLLMs.

RL-based post-training with GRPO is widely used to improve large language models on individual reasoning tasks. However, real-world deployment requires reliable performance across diverse tasks. A straightforward multi-task adaptation of GRPO often leads to imbalanced outcomes, with some tasks dominating optimization while others stagnate. Moreover, tasks can vary widely in how frequently prompts yield zero advantages (and thus zero gradients), which further distorts their effective contribution to the optimization signal. To address these issues, we propose a novel Multi-Task GRPO (MT-GRPO) algorithm that (i) dynamically adapts task weights to explicitly optimize worst-task performance and promote balanced progress across tasks, and (ii) introduces a ratio-preserving sampler to ensure task-wise policy gradients reflect the adapted weights. Experiments on both 3-task and 9-task settings show that MT-GRPO consistently outperforms baselines in worst-task accuracy. In particular, MT-GRPO achieves 16-28% and 6% absolute improvement on worst-task performance over standard GRPO and DAPO, respectively, while maintaining competitive average accuracy. Moreover, MT-GRPO requires 50% fewer training steps to reach 50% worst-task accuracy in the 3-task setting, demonstrating substantially improved efficiency in achieving reliable performance across tasks.

Large Language Models (LLMs) exhibit potential for explainable recommendation systems but overlook collaborative signals, while prevailing methods treat recommendation and explanation as separate tasks, resulting in a memory footprint. We present RGCF-XRec, a hybrid framework that introduces reasoning-guided collaborative filtering (CF) knowledge into a language model to deliver explainable sequential recommendations in a single step. Theoretical grounding and empirical findings reveal that RGCF-XRec offers three key merits over leading CF-aware LLM-based methods: (1) reasoning-guided augmentation of CF knowledge through contextual prompting to discover latent preferences and interpretable reasoning paths; (2) an efficient scoring mechanism based on four dimensions: coherence, completeness, relevance, and consistency to mitigate noisy CF reasoning traces and retain high-quality explanations; (3) a unified representation learning network that encodes collaborative and semantic signals, enabling a structured prompt to condition the LLM for explainable sequential recommendation. RGCF-XRec demonstrates consistent improvements across Amazon datasets, Sports, Toys, and Beauty, comprising 642,503 user-item interactions. It improves HR@10 by 7.38\% in Sports and 4.59\% in Toys, along with ROUGE-L by 8.02\% and 3.49\%, respectively. It reduces the cold warm performance gap, achieving overall gains of 14.5\% in cold-start and 11.9\% in warm start scenarios, and enhances zero-shot HR@5 by 18.54\% in Beauty and 23.16\% in Toys, highlighting effective generalization and robustness. Moreover, RGCF-XRec achieves training efficiency with a lightweight LLaMA 3.2-3B backbone, ensuring scalability for real-world applications.

Matrix multiplication is a fundamental classical computing operation whose efficiency becomes a major challenge at scale, especially for machine learning applications. Quantum computing, with its inherent parallelism and exponential storage capacity, offers a potential solution to these limitations. This work presents a quantum kernel-based matrix multiplication algorithm (QKMM) that achieves an asymptotically optimal computational complexity of $ O(N^2 \log_2 N) $, outperforming the classical optimal complexity of $ O(N^{2.371552}) $, where $N$ denotes the matrix dimension. Through noiseless and noisy quantum simulation experiments, we demonstrate that the proposed algorithm not only exhibits superior theoretical efficiency but also shows practical advantages in runtime performance and stability.

Large Reasoning Models (LRMs) excel at complex reasoning tasks, but their efficiency is often hampered by overly verbose outputs. Prior steering methods attempt to address this issue by applying a single, global vector to hidden representations -- an approach grounded in the restrictive linear representation hypothesis. In this work, we introduce FlowSteer, a nonlinear steering method that goes beyond uniform linear shifts by learning a complete transformation between the distributions associated with verbose and concise reasoning. This transformation is learned via Flow Matching as a velocity field, enabling precise, input-dependent control over the model's reasoning process. By aligning steered representations with the distribution of concise-reasoning activations, FlowSteer yields more compact reasoning than the linear shifts. Across diverse reasoning benchmarks, FlowSteer demonstrates strong task performance and token efficiency compared to leading inference-time baselines. Our work demonstrates that modeling the full distributional transport with generative techniques offers a more effective and principled foundation for controlling LRMs.

Model merging combines multiple fine-tuned models into a single model by adding their weight updates, providing a lightweight alternative to retraining. Existing methods primarily target resolving conflicts between task updates, leaving the failure mode of over-counting shared knowledge unaddressed. We show that when tasks share aligned spectral directions (i.e., overlapping singular vectors), a simple linear combination repeatedly accumulates these directions, inflating the singular values and biasing the merged model toward shared subspaces. To mitigate this issue, we propose Singular Value Calibration (SVC), a training-free and data-free post-processing method that quantifies subspace overlap and rescales inflated singular values to restore a balanced spectrum. Across vision and language benchmarks, SVC consistently improves strong merging baselines and achieves state-of-the-art performance. Furthermore, by modifying only the singular values, SVC improves the performance of Task Arithmetic by 13.0%. Code is available at: https://github.com/lyymuwu/SVC.

Large vision-language models (LVLMs) have shown substantial advances in multimodal understanding and generation. However, when presented with incompetent or adversarial inputs, they frequently produce unreliable or even harmful content, such as fact hallucinations or dangerous instructions. This misalignment with human expectations, referred to as \emph{misbehaviors} of LVLMs, raises serious concerns for deployment in critical applications. These misbehaviors are found to stem from epistemic uncertainty, specifically either conflicting internal knowledge or the absence of supporting information. However, existing uncertainty quantification methods, which typically capture only overall epistemic uncertainty, have shown limited effectiveness in identifying such issues. To address this gap, we propose Evidential Uncertainty Quantification (EUQ), a fine-grained method that captures both information conflict and ignorance for effective detection of LVLM misbehaviors. In particular, we interpret features from the model output head as either supporting (positive) or opposing (negative) evidence. Leveraging Evidence Theory, we model and aggregate this evidence to quantify internal conflict and knowledge gaps within a single forward pass. We extensively evaluate our method across four categories of misbehavior, including hallucinations, jailbreaks, adversarial vulnerabilities, and out-of-distribution (OOD) failures, using state-of-the-art LVLMs, and find that EUQ consistently outperforms strong baselines, showing that hallucinations correspond to high internal conflict and OOD failures to high ignorance. Furthermore, layer-wise evidential uncertainty dynamics analysis helps interpret the evolution of internal representations from a new perspective. The source code is available at https://github.com/HT86159/EUQ.

We study conditional generation in diffusion models under hard constraints, where generated samples must satisfy prescribed events with probability one. Such constraints arise naturally in safety-critical applications and in rare-event simulation, where soft or reward-based guidance methods offer no guarantee of constraint satisfaction. Building on a probabilistic interpretation of diffusion models, we develop a principled conditional diffusion guidance framework based on Doob's h-transform, martingale representation and quadratic variation process. Specifically, the resulting guided dynamics augment a pretrained diffusion with an explicit drift correction involving the logarithmic gradient of a conditioning function, without modifying the pretrained score network. Leveraging martingale and quadratic-variation identities, we propose two novel off-policy learning algorithms based on a martingale loss and a martingale-covariation loss to estimate h and its gradient using only trajectories from the pretrained model. We provide non-asymptotic guarantees for the resulting conditional sampler in both total variation and Wasserstein distances, explicitly characterizing the impact of score approximation and guidance estimation errors. Numerical experiments demonstrate the effectiveness of the proposed methods in enforcing hard constraints and generating rare-event samples.

Detecting misalignment in large language models is challenging because models may learn to conceal misbehavior during training. Standard auditing techniques fall short: black-box methods often cannot distinguish misaligned outputs from benign ones, and mechanistic interpretability does not scale with model capabilities. We introduce Split Personality Training (SPT), which fine-tunes a second ``honest persona'' into LoRA parameters that remain inactive during normal operation. After the main model responds, we activate the LoRA adapter and insert a trigger string, enabling the honest persona to review the response while accessing the main model's latent states. We test our method on the Anthropic Auditing Game Model Organism, a benchmark where Llama-3.3-70B is trained to exploit reward hacks while concealing this behavior. SPT achieves 96% overall accuracy, whereas Anthropic reports near 0% accuracy. The honest persona reveals latent knowledge inaccessible to external observers, such as the fictional biases the compromised model was trained on.

We analyze algorithms for solving stochastic variational inequalities (VI) without the bounded variance or bounded domain assumptions, where our main focus is min-max optimization with possibly unbounded constraint sets. We focus on two classes of problems: monotone VIs; and structured nonmonotone VIs that admit a solution to the weak Minty VI. The latter assumption allows us to solve structured nonconvex-nonconcave min-max problems. For both classes of VIs, to make the expected residual norm less than $\varepsilon$, we show an oracle complexity of $\widetilde{O}(\varepsilon^{-4})$, which is the best-known for constrained VIs. In our setting, this complexity had been obtained with the bounded variance assumption in the literature, which is not even satisfied for bilinear min-max problems with an unbounded domain. We obtain this complexity for stochastic oracles whose variance can grow as fast as the squared norm of the optimization variable.

This study investigates large language model (LLM) -based multi-agent systems (MASs) as a promising approach to inventory management, which is a key component of supply chain management. Although these systems have gained considerable attention for their potential to address the challenges associated with typical inventory management methods, key uncertainties regarding their effectiveness persist. Specifically, it is unclear whether LLM-based MASs can consistently derive optimal ordering policies and adapt to diverse supply chain scenarios. To address these questions, we examine an LLM-based MAS with a fixed-ordering strategy prompt that encodes the stepwise processes of the problem setting and a safe-stock strategy commonly used in inventory management. Our empirical results demonstrate that, even without detailed prompt adjustments, an LLM-based MAS can determine optimal ordering decisions in a restricted scenario. To enhance adaptability, we propose a novel agent called AIM-RM, which leverages similar historical experiences through similarity matching. Our results show that AIM-RM outperforms benchmark methods across various supply chain scenarios, highlighting its robustness and adaptability.

Agentic large language models (LLMs) are increasingly evaluated on cybersecurity tasks using capture-the-flag (CTF) benchmarks. However, existing pointwise benchmarks have limited ability to shed light on the robustness and generalisation abilities of agents across alternative versions of the source code. We introduce CTF challenge families, whereby a single CTF is used as the basis for generating a family of semantically-equivalent challenges via semantics-preserving program transformations. This enables controlled evaluation of agent robustness to source code transformations while keeping the underlying exploit strategy fixed. We introduce a new tool, Evolve-CTF, that generates CTF families from Python challenges using a range of transformations. Using Evolve-CTF to derive families from Cybench and Intercode challenges, we evaluate 13 agentic LLM configurations with tool access. We find that models are remarkably robust to intrusive renaming and code insertion-based transformations, but that composed transformations and deeper obfuscation affect performance by requiring more sophisticated use of tools. We also find that enabling explicit reasoning has little effect on solution success rates across challenge families. Our work contributes a valuable technique and tool for future LLM evaluations, and a large dataset characterising the capabilities of current state-of-the-art models in this domain.

Artificial intelligence (AI)-enabled diagnostics in maxillofacial pathology require structured, high-quality multimodal datasets. However, existing resources provide limited ameloblastoma coverage and lack the format consistency needed for direct model training. We present a newly curated multimodal dataset specifically focused on ameloblastoma, integrating annotated radiological, histopathological, and intraoral clinical images with structured data derived from case reports. Natural language processing techniques were employed to extract clinically relevant features from textual reports, while image data underwent domain specific preprocessing and augmentation. Using this dataset, a multimodal deep learning model was developed to classify ameloblastoma variants, assess behavioral patterns such as recurrence risk, and support surgical planning. The model is designed to accept clinical inputs such as presenting complaint, age, and gender during deployment to enhance personalized inference. Quantitative evaluation demonstrated substantial improvements; variant classification accuracy increased from 46.2 percent to 65.9 percent, and abnormal tissue detection F1-score improved from 43.0 percent to 90.3 percent. Benchmarked against resources like MultiCaRe, this work advances patient-specific decision support by providing both a robust dataset and an adaptable multimodal AI framework.

Overview of the Proposed DECO Framework.} DECO is a DiT-based policy that decouples multimodal conditioning. Image and action tokens interact via joint self attention, while proprioceptive states and optional conditions are injected through adaptive layer normalization. Tactile signals are injected via cross attention, while a lightweight LoRA-based adapter is used to efficiently fine-tune the pretrained policy. DECO is also accompanied by DECO-50, a bimanual dexterous manipulation dataset with tactile sensing, consisting of 4 scenarios and 28 sub-tasks, covering more than 50 hours of data, approximately 5 million frames, and 8,000 successful trajectories.

Large Language Models (LLMs) excel at language understanding but remain limited in knowledge-intensive domains due to hallucinations, outdated information, and limited explainability. Text-based retrieval-augmented generation (RAG) helps ground model outputs in external sources but struggles with multi-hop reasoning. Knowledge Graphs (KGs), in contrast, support precise, explainable querying, yet require a knowledge of query languages. This work introduces an interactive framework in which LLMs generate and explain Cypher graph queries and users iteratively refine them through natural language. Applied to real-world KGs, the framework improves accessibility to complex datasets while preserving factual accuracy and semantic rigor and provides insight into how model performance varies across domains. Our core quantitative evaluation is a 90-query benchmark on a synthetic movie KG that measures query explanation quality and fault detection across multiple LLMs, complemented by two smaller real-life query-generation experiments on a Hyena KG and the MaRDI (Mathematical Research Data Initiative) KG.

Large language models (LLMs) underpin interactive multimedia applications such as captioning, retrieval, recommendation, and creative content generation, yet their autoregressive decoding incurs substantial latency. Speculative decoding reduces latency using a lightweight draft model, but deployment is often limited by the cost and complexity of acquiring, tuning, and maintaining an effective draft model. Recent approaches usually require auxiliary training or specialization, and even training-free methods incur costly search or optimization. We propose SDFP, a fully training-free and plug-and-play framework that builds the draft model via Fisher Information Trace (FIT)-based layer pruning of a given LLM. Using layer sensitivity as a proxy for output perturbation, SDFP removes low-impact layers to obtain a compact draft while preserving compatibility with the original model for standard speculative verification. SDFP needs no additional training, hyperparameter tuning, or separately maintained drafts, enabling rapid, deployment-friendly draft construction. Across benchmarks, SDFP delivers 1.32x-1.5x decoding speedup without altering the target model's output distribution, supporting low-latency multimedia applications.

Explainable Multimodal Emotion Recognition plays a crucial role in applications such as human-computer interaction and social media analytics. However, current approaches struggle with cue-level perception and reasoning due to two main challenges: 1) general-purpose modality encoders are pretrained to capture global structures and general semantics rather than fine-grained emotional cues, resulting in limited sensitivity to emotional signals; and 2) available datasets usually involve a trade-off between annotation quality and scale, which leads to insufficient supervision for emotional cues and ultimately limits cue-level reasoning. Moreover, existing evaluation metrics are inadequate for assessing cue-level reasoning performance. To address these challenges, we propose eXplainable Emotion GPT (XEmoGPT), a novel EMER framework capable of both perceiving and reasoning over emotional cues. It incorporates two specialized modules: the Video Emotional Cue Bridge (VECB) and the Audio Emotional Cue Bridge (AECB), which enhance the video and audio encoders through carefully designed tasks for fine-grained emotional cue perception. To further support cue-level reasoning, we construct a large-scale dataset, EmoCue, designed to teach XEmoGPT how to reason over multimodal emotional cues. In addition, we introduce EmoCue-360, an automated metric that extracts and matches emotional cues using semantic similarity, and release EmoCue-Eval, a benchmark of 400 expert-annotated samples covering diverse emotional scenarios. Experimental results show that XEmoGPT achieves strong performance in both emotional cue perception and reasoning.

Large language models (LLMs) achieve strong capabilities by scaling model capacity and training data, yet many real-world deployments rely on smaller models trained or adapted from low-resource data. This gap motivates the need for mechanisms to transfer knowledge from large, high-resource models to smaller, low-resource targets. While model merging provides an effective transfer mechanism, most existing approaches assume architecture-compatible models and therefore cannot directly transfer knowledge from large high-resource LLMs to heterogeneous low-resource targets. In this work, we propose a cross-architecture merging framework based on optimal transport (OT) that aligns activations to infer cross-neuron correspondences between heterogeneous models. The resulting transport plans are then used to guide direct weight-space fusion, enabling effective high-resource to low-resource transfer using only a small set of inputs. Extensive experiments across low-resource languages and specialized domains demonstrate consistent improvements over target models.

Reinforcement Learning with Verified Reward (RLVR) has emerged as a critical paradigm for advancing the reasoning capabilities of Large Language Models (LLMs). Most existing RLVR methods, such as GRPO and its variants, ensure stable updates by constraining policy divergence through clipping likelihood ratios. This paper introduces a unified clipping framework that characterizes existing methods via a general notion of policy divergence, encompassing both likelihood ratios and Kullback-Leibler (KL) divergences and extending to alternative measures. The framework provides a principled foundation for systematically analyzing how different policy divergence measures affect exploration and performance. We further identify the KL3 estimator, a variance-reduced Monte Carlo estimator of the KL divergence, as a key policy divergence constraint. We theoretically demonstrate that the KL3-based constraint is mathematically equivalent to an asymmetric ratio-based clipping that reallocates probability mass toward high-confidence actions, promoting stronger exploration while retaining the simplicity of GRPO-style methods. Empirical results on mathematical reasoning benchmarks demonstrate that incorporating the KL3 estimator into GRPO improves both training stability and final performance, highlighting the importance of principled policy divergence constraints in policy optimization.

Data annotation remains a significant bottleneck in the Humanities and Social Sciences, particularly for complex semantic tasks such as metaphor identification. While Large Language Models (LLMs) show promise, a significant gap remains between the theoretical capability of LLMs and their practical utility for researchers. This paper introduces LinguistAgent, an integrated, user-friendly platform that leverages a reflective multi-model architecture to automate linguistic annotation. The system implements a dual-agent workflow, comprising an Annotator and a Reviewer, to simulate a professional peer-review process. LinguistAgent supports comparative experiments across three paradigms: Prompt Engineering (Zero/Few-shot), Retrieval-Augmented Generation, and Fine-tuning. We demonstrate LinguistAgent's efficacy using the task of metaphor identification as an example, providing real-time token-level evaluation (Precision, Recall, and $F_1$ score) against human gold standards. The application and codes are released on https://github.com/Bingru-Li/LinguistAgent.

We analyze two classical algorithms for solving additively composite convex optimization problems where the objective is the sum of a smooth term and a nonsmooth regularizer: proximal stochastic gradient method for a single regularizer; and the randomized incremental proximal method, which uses the proximal operator of a randomly selected function when the regularizer is given as the sum of many nonsmooth functions. We focus on relaxing the bounded variance assumption that is common, yet stringent, for getting last iterate convergence rates. We prove the $\widetilde{O}(1/\sqrt{T})$ rate of convergence for the last iterate of both algorithms under componentwise convexity and smoothness, which is optimal up to log terms. Our results apply directly to graph-guided regularizers that arise in multi-task and federated learning, where the regularizer decomposes as a sum over edges of a collaboration graph.

Digital sovereignty has emerged as a central concern for modern software-intensive systems, driven by the dominance of non-sovereign cloud infrastructures, the rapid adoption of Generative AI, and increasingly stringent regulatory requirements. While existing initiatives address governance, compliance, and security in isolation, they provide limited guidance on how sovereignty can be operationalized at the architectural level. In this paper, we argue that sovereignty must be treated as a first-class architectural property rather than a purely regulatory objective. We introduce a Sovereign Reference Architecture that integrates self-sovereign identity, blockchain-based trust and auditability, sovereign data governance, and Generative AI deployed under explicit architectural control. The architecture explicitly captures the dual role of Generative AI as both a source of governance risk and an enabler of compliance, accountability, and continuous assurance when properly constrained. By framing sovereignty as an architectural quality attribute, our work bridges regulatory intent and concrete system design, offering a coherent foundation for building auditable, evolvable, and jurisdiction-aware AI-enabled systems. The proposed reference architecture provides a principled starting point for future research and practice at the intersection of software architecture, Generative AI, and digital sovereignty.

Drug discovery remains time-consuming, labor-intensive, and expensive, often requiring years and substantial investment per drug candidate. Predicting compound-protein interactions (CPIs) is a critical component in this process, enabling the identification of molecular interactions between drug candidates and target proteins. Recent deep learning methods have successfully modeled CPIs at the atomic level, achieving improved efficiency and accuracy over traditional energy-based approaches. However, these models do not always align with chemical realities, as molecular fragments (motifs or functional groups) typically serve as the primary units of biological recognition and binding. In this paper, we propose Phi-former, a pairwise hierarchical interaction representation learning method that addresses this gap by incorporating the biological role of motifs in CPIs. Phi-former represents compounds and proteins hierarchically and employs a pairwise pre-training framework to model interactions systematically across atom-atom, motif-motif, and atom-motif levels, reflecting how biological systems recognize molecular partners. We design intra-level and inter-level learning pipelines that make different interaction levels mutually beneficial. Experimental results demonstrate that Phi-former achieves superior performance on CPI-related tasks. A case study shows that our method accurately identifies specific atoms or motifs activated in CPIs, providing interpretable model explanations. These insights may guide rational drug design and support precision medicine applications.

Motivation-based recommendation systems uncover user behavior drivers. Motivation modeling, crucial for decision-making and content preference, explains recommendation generation. Existing methods often treat motivation as latent variables from interaction data, neglecting heterogeneous information like review text. In multimodal motivation fusion, two challenges arise: 1) achieving stable cross-modal alignment amid noise, and 2) identifying features reflecting the same underlying motivation across modalities. To address these, we propose LLM-driven Motivation-aware Multimodal Recommendation (LMMRec), a model-agnostic framework leveraging large language models for deep semantic priors and motivation understanding. LMMRec uses chain-of-thought prompting to extract fine-grained user and item motivations from text. A dual-encoder architecture models textual and interaction-based motivations for cross-modal alignment, while Motivation Coordination Strategy and Interaction-Text Correspondence Method mitigate noise and semantic drift through contrastive learning and momentum updates. Experiments on three datasets show LMMRec achieves up to a 4.98\% performance improvement.

The quest for expert-level reasoning in Large Language Models (LLMs) has been hampered by a persistent \textit{reward bottleneck}: traditional reinforcement learning (RL) relies on scalar rewards that are \textbf{costly} to scale, \textbf{brittle} across domains, and \textbf{blind} to the underlying logic of a solution. This reliance on external, impoverished signals prevents models from developing a deep, self-contained understanding of reasoning principles. We introduce \textbf{ALIVE} (\emph{Adversarial Learning with Instructive Verbal Evaluation}), a hands-free alignment framework that moves beyond scalar reward optimization toward intrinsic reasoning acquisition. Grounded in the principle of \emph{Cognitive Synergy}, ALIVE unifies problem posing, solving, and judging within a single policy model to internalize the logic of correctness. By coupling adversarial learning with instructive verbal feedback, ALIVE enables models to internalize evaluative criteria directly from raw corpora, effectively transforming external critiques into an endogenous reasoning faculty. Empirical evaluations across mathematical reasoning, code generation, and general logical inference benchmarks demonstrate that ALIVE consistently mitigates reward signal limitations. With identical data and compute, it achieves accuracy gains, markedly improved cross-domain generalization, and higher self-correction rates. These results indicate that the reasoning trinity fosters a self-sustaining trajectory of capability growth, positioning ALIVE as a scalable foundation for general-purpose reasoning alignment without human-in-the-loop supervision.

Contemporary knowledge-based systems increasingly rely on multilingual emotion identification to support intelligent decision-making, yet they face major challenges due to emotional ambiguity and incomplete supervision. Emotion recognition from text is inherently uncertain because multiple emotional states often co-occur and emotion annotations are frequently missing or heterogeneous. Most existing multi-label emotion classification methods assume fully observed labels and rely on deterministic learning objectives, which can lead to biased learning and unreliable predictions under partial supervision. This paper introduces Reasoning under Ambiguity, an uncertainty-aware framework for multilingual multi-label emotion classification that explicitly aligns learning with annotation uncertainty. The proposed approach uses a shared multilingual encoder with language-specific optimization and an entropy-based ambiguity weighting mechanism that down-weights highly ambiguous training instances rather than treating missing labels as negative evidence. A mask-aware objective with positive-unlabeled regularization is further incorporated to enable robust learning under partial supervision. Experiments on English, Spanish, and Arabic emotion classification benchmarks demonstrate consistent improvements over strong baselines across multiple evaluation metrics, along with improved training stability, robustness to annotation sparsity, and enhanced interpretability.

Visual Language Navigation (VLN) is one of the fundamental capabilities for embodied intelligence and a critical challenge that urgently needs to be addressed. However, existing methods are still unsatisfactory in terms of both success rate (SR) and generalization: Supervised Fine-Tuning (SFT) approaches typically achieve higher SR, while Training-Free (TF) approaches often generalize better, but it is difficult to obtain both simultaneously. To this end, we propose a Memory-Execute-Review framework. It consists of three parts: a hierarchical memory module for providing information support, an execute module for routine decision-making and actions, and a review module for handling abnormal situations and correcting behavior. We validated the effectiveness of this framework on the Object Goal Navigation task. Across 4 datasets, our average SR achieved absolute improvements of 7% and 5% compared to all baseline methods under TF and Zero-Shot (ZS) settings, respectively. On the most commonly used HM3D_v0.1 and the more challenging open vocabulary dataset HM3D_OVON, the SR improved by 8% and 6%, under ZS settings. Furthermore, on the MP3D and HM3D_OVON datasets, our method not only outperformed all TF methods but also surpassed all SFT methods, achieving comprehensive leadership in both SR (5% and 2%) and generalization.

Black-box optimization is increasingly used in engineering design problems where simulation-based evaluations are costly and gradients are unavailable. In this context, the optimization community has largely analyzed algorithm performance in context-free setups, while not enough attention has been devoted to how problem formulation and domain knowledge may affect the optimization outcomes. We address this gap through a case study in the topology optimization of laminated composite structures, formulated as a black-box optimization problem. Specifically, we consider the design of a cantilever beam under a volume constraint, intending to minimize compliance while optimizing both the structural topology and fiber orientations. To assess the impact of problem formulation, we explicitly separate topology and material design variables and compare two strategies: a concurrent approach that optimizes all variables simultaneously without leveraging physical insight, and a sequential approach that optimizes variables of the same nature in stages. Our results show that context-agnostic strategies consistently lead to suboptimal or non-physical designs. In contrast, the sequential strategy yields better-performing and more interpretable solutions. These findings underscore the value of incorporating, when available, domain knowledge into the optimization process and motivate the development of new black-box benchmarks that reward physically informed and context-aware optimization strategies.

Flaky tests yield inconsistent results when they are repeatedly executed on the same code revision. They interfere with automated quality assurance of code changes and hinder efficient software testing. Previous work evaluated approaches to train machine learning models to classify flaky tests based on identifiers in the test code. However, the resulting classifiers have been shown to lack generalizability, hindering their applicability in practical environments. Recently, pre-trained Large Language Models (LLMs) have shown the capability to generalize across various tasks. Thus, they represent a promising approach to address the generalizability problem of previous approaches. In this study, we evaluated three LLMs (two general-purpose models, one code-specific model) using three prompting techniques on two benchmark datasets from prior studies on flaky test classification. Furthermore, we manually investigated 50 samples from the given datasets to determine whether classifying flaky tests based only on test code is feasible for humans. Our findings indicate that LLMs struggle to classify flaky tests given only the test code. The results of our best prompt-model combination were only marginally better than random guessing. In our manual analysis, we found that the test code does not necessarily contain sufficient information for a flakiness classification. Our findings motivate future work to evaluate LLMs for flakiness classification with additional context, for example, using retrieval-augmented generation or agentic AI.

Conditional representation learning aims to extract criterion-specific features for customized tasks. Recent studies project universal features onto the conditional feature subspace spanned by an LLM-generated text basis to obtain conditional representations. However, such methods face two key limitations: sensitivity to subspace basis and vulnerability to inter-subspace interference. To address these challenges, we propose OD-CRL, a novel framework integrating Adaptive Orthogonal Basis Optimization (AOBO) and Null-Space Denoising Projection (NSDP). Specifically, AOBO constructs orthogonal semantic bases via singular value decomposition with a curvature-based truncation. NSDP suppresses non-target semantic interference by projecting embeddings onto the null space of irrelevant subspaces. Extensive experiments conducted across customized clustering, customized classification, and customized retrieval tasks demonstrate that OD-CRL achieves a new state-of-the-art performance with superior generalization.

Modern AI systems achieve remarkable capabilities at the cost of substantial energy consumption. To connect intelligence to physical efficiency, we propose two complementary bits-per-joule metrics under explicit accounting conventions: (1) Thermodynamic Epiplexity per Joule -- bits of structural information about a theoretical environment-instance variable newly encoded in an agent's internal state per unit measured energy within a stated boundary -- and (2) Empowerment per Joule -- the embodied sensorimotor channel capacity (control information) per expected energetic cost over a fixed horizon. These provide two axes of physical intelligence: recognition (model-building) vs.control (action influence). Drawing on stochastic thermodynamics, we show how a Landauer-scale closed-cycle benchmark for epiplexity acquisition follows as a corollary of a standard thermodynamic-learning inequality under explicit subsystem assumptions, and we clarify how Landauer-scaled costs act as closed-cycle benchmarks under explicit reset/reuse and boundary-closure assumptions; conversely, we give a simple decoupling construction showing that without such assumptions -- and without charging for externally prepared low-entropy resources (e.g.fresh memory) crossing the boundary -- information gain and in-boundary dissipation need not be tightly linked. For empirical settings where the latent structure variable is unavailable, we align the operational notion of epiplexity with compute-bounded MDL epiplexity and recommend reporting MDL-epiplexity / compression-gain surrogates as companions. Finally, we propose a unified efficiency framework that reports both metrics together with a minimal checklist of boundary/energy accounting, coarse-graining/noise, horizon/reset, and cost conventions to reduce ambiguity and support consistent bits-per-joule comparisons, and we sketch connections to energy-adjusted scaling analyses.

Hyperparameter sensitivity in Deep Reinforcement Learning (RL) is often accepted as unavoidable. However, it remains unclear whether it is intrinsic to the RL problem or exacerbated by specific training mechanisms. We investigate this question in offline goal-conditioned RL, where data distributions are fixed, and non-stationarity can be explicitly controlled via scheduled shifts in data quality. Additionally, we study varying data qualities under both stationary and non-stationary regimes, and cover two representative algorithms: HIQL (bootstrapped TD-learning) and QRL (quasimetric representation learning). Overall, we observe substantially greater robustness to changes in hyperparameter configurations than commonly reported for online RL, even under controlled non-stationarity. Once modest expert data is present ($\approx$ 20\%), QRL maintains broad, stable near-optimal regions, while HIQL exhibits sharp optima that drift significantly across training phases. To explain this divergence, we introduce an inter-goal gradient alignment diagnostic. We find that bootstrapped objectives exhibit stronger destructive gradient interference, which coincides directly with hyperparameter sensitivity. These results suggest that high sensitivity to changes in hyperparameter configurations during training is not inevitable in RL, but is amplified by the dynamics of bootstrapping, offering a pathway toward more robust algorithmic objective design.

SLO-as-code has made per-service} reliability declarative, but user experience is defined by journeys whose reliability is an emergent property of microservice topology, routing, redundancy, timeouts/fallbacks, shared failure domains, and tail amplification. As a result, journey objectives (e.g., "checkout p99 < 400 ms") are often maintained outside code and drift as the system evolves, forcing teams to either miss user expectations or over-provision and gate releases with ad-hoc heuristics. We propose Emergence-as-Code (EmaC), a vision for making journey reliability computable and governable via intent plus evidence. An EmaC spec declares journey intent (objective, control-flow operators, allowed actions) and binds it to atomic SLOs and telemetry. A runtime inference component consumes operational artifacts (e.g., tracing and traffic configuration) to synthesize a candidate journey model with provenance and confidence. From the last accepted model, the EmaC compiler/controller derives bounded journey SLOs and budgets under explicit correlation assumptions (optimistic independence vs. pessimistic shared fate), and emits control-plane artifacts (burn-rate alerts, rollout gates, action guards) that are reviewable in a Git workflow. An anonymized artifact repository provides a runnable example specification and generated outputs.

This paper addresses robotic system engineering for safety- and mission-critical applications by bridging the gap between high-level objectives and formal, executable specifications. The proposed method, Robotic System Task to Model Transformation Methodology (RSTM2) is an ontology-driven, hierarchical approach using stochastic timed Petri nets with resources, enabling Monte Carlo simulations at mission, system, and subsystem levels. A hypothetical case study demonstrates how the RSTM2 method supports architectural trades, resource allocation, and performance analysis under uncertainty. Ontological concepts further enable explainable AI-based assistants, facilitating fully autonomous specification synthesis. The methodology offers particular benefits to complex multi-robot systems, such as the NASA CADRE mission, representing decentralized, resource-aware, and adaptive autonomous systems of the future.

Continual learning (CL) empowers AI systems to progressively acquire knowledge from non-stationary data streams. However, catastrophic forgetting remains a critical challenge. In this work, we identify attention drift in Vision Transformers as a primary source of catastrophic forgetting, where the attention to previously learned visual concepts shifts significantly after learning new tasks. Inspired by neuroscientific insights into the selective attention in the human visual system, we propose a novel attention-retaining framework to mitigate forgetting in CL. Our method constrains attention drift by explicitly modifying gradients during backpropagation through a two-step process: 1) extracting attention maps of the previous task using a layer-wise rollout mechanism and generating instance-adaptive binary masks, and 2) when learning a new task, applying these masks to zero out gradients associated with previous attention regions, thereby preventing disruption of learned visual concepts. For compatibility with modern optimizers, the gradient masking process is further enhanced by scaling parameter updates proportionally to maintain their relative magnitudes. Experiments and visualizations demonstrate the effectiveness of our method in mitigating catastrophic forgetting and preserving visual concepts. It achieves state-of-the-art performance and exhibits robust generalizability across diverse CL scenarios.

Liver tumour ablation presents a significant clinical challenge: whilst tumours are clearly visible on pre-operative MRI, they are often effectively invisible on intra-operative CT due to minimal contrast between pathological and healthy tissue. This work investigates the feasibility of cross-modality weak supervision for scenarios where pathology is visible in one modality (MRI) but absent in another (CT). We present a hybrid registration-segmentation framework that combines MSCGUNet for inter-modal image registration with a UNet-based segmentation module, enabling registration-assisted pseudo-label generation for CT images. Our evaluation on the CHAOS dataset demonstrates that the pipeline can successfully register and segment healthy liver anatomy, achieving a Dice score of 0.72. However, when applied to clinical data containing tumours, performance degrades substantially (Dice score of 0.16), revealing the fundamental limitations of current registration methods when the target pathology lacks corresponding visual features in the target modality. We analyse the "domain gap" and "feature absence" problems, demonstrating that whilst spatial propagation of labels via registration is feasible for visible structures, segmenting truly invisible pathology remains an open challenge. Our findings highlight that registration-based label transfer cannot compensate for the absence of discriminative features in the target modality, providing important insights for future research in cross-modality medical image analysis. Code an weights are available at: https://github.com/BudhaTronix/Weakly-Supervised-Tumour-Detection

While diffusion models have achieved great success in the field of video generation, this progress is accompanied by a rapidly escalating computational burden. Among the existing acceleration methods, Feature Caching is popular due to its training-free property and considerable speedup performance, but it inevitably faces semantic and detail drop with further compression. Another widely adopted method, training-aware step-distillation, though successful in image generation, also faces drastic degradation in video generation with a few steps. Furthermore, the quality loss becomes more severe when simply applying training-free feature caching to the step-distilled models, due to the sparser sampling steps. This paper novelly introduces a distillation-compatible learnable feature caching mechanism for the first time. We employ a lightweight learnable neural predictor instead of traditional training-free heuristics for diffusion models, enabling a more accurate capture of the high-dimensional feature evolution process. Furthermore, we explore the challenges of highly compressed distillation on large-scale video models and propose a conservative Restricted MeanFlow approach to achieve more stable and lossless distillation. By undertaking these initiatives, we further push the acceleration boundaries to $11.8\times$ while preserving generation quality. Extensive experiments demonstrate the effectiveness of our method. The code is in the supplementary materials and will be publicly available.

Large language models have emerged as powerful zero-shot rerankers for retrieval-augmented generation, offering strong generalization without task-specific training. However, existing LLM reranking methods either rely on heuristics that fail to fully exploit the information revealed by each ranking decision or are inefficient when they do. We introduce a tournament graph framework that provides a principled foundation for $k$-wise reranking. Our key observation is that each $k$-document comparison reveals a complete tournament of $\binom{k}{2}$ pairwise preferences. These tournaments are aggregated into a global preference graph, whose transitive closure yields many additional orderings without further model invocations. We formalize when a candidate's rank is certifiably determined and design a query schedule that greedily maximizes information gain towards identifying the top-$m$ items. Our framework also gracefully handles non-transitive preferences - cycles induced by LLM judgments - by collapsing them into equivalence classes that yield principled tiered rankings. Empirically, across 14 benchmarks and 5 LLMs, our method achieves Pareto dominance over existing methods: matching or exceeding accuracy while requiring 25-40% fewer tokens than comparable approaches, and 7$\times$ fewer than pairwise methods at near-identical quality.

Large Language Model agents increasingly operate external systems through programmatic interfaces, yet practitioners lack empirical guidance on how to structure the context these agents consume. Using SQL generation as a proxy for programmatic agent operations, we present a systematic study of context engineering for structured data, comprising 9,649 experiments across 11 models, 4 formats (YAML, Markdown, JSON, Token-Oriented Object Notation [TOON]), and schemas ranging from 10 to 10,000 tables. Our findings challenge common assumptions. First, architecture choice is model-dependent: file-based context retrieval improves accuracy for frontier-tier models (Claude, GPT, Gemini; +2.7%, p=0.029) but shows mixed results for open source models (aggregate -7.7%, p<0.001), with deficits varying substantially by model. Second, format does not significantly affect aggregate accuracy (chi-squared=2.45, p=0.484), though individual models, particularly open source, exhibit format-specific sensitivities. Third, model capability is the dominant factor, with a 21 percentage point accuracy gap between frontier and open source tiers that dwarfs any format or architecture effect. Fourth, file-native agents scale to 10,000 tables through domain-partitioned schemas while maintaining high navigation accuracy. Fifth, file size does not predict runtime efficiency: compact formats can consume significantly more tokens at scale due to format-unfamiliar search patterns. These findings provide practitioners with evidence-based guidance for deploying LLM agents on structured systems, demonstrating that architectural decisions should be tailored to model capability rather than assuming universal best practices.

Safety alignment mechanisms in Large Language Models (LLMs) often operate as latent internal states, obscuring the model's inherent capabilities. Building on this observation, we model the safety mechanism as an unobserved confounder from a causal perspective. Then, we propose the \textbf{C}ausal \textbf{F}ront-Door \textbf{A}djustment \textbf{A}ttack ({\textbf{CFA}}$^2$) to jailbreak LLM, which is a framework that leverages Pearl's Front-Door Criterion to sever the confounding associations for robust jailbreaking. Specifically, we employ Sparse Autoencoders (SAEs) to physically strip defense-related features, isolating the core task intent. We further reduce computationally expensive marginalization to a deterministic intervention with low inference complexity. Experiments demonstrate that {CFA}$^2$ achieves state-of-the-art attack success rates while offering a mechanistic interpretation of the jailbreaking process.

Vision-Language-Action (VLA) models have been attracting the attention of researchers and practitioners thanks to their promise of generalization. Although single-task policies still offer competitive performance, VLAs are increasingly able to handle commands and environments unseen in their training set. While generalization in vision and language space is undoubtedly important for robust versatile behaviors, a key meta-skill VLAs need to possess is affordance generalization -- the ability to manipulate new objects with familiar physical features. In this work, we present BusyBox, a physical benchmark for systematic semi-automatic evaluation of VLAs' affordance generalization. BusyBox consists of 6 modules with switches, sliders, wires, buttons, a display, and a dial. The modules can be swapped and rotated to create a multitude of BusyBox variations with different visual appearances but the same set of affordances. We empirically demonstrate that generalization across BusyBox variants is highly challenging even for strong open-weights VLAs such as $π_{0.5}$ and GR00T-N1.6. To encourage the research community to evaluate their own VLAs on BusyBox and to propose new affordance generalization experiments, we have designed BusyBox to be easy to build in most robotics labs. We release the full set of CAD files for 3D-printing its parts as well as a bill of materials for (optionally) assembling its electronics. We also publish a dataset of language-annotated demonstrations that we collected using the common bimanual Mobile Aloha robot on the canonical BusyBox configuration. All of the released materials are available at https://microsoft.github.io/BusyBox.

Vision-language models (VLMs) can achieve high accuracy while still accepting culturally plausible but visually incorrect interpretations. Existing hallucination benchmarks rarely test this failure mode, particularly outside Western contexts and English. We introduce M2CQA, a culturally grounded multimodal benchmark built from images spanning 17 MENA countries, paired with contrastive true and counterfactual statements in English, Arabic, and its dialects. To isolate hallucination beyond raw accuracy, we propose the CounterFactual Hallucination Rate (CFHR), which measures counterfactual acceptance conditioned on correctly answering the true statement. Evaluating state-of-the-art VLMs under multiple prompting strategies, we find that CFHR rises sharply in Arabic, especially in dialects, even when true-statement accuracy remains high. Moreover, reasoning-first prompting consistently increases counterfactual hallucination, while answering before justifying improves robustness. We will make the experimental resources and dataset publicly available for the community.

Electricity price forecasting (EPF) is essential for energy markets stakeholders (e.g. grid operators, energy traders, policymakers) but remains challenging due to the inherent volatility and nonlinearity of price signals. Traditional statistical and deep learning (DL) models often struggle to capture complex temporal dependencies and integrate heterogeneous data effectively. While time series foundation models (TSFMs) have shown strong performance in general time series forecasting tasks, such as traffic forecasting and weather forecasting. However, their effectiveness in day-ahead EPF, particularly in volatile markets, remains underexplored. This paper presents a spike regularization strategy and evaluates a wide range of TSFMs, including Tiny Time Mixers (TTMs), MOIRAI, MOMENT, and TimesFM, against traditional statistical and DL models such as Autoregressive Integrated Moving Average (ARIMA), Long-short Term Memory (LSTM), and Convolutional Neural Network - LSTM (CNN-LSTM) using half-hourly wholesale market data with volatile trends in Singapore. Exogenous factors (e.g. weather and calendar variables) are also incorporated into models where applicable. Results demonstrate that TSFMs consistently outperform traditional approaches, achieving up to 37.4% improvement in MAPE across various evaluation settings. The findings offer practical guidance for improving forecast accuracy and decision-making in volatile electricity markets.

Graphical User Interface (GUI) agent is pivotal to advancing intelligent human-computer interaction paradigms. Constructing powerful GUI agents necessitates the large-scale annotation of high-quality user-behavior trajectory data (i.e., intent-trajectory pairs) for training. However, manual annotation methods and current GUI agent data mining approaches typically face three critical challenges: high construction cost, poor data quality, and low data richness. To address these issues, we propose M$^2$-Miner, the first low-cost and automated mobile GUI agent data-mining framework based on Monte Carlo Tree Search (MCTS). For better data mining efficiency and quality, we present a collaborative multi-agent framework, comprising InferAgent, OrchestraAgent, and JudgeAgent for guidance, acceleration, and evaluation. To further enhance the efficiency of mining and enrich intent diversity, we design an intent recycling strategy to extract extra valuable interaction trajectories. Additionally, a progressive model-in-the-loop training strategy is introduced to improve the success rate of data mining. Extensive experiments have demonstrated that the GUI agent fine-tuned using our mined data achieves state-of-the-art performance on several commonly used mobile GUI benchmarks. Our work will be released to facilitate the community research.

Knowledge graphs (KGs) have become a key ingredient supporting a variety of applications. Beyond the traditional triplet representation of facts where a relation connects two entities, modern KGs observe an increasing number of hyper-relational facts, where an arbitrary number of qualifiers associated with a triplet provide auxiliary information to further describe the rich semantics of the triplet, which can effectively boost the reasoning performance in link prediction tasks. However, existing link prediction techniques over such hyper-relational KGs (HKGs) mostly focus on a transductive setting, where KG embedding models are learned from the specific vocabulary of a given KG and subsequently can only make predictions within the same vocabulary, limiting their generalizability to previously unseen vocabularies. Against this background, we propose THOR, an inducTive link prediction technique for Hyper-relational knOwledge gRaphs. Specifically, we first introduce both relation and entity foundation graphs, modeling their fundamental inter- and intra-fact interactions in HKGs, which are agnostic to any specific relations and entities. Afterward, THOR is designed to learn from the two foundation graphs with two parallel graph encoders followed by a transformer decoder, which supports efficient masked training and fully-inductive inference. We conduct a thorough evaluation of THOR in hyper-relational link prediction tasks on 12 datasets with different settings. Results show that THOR outperforms a sizable collection of baselines, yielding 66.1%, 55.9%, and 20.4% improvement over the best-performing rule-based, semi-inductive, and fully-inductive techniques, respectively. A series of ablation studies also reveals our key design factors capturing the structural invariance transferable across HKGs for inductive tasks.

Accurate cell instance segmentation is foundational for digital pathology analysis. Existing methods based on contour detection and distance mapping still face significant challenges in processing complex and dense cellular regions. Graph coloring-based methods provide a new paradigm for this task, yet the effectiveness of this paradigm in real-world scenarios with dense overlaps and complex topologies has not been verified. Addressing this issue, we release a large-scale dataset GBC-FS 2025, which contains highly complex and dense sub-cellular nuclear arrangements. We conduct the first systematic analysis of the chromatic properties of cell adjacency graphs across four diverse datasets and reveal an important discovery: most real-world cell graphs are non-bipartite, with a high prevalence of odd-length cycles (predominantly triangles). This makes simple 2-coloring theory insufficient for handling complex tissues, while higher-chromaticity models would cause representational redundancy and optimization difficulties. Building on this observation of complex real-world contexts, we propose Disco (Densely-overlapping Cell Instance Segmentation via Adjacency-aware COllaborative Coloring), an adjacency-aware framework based on the "divide and conquer" principle. It uniquely combines a data-driven topological labeling strategy with a constrained deep learning system to resolve complex adjacency conflicts. First, "Explicit Marking" strategy transforms the topological challenge into a learnable classification task by recursively decomposing the cell graph and isolating a "conflict set." Second, "Implicit Disambiguation" mechanism resolves ambiguities in conflict regions by enforcing feature dissimilarity between different instances, enabling the model to learn separable feature representations.

Automatic evaluation in grammatical error correction (GEC) is crucial for selecting the best-performing systems. Currently, reference-based metrics are a popular choice, which basically measure the similarity between hypothesis and reference sentences. However, similarity measures based on embeddings, such as BERTScore, are often ineffective, since many words in the source sentences remain unchanged in both the hypothesis and the reference. This study focuses on edits specifically designed for GEC, i.e., ERRANT, and computes similarity measured over the edits from the source sentence. To this end, we propose edit vector, a representation for an edit, and introduce a new metric, UOT-ERRANT, which transports these edit vectors from hypothesis to reference using unbalanced optimal transport. Experiments with SEEDA meta-evaluation show that UOT-ERRANT improves evaluation performance, particularly in the +Fluency domain where many edits occur. Moreover, our method is highly interpretable because the transport plan can be interpreted as a soft edit alignment, making UOT-ERRANT a useful metric for both system ranking and analyzing GEC systems. Our code is available from https://github.com/gotutiyan/uot-errant.

While POD-based surrogates are widely explored for hydrodynamic applications, the use of Koopman Autoencoders for real-world coastal-ocean modelling remains relatively limited. This paper introduces a flexible Koopman autoencoder formulation that incorporates meteorological forcings and boundary conditions, and systematically compares its performance against POD-based surrogates. The Koopman autoencoder employs a learned linear temporal operator in latent space, enabling eigenvalue regularization to promote temporal stability. This strategy is evaluated alongside temporal unrolling techniques for achieving stable and accurate long-term predictions. The models are assessed on three test cases spanning distinct dynamical regimes, with prediction horizons up to one year at 30-minute temporal resolution. Across all cases, the Koopman autoencoder with temporal unrolling yields the best overall accuracy compared to the POD-based surrogates, achieving relative root-mean-squared-errors of 0.01-0.13 and $R^2$-values of 0.65-0.996. Prediction errors are largest for current velocities, and smallest for water surface elevations. Comparing to in-situ observations, the surrogate yields -0.65% to 12% change in water surface elevation prediction error when compared to prediction errors of the physics-based model. These error levels, corresponding to a few centimeters, are acceptable for many practical applications, while inference speed-ups of 300-1400x enables workflows such as ensemble forecasting and long climate simulations for coastal-ocean modelling.

Definitions are the foundation for any scientific work, but with a significant increase in publication numbers, gathering definitions relevant to any keyword has become challenging. We therefore introduce SciDef, an LLM-based pipeline for automated definition extraction. We test SciDef on DefExtra & DefSim, novel datasets of human-extracted definitions and definition-pairs' similarity, respectively. Evaluating 16 language models across prompting strategies, we demonstrate that multi-step and DSPy-optimized prompting improve extraction performance. To evaluate extraction, we test various metrics and show that an NLI-based method yields the most reliable results. We show that LLMs are largely able to extract definitions from scientific literature (86.4% of definitions from our test-set); yet future work should focus not just on finding definitions, but on identifying relevant ones, as models tend to over-generate them. Code & datasets are available at https://github.com/Media-Bias-Group/SciDef.

Federated Learning (FL) aims to train a collaborative model while preserving data privacy. However, the distributed nature of this approach still raises privacy and security issues, such as the exposure of sensitive data due to inference attacks and the influence of Byzantine behaviors on the trained model. In particular, achieving both secure aggregation and Byzantine resilience remains challenging, as existing solutions often address these aspects independently. In this work, we propose to address these challenges through a novel approach that combines homomorphic encryption for privacy-preserving aggregation with property-inference-inspired meta-classifiers for Byzantine filtering. First, following the property-inference attacks blueprint, we train a set of filtering meta-classifiers on labeled shadow updates, reproducing a diverse ensemble of Byzantine misbehaviors in FL, including backdoor, gradient-inversion, label-flipping and shuffling attacks. The outputs of these meta-classifiers are then used to cancel the Byzantine encrypted updates by reweighting. Second, we propose an automated method for selecting the optimal kernel and the dimensionality hyperparameters with respect to homomorphic inference, aggregation constraints and efficiency over the CKKS cryptosystem. Finally, we demonstrate through extensive experiments the effectiveness of our approach against Byzantine participants on the FEMNIST, CIFAR10, GTSRB, and acsincome benchmarks. More precisely, our SVM filtering achieves accuracies between $90$% and $94$% for identifying Byzantine updates at the cost of marginal losses in model utility and encrypted inference runtimes ranging from $6$ to $24$ seconds and from $9$ to $26$ seconds for an overall aggregation.

Hospital administration departments handle a wide range of operational tasks and, in large hospitals, process over 10,000 requests per day, driving growing interest in LLM-based automation. However, prior work has focused primarily on patient--physician interactions or isolated administrative subtasks, failing to capture the complexity of real administrative workflows. To address this gap, we propose H-AdminSim, a comprehensive end-to-end simulation framework that combines realistic data generation with multi-agent-based simulation of hospital administrative workflows. These tasks are quantitatively evaluated using detailed rubrics, enabling systematic comparison of LLMs. Through FHIR integration, H-AdminSim provides a unified and interoperable environment for testing administrative workflows across heterogeneous hospital settings, serving as a standardized testbed for assessing the feasibility and performance of LLM-driven administrative automation.

The lack of impaired speech data hinders advancements in the development of inclusive speech technologies, particularly in low-resource languages such as Akan. To address this gap, this study presents a curated corpus of speech samples from native Akan speakers with speech impairment. The dataset comprises of 50.01 hours of audio recordings cutting across four classes of impaired speech namely stammering, cerebral palsy, cleft palate, and stroke induced speech disorder. Recordings were done in controlled supervised environments were participants described pre-selected images in their own words. The resulting dataset is a collection of audio recordings, transcriptions, and associated metadata on speaker demographics, class of impairment, recording environment and device. The dataset is intended to support research in low-resource automatic disordered speech recognition systems and assistive speech technology.

Advanced opinion dynamics modeling is vital for deciphering social behavior, emphasizing its role in mitigating polarization and securing cyberspace. To synergize mechanistic interpretability with data-driven flexibility, recent studies have explored the integration of Physics-Informed Neural Networks (PINNs) for opinion modeling. Despite this promise, existing methods are tailored to incomplete priors, lacking a comprehensive physical system to integrate dynamics from local, global, and endogenous levels. Moreover, penalty-based constraints adopted in existing methods struggle to deeply encode physical priors, leading to optimization pathologies and discrepancy between latent representations and physical transparency. To this end, we offer a physical view to interpret opinion dynamics via Diffusion-Convection-Reaction (DCR) system inspired by interacting particle theory. Building upon the Neural ODEs, we define the neural opinion dynamics to coordinate neural networks with physical priors, and further present the OPINN, a physics-informed neural framework for opinion dynamics modeling. Evaluated on real-world and synthetic datasets, OPINN achieves state-of-the-art performance in opinion evolution forecasting, offering a promising paradigm for the nexus of cyber, physical, and social systems.

As high-quality public text approaches exhaustion, a phenomenon known as the Data Wall, pre-training is shifting from more tokens to better tokens. However, existing methods either rely on heuristic static filters that ignore training dynamics, or use dynamic yet optimizer-agnostic criteria based on raw gradients. We propose OPUS (Optimizer-induced Projected Utility Selection), a dynamic data selection framework that defines utility in the optimizer-induced update space. OPUS scores candidates by projecting their effective updates, shaped by modern optimizers, onto a target direction derived from a stable, in-distribution proxy. To ensure scalability, we employ Ghost technique with CountSketch for computational efficiency, and Boltzmann sampling for data diversity, incurring only 4.7\% additional compute overhead. OPUS achieves remarkable results across diverse corpora, quality tiers, optimizers, and model scales. In pre-training of GPT-2 Large/XL on FineWeb and FineWeb-Edu with 30B tokens, OPUS outperforms industrial-level baselines and even full 200B-token training. Moreover, when combined with industrial-level static filters, OPUS further improves pre-training efficiency, even with lower-quality data. Furthermore, in continued pre-training of Qwen3-8B-Base on SciencePedia, OPUS achieves superior performance using only 0.5B tokens compared to full training with 3B tokens, demonstrating significant data efficiency gains in specialized domains.

A simple strategy for improving LLM accuracy, especially in math and reasoning problems, is to sample multiple responses and submit the answer most consistently reached. In this paper we leverage Bayesian prior information to save on sampling costs, stopping once sufficient consistency is reached. Although the exact posterior is computationally intractable, we further introduce an efficient "L-aggregated" stopping policy that tracks only the L-1 most frequent answer counts. Theoretically, we prove that L=3 is all you need: this coarse approximation is sufficient to achieve asymptotic optimality, and strictly dominates prior-free baselines, while having a fast posterior computation. Empirically, this identifies the most consistent (i.e., mode) LLM answer using fewer samples, and can achieve similar answer accuracy while cutting the number of LLM calls (i.e., saving on LLM inference costs) by up to 50%.

As Large Language Models (LLMs) achieve remarkable empirical success through scaling model and data size, pretraining has become increasingly critical yet computationally prohibitive, hindering rapid development. Despite the availability of numerous pretrained LLMs developed at significant computational expense, a fundamental real-world question remains underexplored: \textit{Can we leverage existing small pretrained models to accelerate the training of larger models?} In this paper, we propose a Late-to-Early Training (LET) paradigm that enables LLMs to explicitly learn later knowledge in earlier steps and earlier layers. The core idea is to guide the early layers of an LLM during early training using representations from the late layers of a pretrained (i.e. late training phase) model. We identify two key mechanisms that drive LET's effectiveness: late-to-early-step learning and late-to-early-layer learning. These mechanisms significantly accelerate training convergence while robustly enhancing both language modeling capabilities and downstream task performance, enabling faster training with superior performance. Extensive experiments on 1.4B and 7B parameter models demonstrate LET's efficiency and effectiveness. Notably, when training a 1.4B LLM on the Pile dataset, our method achieves up to 1.6$\times$ speedup with nearly 5\% improvement in downstream task accuracy compared to standard training, even when using a pretrained model with 10$\times$ fewer parameters than the target model.

Evaluating the quality of children's utterances in adult-child dialogue remains challenging due to insufficient context-sensitive metrics. Common proxies such as Mean Length of Utterance (MLU), lexical diversity (vocd-D), and readability indices (Flesch-Kincaid Grade Level, Gunning Fog Index) are dominated by length and ignore conversational context, missing aspects of response quality such as reasoning depth, topic maintenance, and discourse planning. We introduce an LLM-as-a-judge framework that first classifies the Previous Adult Utterance Type and then scores the child's response along two axes: Expansion (contextual elaboration and inferential depth) and Independence (the child's contribution to advancing the discourse). These axes reflect fundamental dimensions in child language development, where Expansion captures elaboration, clause combining, and causal and contrastive connectives. Independence captures initiative, topic control, decreasing reliance on adult scaffolding through growing self-regulation, and audience design. We establish developmental validity by showing age-related patterns and demonstrate predictive value by improving age estimation over common baselines. We further confirm semantic sensitivity by detecting differences tied to discourse relations. Our metrics align with human judgments, enabling large-scale evaluation. This shifts child utterance assessment from simply measuring length to evaluating how meaningfully the child's speech contributes to and advances the conversation within its context.

Time-series foundation models have emerged as a new paradigm for forecasting, yet their ability to effectively leverage exogenous features -- critical for electricity demand forecasting -- remains unclear. This paper empirically evaluates foundation models capable of modeling cross-channel correlations against a baseline LSTM with reversible instance normalization across Singaporean and Australian electricity markets at hourly and daily granularities. We systematically assess MOIRAI, MOMENT, TinyTimeMixers, ChronosX, and Chronos-2 under three feature configurations: all features, selected features, and target-only. Our findings reveal highly variable effectiveness: while Chronos-2 achieves the best performance among foundation models (in zero-shot settings), the simple baseline frequently outperforms all foundation models in Singapore's stable climate, particularly for short-term horizons. Model architecture proves critical, with synergistic architectural implementations (TTM's channel-mixing, Chronos-2's grouped attention) consistently leveraging exogenous features, while other approaches show inconsistent benefits. Geographic context emerges as equally important, with foundation models demonstrating advantages primarily in variable climates. These results challenge assumptions about universal foundation model superiority and highlight the need for domain-specific models, specifically in the energy domain.

Multivariate time series (MTS) forecasting is crucial for decision-making in domains such as weather, energy, and finance. It remains challenging because real-world sequences intertwine slow trends, multi-rate seasonalities, and irregular residuals. Existing methods often rely on rigid, hand-crafted decompositions or generic end-to-end architectures that entangle components and underuse structure shared across variables. To address these limitations, we propose DecompSSM, an end-to-end decomposition framework using three parallel deep state space model branches to capture trend, seasonal, and residual components. The model features adaptive temporal scales via an input-dependent predictor, a refinement module for shared cross-variable context, and an auxiliary loss that enforces reconstruction and orthogonality. Across standard benchmarks (ECL, Weather, ETTm2, and PEMS04), DecompSSM outperformed strong baselines, indicating the effectiveness of combining component-wise deep state space models and global context refinement.

MRI provides superior soft tissue contrast without ionizing radiation; however, the absence of electron density information limits its direct use for dose calculation. As a result, current radiotherapy workflows rely on combined MRI and CT acquisitions, increasing registration uncertainty and procedural complexity. Synthetic CT generation enables MRI only planning but remains challenging due to nonlinear MRI-CT relationships and anatomical variability. We propose Parallel Swin Transformer-Enhanced Med2Transformer, a 3D architecture that integrates convolutional encoding with dual Swin Transformer branches to model both local anatomical detail and long-range contextual dependencies. Multi-scale shifted window attention with hierarchical feature aggregation improves anatomical fidelity. Experiments on public and clinical datasets demonstrate higher image similarity and improved geometric accuracy compared with baseline methods. Dosimetric evaluation shows clinically acceptable performance, with a mean target dose error of 1.69%. Code is available at: https://github.com/mobaidoctor/med2transformer.

As large language models (LLMs) evolve into autonomous agents, their real-world applicability has expanded significantly, accompanied by new security challenges. Most existing agent defense mechanisms adopt a mandatory checking paradigm, in which security validation is forcibly triggered at predefined stages of the agent lifecycle. In this work, we argue that effective agent security should be intrinsic and selective rather than architecturally decoupled and mandatory. We propose Spider-Sense framework, an event-driven defense framework based on Intrinsic Risk Sensing (IRS), which allows agents to maintain latent vigilance and trigger defenses only upon risk perception. Once triggered, the Spider-Sense invokes a hierarchical defence mechanism that trades off efficiency and precision: it resolves known patterns via lightweight similarity matching while escalating ambiguous cases to deep internal reasoning, thereby eliminating reliance on external models. To facilitate rigorous evaluation, we introduce S$^2$Bench, a lifecycle-aware benchmark featuring realistic tool execution and multi-stage attacks. Extensive experiments demonstrate that Spider-Sense achieves competitive or superior defense performance, attaining the lowest Attack Success Rate (ASR) and False Positive Rate (FPR), with only a marginal latency overhead of 8.3\%.

Text-to-SQL is a key natural language processing task that maps natural language questions to SQL queries, enabling intuitive interaction with web-based databases. Although current methods perform well on benchmarks like BIRD and Spider, they struggle with complex reasoning, domain knowledge, and hypothetical queries, and remain costly in enterprise deployment. To address these issues, we propose a framework named IESR(Information Enhanced Structured Reasoning) for lightweight large language models: (i) leverages LLMs for key information understanding and schema linking, and decoupling mathematical computation and SQL generation, (ii) integrates a multi-path reasoning mechanism based on Monte Carlo Tree Search (MCTS) with majority voting, and (iii) introduces a trajectory consistency verification module with a discriminator model to ensure accuracy and consistency. Experimental results demonstrate that IESR achieves state-of-the-art performance on the complex reasoning benchmark LogicCat (24.28 EX) and the Archer dataset (37.28 EX) using only compact lightweight models without fine-tuning. Furthermore, our analysis reveals that current coder models exhibit notable biases and deficiencies in physical knowledge, mathematical computation, and common-sense reasoning, highlighting important directions for future research. We released code at https://github.com/Ffunkytao/IESR-SLM.

Recent progress in Vision Language Models (VLMs) has raised the question of whether they can reliably perform nonverbal reasoning. To this end, we introduce VRIQ (Visual Reasoning IQ), a novel benchmark designed to assess and analyze the visual reasoning ability of VLMs. We evaluate models on two sets of tasks: abstract puzzle-style and natural-image reasoning tasks. We find that on abstract puzzles, performance remains near random with an average accuracy of around 28%, while natural tasks yield better but still weak results with 45% accuracy. We also find that tool-augmented reasoning demonstrates only modest improvements. To uncover the source of this weakness, we introduce diagnostic probes targeting perception and reasoning. Our analysis demonstrates that around 56% of failures arise from perception alone, 43% from both perception and reasoning, and only a mere 1% from reasoning alone. This motivates us to design fine-grained diagnostic probe questions targeting specific perception categories (e.g., shape, count, position, 3D/depth), revealing that certain categories cause more failures than others. Our benchmark and analysis establish that current VLMs, even with visual reasoning tools, remain unreliable abstract reasoners, mostly due to perception limitations, and offer a principled basis for improving visual reasoning in multimodal systems.

Domain specific large language models are increasingly used to support patient education, triage, and clinical decision making in ophthalmology, making rigorous evaluation essential to ensure safety and accuracy. This study evaluated four small medical LLMs Meerkat-7B, BioMistral-7B, OpenBioLLM-8B, and MedLLaMA3-v20 in answering ophthalmology related patient queries and assessed the feasibility of LLM based evaluation against clinician grading. In this cross sectional study, 180 ophthalmology patient queries were answered by each model, generating 2160 responses. Models were selected for parameter sizes under 10 billion to enable resource efficient deployment. Responses were evaluated by three ophthalmologists of differing seniority and by GPT-4-Turbo using the S.C.O.R.E. framework assessing safety, consensus and context, objectivity, reproducibility, and explainability, with ratings assigned on a five point Likert scale. Agreement between LLM and clinician grading was assessed using Spearman rank correlation, Kendall tau statistics, and kernel density estimate analyses. Meerkat-7B achieved the highest performance with mean scores of 3.44 from Senior Consultants, 4.08 from Consultants, and 4.18 from Residents. MedLLaMA3-v20 performed poorest, with 25.5 percent of responses containing hallucinations or clinically misleading content, including fabricated terminology. GPT-4-Turbo grading showed strong alignment with clinician assessments overall, with Spearman rho of 0.80 and Kendall tau of 0.67, though Senior Consultants graded more conservatively. Overall, medical LLMs demonstrated potential for safe ophthalmic question answering, but gaps remained in clinical depth and consensus, supporting the feasibility of LLM based evaluation for large scale benchmarking and the need for hybrid automated and clinician review frameworks to guide safe clinical deployment.

Effective reinforcement learning (RL) for complex stochastic systems requires leveraging historical data collected in previous iterations to accelerate policy optimization. Classical experience replay treats all past observations uniformly and fails to account for their varying contributions to learning. To overcome this limitation, we propose Variance Reduction Experience Replay (VRER), a principled framework that selectively reuses informative samples to reduce variance in policy gradient estimation. VRER is algorithm-agnostic and integrates seamlessly with existing policy optimization methods, forming the basis of our sample-efficient off-policy algorithm, Policy Gradient with VRER (PG-VRER). Motivated by the lack of rigorous theoretical analysis of experience replay, we develop a novel framework that explicitly captures dependencies introduced by Markovian dynamics and behavior-policy interactions. Using this framework, we establish finite-time convergence guarantees for PG-VRER and reveal a fundamental bias-variance trade-off: reusing older experience increases bias but simultaneously reduces gradient variance. Extensive empirical experiments demonstrate that VRER consistently accelerates policy learning and improves performance over state-of-the-art policy optimization algorithms.

Many approximate machine unlearning methods demonstrate strong logit-level forgetting -- such as near-zero accuracy on the forget set -- yet continue to preserve substantial information within their internal feature representations. We refer to this discrepancy as superficial forgetting. Recent studies indicate that most existing unlearning approaches primarily alter the final classifier, leaving intermediate representations largely unchanged and highly similar to those of the original model. To address this limitation, we introduce the Erase at the Core (EC), a framework designed to enforce forgetting throughout the entire network hierarchy. EC integrates multi-layer contrastive unlearning on the forget set with retain set preservation through deeply supervised learning. Concretely, EC attaches auxiliary modules to intermediate layers and applies both contrastive unlearning and cross-entropy losses at each supervision point, with layer-wise weighted losses. Experimental results show that EC not only achieves effective logit-level forgetting, but also substantially reduces representational similarity to the original model across intermediate layers. Furthermore, EC is model-agnostic and can be incorporated as a plug-in module into existing unlearning methods, improving representation-level forgetting while maintaining performance on the retain set.

In recent years, Large Language Models (LLMs) have become widely used in medical applications, such as clinical decision support, medical education, and medical question answering. Yet, these models are often English-centric, limiting their robustness and reliability for linguistically diverse communities. Recent work has highlighted discrepancies in performance in low-resource languages for various medical tasks, but the underlying causes remain poorly understood. In this study, we conduct a cross-lingual empirical analysis of LLM performance on Arabic and English medical question and answering. Our findings reveal a persistent language-driven performance gap that intensifies with increasing task complexity. Tokenization analysis exposes structural fragmentation in Arabic medical text, while reliability analysis suggests that model-reported confidence and explanations exhibit limited correlation with correctness. Together, these findings underscore the need for language-aware design and evaluation strategies in LLMs for medical tasks.

Oblique decision trees combine the transparency of trees with the power of multivariate decision boundaries, but learning high-quality oblique splits is NP-hard, and practical methods still rely on slow search or theory-free heuristics. We present the Hinge Regression Tree (HRT), which reframes each split as a non-linear least-squares problem over two linear predictors whose max/min envelope induces ReLU-like expressive power. The resulting alternating fitting procedure is exactly equivalent to a damped Newton (Gauss-Newton) method within fixed partitions. We analyze this node-level optimization and, for a backtracking line-search variant, prove that the local objective decreases monotonically and converges; in practice, both fixed and adaptive damping yield fast, stable convergence and can be combined with optional ridge regularization. We further prove that HRT's model class is a universal approximator with an explicit $O(δ^2)$ approximation rate, and show on synthetic and real-world benchmarks that it matches or outperforms single-tree baselines with more compact structures.

Iterative Direct Preference Optimization has emerged as the state-of-the-art paradigm for aligning Large Language Models on reasoning tasks. Standard implementations (DPO-R1) rely on Best-of-N sampling (e.g., $N \ge 8$) to mine golden trajectories from the distribution tail. In this paper, we challenge this scaling hypothesis and reveal a counter-intuitive phenomenon: in mathematical reasoning, aggressive exploration yields diminishing returns and even catastrophic policy collapse. We theoretically demonstrate that scaling $N$ amplifies verifier noise and induces detrimental distribution shifts. To resolve this, we introduce \textbf{PACE} (Proximal Alignment via Corrective Exploration), which replaces brute-force mining with a generation-based corrective strategy. Operating with a minimal budget ($2<N<3$), PACE synthesizes high-fidelity preference pairs from failed explorations. Empirical evaluations show that PACE outperforms DPO-R1 $(N=16)$ while using only about $1/5$ of the compute, demonstrating superior robustness against reward hacking and label noise.

Efficient deployment of large language models (LLMs) requires extreme quantization, forcing a critical trade-off between low-bit efficiency and performance. Residual binarization enables hardware-friendly, matmul-free inference by stacking binary ($\pm$1) layers, but is plagued by pathological feature co-adaptation. We identify a key failure mode, which we term inter-path adaptation: during quantization-aware training (QAT), parallel residual binary paths learn redundant features, degrading the error-compensation structure and limiting the expressive capacity of the model. While prior work relies on heuristic workarounds (e.g., path freezing) that constrain the solution space, we propose RaBiT, a novel quantization framework that resolves co-adaptation by algorithmically enforcing a residual hierarchy. Its core mechanism sequentially derives each binary path from a single shared full-precision weight, which ensures that every path corrects the error of the preceding one. This process is stabilized by a robust initialization that prioritizes functional preservation over mere weight approximation. RaBiT redefines the 2-bit accuracy-efficiency frontier: it achieves state-of-the-art performance, rivals even hardware-intensive Vector Quantization (VQ) methods, and delivers a $4.49\times$ inference speed-up over full-precision models on an RTX 4090.

Integrating external tools enables Large Language Models (LLMs) to interact with real-world environments and solve complex tasks. Given the growing scale of available tools, effective tool retrieval is essential to mitigate constraints of LLMs' context windows and ensure computational efficiency. Existing approaches typically treat tool retrieval as a traditional ad-hoc retrieval task, matching user queries against the entire raw tool documentation. In this paper, we identify three fundamental challenges that limit the effectiveness of this paradigm: (i) the incompleteness and structural inconsistency of tool documentation; (ii) the significant semantic and granular mismatch between user queries and technical tool documents; and, most importantly, (iii) the multi-aspect nature of tool utility, that involves distinct dimensions, such as functionality, input constraints, and output formats, varying in format and importance. To address these challenges, we introduce Multi-Field Tool Retrieval, a framework designed to align user intent with tool representations through fine-grained, multi-field modeling. Experimental results show that our framework achieves SOTA performance on five datasets and a mixed benchmark, exhibiting superior generalizability and robustness.

The neighborhood scope (i.e., number of hops) where graph neural networks (GNNs) aggregate information to characterize a node's statistical property is critical to GNNs' performance. Two-stage approaches, training and validating GNNs for every pre-specified neighborhood scope to search for the best setting, is a time-consuming task and tends to be biased due to the search space design. How to adaptively determine proper neighborhood scopes for the aggregation process for both homophilic and heterophilic graphs remains largely unexplored. We thus propose to model the GNNs' message-passing behavior on a graph as a stochastic process by treating the number of hops as a beta process. This Bayesian framework allows us to infer the most plausible neighborhood scope for message aggregation simultaneously with the optimization of GNN parameters. Our theoretical analysis shows that the scope inference improves the expressivity of a GNN. Experiments on benchmark homophilic and heterophilic datasets show that the proposed method is compatible with state-of-the-art GNN variants, achieving competitive or superior performance on the node classification task, and providing well-calibrated predictions.

The Dolev-Reischuk lower bound establishes that any deterministic Byzantine Agreement (BA) protocol for $n$ processors tolerating $f$ faults requires $Ω(f^2+n)$ messages. But what exactly does this quadratic cost pay for? Even the minimal requirement that every correct processor \emph{receive at least one message} already necessitates $Ω(f^2 + n)$ messages. This raises a fundamental question: is the Dolev-Reischuk bound about the difficulty of \emph{reaching univalency} -- the point at which the protocol's outcome is determined -- or merely about \emph{disseminating} the outcome to all processors afterward? We resolve this question by showing that reaching univalency does \emph{not} require quadratic communication. Specifically, we introduce $ε$-BA, a relaxation allowing an $ε$-fraction of correct processors to output incorrectly, and prove it can be solved deterministically with $O(n \log n)$ communication complexity when $f < n(1/3 - ε)$. Crucially, any $ε$-BA protocol can serve as the first phase of a full BA protocol: after $ε$-BA, a single all-to-all exchange and majority vote completes BA. Since the outcome is already determined after $ε$-BA, this demonstrates that the quadratic cost in Dolev-Reischuk stems entirely from dissemination, rather than from reaching univalency. We also define Extractable BA for authenticated settings, capturing when processors collectively hold enough signed messages to determine the agreed value, and show it can be solved with communication complexity $O(f \log f)$.

We introduce PATHWAYS, a benchmark of 250 multi-step decision tasks that test whether web-based agents can discover and correctly use hidden contextual information. Across both closed and open models, agents typically navigate to relevant pages but retrieve decisive hidden evidence in only a small fraction of cases. When tasks require overturning misleading surface-level signals, performance drops sharply to near chance accuracy. Agents frequently hallucinate investigative reasoning by claiming to rely on evidence they never accessed. Even when correct context is discovered, agents often fail to integrate it into their final decision. Providing more explicit instructions improves context discovery but often reduces overall accuracy, revealing a tradeoff between procedural compliance and effective judgement. Together, these results show that current web agent architectures lack reliable mechanisms for adaptive investigation, evidence integration, and judgement override.

Large Language Models have shown strong capabilities in complex problem solving, yet many agentic systems remain difficult to interpret and control due to opaque internal workflows. While some frameworks offer explicit architectures for collaboration, many deployed agentic systems operate as black boxes to users. We address this by introducing Agentic Workflow Reconstruction (AWR), a new task aiming to synthesize an explicit, interpretable stand-in workflow that approximates a black-box system using only input--output access. We propose AgentXRay, a search-based framework that formulates AWR as a combinatorial optimization problem over discrete agent roles and tool invocations in a chain-structured workflow space. Unlike model distillation, AgentXRay produces editable white-box workflows that match target outputs under an observable, output-based proxy metric, without accessing model parameters. To navigate the vast search space, AgentXRay employs Monte Carlo Tree Search enhanced by a scoring-based Red-Black Pruning mechanism, which dynamically integrates proxy quality with search depth. Experiments across diverse domains demonstrate that AgentXRay achieves higher proxy similarity and reduces token consumption compared to unpruned search, enabling deeper workflow exploration under fixed iteration budgets.

Modern neural networks have shown promise for solving partial differential equations over surfaces, often by discretizing the surface as a mesh and learning with a mesh-aware graph neural network. However, graph neural networks suffer from oversmoothing, where a node's features become increasingly similar to those of its neighbors. Unitary graph convolutions, which are mathematically constrained to preserve smoothness, have been proposed to address this issue. Despite this, in many physical systems, such as diffusion processes, smoothness naturally increases and unitarity may be overconstraining. In this paper, we systematically study the smoothing effects of different GNNs for dynamics modeling and prove that unitary convolutions hurt performance for such tasks. We propose relaxed unitary convolutions that balance smoothness preservation with the natural smoothing required for physical systems. We also generalize unitary and relaxed unitary convolutions from graphs to meshes. In experiments on PDEs such as the heat and wave equations over complex meshes and on weather forecasting, we find that our method outperforms several strong baselines, including mesh-aware transformers and equivariant neural networks.

Language models (LMs) have been reported to implicitly encode character-level information, despite not being explicitly provided during training. However, the mechanisms underlying this phenomenon remain largely unexplored. To reveal the mechanisms, we analyze how models acquire character-level knowledge by comparing LMs trained under controlled settings, such as specifying the pre-training dataset or tokenizer, with those trained under standard settings. We categorize the contributing factors into those independent of tokenization. Our analysis reveals that merge rules and orthographic constraints constitute primary factors arising from tokenization, whereas semantic associations of substrings and syntactic information function as key factors independent of tokenization.

Distributed ledgers are increasingly relied upon by industry to provide trustworthy accountability, strong integrity protection, and high availability for critical data without centralizing trust. Recently, distributed append-only logs are opting for a layered approach, combining crash-fault-tolerant (CFT) consensus with hardware-based Trusted Execution Environments (TEEs) for greater resiliency. Unfortunately, hardware TEEs can be subject to (rare) attacks, undermining the very guarantees that distributed ledgers are carefully designed to achieve. In response, we present Proteus, a new distributed consensus protocol that cautiously trusts the guarantees of TEEs. Proteus carefully embeds a Byzantine fault-tolerant (BFT) protocol inside of a CFT protocol with no additional messages. This is made possible through careful refactoring of both the CFT and BFT protocols such that their structure aligns. Proteus achieves performance in line with regular TEE-enabled consensus protocols, while guaranteeing integrity in the face of TEE platform compromises.

We consider the sequential experimental design problem in the predict-then-optimize paradigm. In this paradigm, the outputs of the prediction model are used as coefficient vectors in a downstream linear optimization problem. Traditional sequential experimental design aims to control the input variables (features) so that the improvement in prediction accuracy from each experimental outcome (label) is maximized. However, in the predict-then-optimize setting, performance is ultimately evaluated based on the decision loss induced by the downstream optimization, rather than by prediction error. This mismatch between prediction accuracy and decision loss renders traditional decision-blind designs inefficient. To address this issue, we propose a directional-based metric to quantify predictive uncertainty. This metric does not require solving an optimization oracle and is therefore computationally tractable. We show that the resulting sequential design criterion enjoys strong consistency and convergence guarantees. Under a broad class of distributions, we demonstrate that our directional uncertainty-based design attains an earlier stopping time than decision-blind designs. This advantage is further supported by real-world experiments on an LLM job allocation problem.

With the increasing versatility of text-to-image diffusion models, the ability to selectively erase undesirable concepts (e.g., harmful content) has become indispensable. However, existing concept erasure approaches primarily focus on removing unsafe concepts without providing guidance toward corresponding safe alternatives, which often leads to failure in preserving the structural and semantic consistency between the original and erased generations. In this paper, we propose a novel framework, PAIRed Erasing (PAIR), which reframes concept erasure from simple removal to consistency-preserving semantic realignment using unsafe-safe pairs. We first generate safe counterparts from unsafe inputs while preserving structural and semantic fidelity, forming paired unsafe-safe multimodal data. Leveraging these pairs, we introduce two key components: (1) Paired Semantic Realignment, a guided objective that uses unsafe-safe pairs to explicitly map target concepts to semantically aligned safe anchors; and (2) Fisher-weighted Initialization for DoRA, which initializes parameter-efficient low-rank adaptation matrices using unsafe-safe pairs, encouraging the generation of safe alternatives while selectively suppressing unsafe concepts. Together, these components enable fine-grained erasure that removes only the targeted concepts while maintaining overall semantic consistency. Extensive experiments demonstrate that our approach significantly outperforms state-of-the-art baselines, achieving effective concept erasure while preserving structural integrity, semantic coherence, and generation quality.

This paper proposes an information-theoretic framework for analyzing the theoretical limits of pool-based active learning (AL), in which a subset of instances is selectively labeled. The proposed framework reformulates pool-based AL as a noisy lossy compression problem by mapping pool observations to noisy symbol observations, data selection to compression, and learning to decoding. This correspondence enables a unified information-theoretic analysis of data selection and learning in pool-based AL. Applying finite blocklength analysis of noisy lossy compression, we derive information-theoretic lower bounds on label complexity and generalization error that serve as theoretical limits for a given learning algorithm under its associated optimal data selection strategy. Specifically, our bounds include terms that reflect overfitting induced by the learning algorithm and the discrepancy between its inductive bias and the target task, and are closely related to established information-theoretic bounds and stability theory, which have not been previously applied to the analysis of pool-based AL. These properties yield a new theoretical perspective on pool-based AL.

Existing Large Language Model (LLM) agents struggle in interactive environments requiring long-horizon planning, primarily due to compounding errors when simulating future states. To address this, we propose ProAct, a framework that enables agents to internalize accurate lookahead reasoning through a two-stage training paradigm. First, we introduce Grounded LookAhead Distillation (GLAD), where the agent undergoes supervised fine-tuning on trajectories derived from environment-based search. By compressing complex search trees into concise, causal reasoning chains, the agent learns the logic of foresight without the computational overhead of inference-time search. Second, to further refine decision accuracy, we propose the Monte-Carlo Critic (MC-Critic), a plug-and-play auxiliary value estimator designed to enhance policy-gradient algorithms like PPO and GRPO. By leveraging lightweight environment rollouts to calibrate value estimates, MC-Critic provides a low-variance signal that facilitates stable policy optimization without relying on expensive model-based value approximation. Experiments on both stochastic (e.g., 2048) and deterministic (e.g., Sokoban) environments demonstrate that ProAct significantly improves planning accuracy. Notably, a 4B parameter model trained with ProAct outperforms all open-source baselines and rivals state-of-the-art closed-source models, while demonstrating robust generalization to unseen environments. The codes and models are available at https://github.com/GreatX3/ProAct

Dynamic games are powerful tools to model multi-agent decision-making, yet computing Nash (generalized Nash) equilibria remains a central challenge in such settings. Complexity arises from tightly coupled optimality conditions, nested optimization structures, and poor numerical conditioning. Existing game-theoretic solvers address these challenges by directly solving the joint game, typically requiring explicit modeling of all agents' objective functions and constraints, while learning-based approaches often decouple interaction through prediction or policy approximation, sacrificing equilibrium consistency. This paper introduces a conceptually novel formulation for dynamic games by restructuring the equilibrium computation. Rather than solving a fully coupled game or decoupling agents through prediction or policy approximation, a data-driven structural reduction of the game is proposed that removes nested optimization layers and derivative coupling by embedding an offline-compiled best-response map as a feasibility constraint. Under standard regularity conditions, when the best-response operator is exact, any converged solution of the reduced problem corresponds to a local open-loop Nash (GNE) equilibrium of the original game; with a learned surrogate, the solution is approximately equilibrium-consistent up to the best-response approximation error. The proposed formulation is supported by mathematical proofs, accompanying a large-scale Monte Carlo study in a two-player open-loop dynamic game motivated by the autonomous racing problem. Comparisons are made against state-of-the-art joint game solvers, and results are reported on solution quality, computational cost, and constraint satisfaction.

Offline Safe Reinforcement Learning (OSRL) aims to learn a policy to achieve high performance in sequential decision-making while satisfying constraints, using only pre-collected datasets. Recent works, inspired by the strong capabilities of Generative Models (GMs), reformulate decision-making in OSRL as a conditional generative process, where GMs generate desirable actions conditioned on predefined reward and cost values. However, GM-assisted methods face two major challenges in OSRL: (1) lacking the ability to "stitch" optimal transitions from suboptimal trajectories within the dataset, and (2) struggling to balance reward targets with cost targets, particularly when they are conflict. To address these issues, we propose Goal-Assisted Stitching (GAS), a novel algorithm designed to enhance stitching capabilities while effectively balancing reward maximization and constraint satisfaction. To enhance the stitching ability, GAS first augments and relabels the dataset at the transition level, enabling the construction of high-quality trajectories from suboptimal ones. GAS also introduces novel goal functions, which estimate the optimal achievable reward and cost goals from the dataset. These goal functions, trained using expectile regression on the relabeled and augmented dataset, allow GAS to accommodate a broader range of reward-cost return pairs and achieve a better tradeoff between reward maximization and constraint satisfaction compared to human-specified values. The estimated goals then guide policy training, ensuring robust performance under constrained settings. Furthermore, to improve training stability and efficiency, we reshape the dataset to achieve a more uniform reward-cost return distribution. Empirical results validate the effectiveness of GAS, demonstrating superior performance in balancing reward maximization and constraint satisfaction compared to existing methods.

Sequential prediction from streaming observations is a fundamental problem in stochastic dynamical systems, where inherent uncertainty often leads to multiple plausible futures. While diffusion and flow-matching models are capable of modeling complex, multi-modal trajectories, their deployment in real-time streaming environments typically relies on repeated sampling from a non-informative initial distribution, incurring substantial inference latency and potential system backlogs. In this work, we introduce Sequential Flow Matching, a principled framework grounded in Bayesian filtering. By treating streaming inference as learning a probability flow that transports the predictive distribution from one time step to the next, our approach naturally aligns with the recursive structure of Bayesian belief updates. We provide theoretical justification that initializing generation from the previous posterior offers a principled warm start that can accelerate sampling compared to naïve re-sampling. Across a wide range of forecasting, decision-making and state estimation tasks, our method achieves performance competitive with full-step diffusion while requiring only one or very few sampling steps, therefore with faster sampling. It suggests that framing sequential inference via Bayesian filtering provides a new and principled perspective towards efficient real-time deployment of flow-based models.

Fuzzing has become a popular technique for automatically detecting vulnerabilities and bugs by generating unexpected inputs. In recent years, the fuzzing process has been integrated into continuous integration workflows (i.e., continuous fuzzing), enabling short and frequent testing cycles. Despite its widespread adoption, prior research has not examined whether the effectiveness of continuous fuzzing varies across programming languages. This study conducts a large-scale cross-language analysis to examine how fuzzing bug characteristics and detection efficiency differ among languages. We analyze 61,444 fuzzing bugs and 999,248 builds from 559 OSS-Fuzz projects categorized by primary language. Our findings reveal that (i) C++ and Rust exhibit higher fuzzing bug detection frequencies, (ii) Rust and Python show low vulnerability ratios but tend to expose more critical vulnerabilities, (iii) crash types vary across languages and unreproducible bugs are more frequent in Go but rare in Rust, and (iv) Python attains higher patch coverage but suffers from longer time-to-detection. These results demonstrate that fuzzing behavior and effectiveness are strongly shaped by language design, providing insights for language-aware fuzzing strategies and tool development.

Neural Lyapunov and barrier certificates have recently been used as powerful tools for verifying the safety and stability properties of deep reinforcement learning (RL) controllers. However, existing methods offer guarantees only under fixed ideal unperturbed dynamics, limiting their reliability in real-world applications where dynamics may deviate due to uncertainties. In this work, we study the problem of synthesizing \emph{robust neural Lyapunov barrier certificates} that maintain their guarantees under perturbations in system dynamics. We formally define a robust Lyapunov barrier function and specify sufficient conditions based on Lipschitz continuity that ensure robustness against bounded perturbations. We propose practical training objectives that enforce these conditions via adversarial training, Lipschitz neighborhood bound, and global Lipschitz regularization. We validate our approach in two practically relevant environments, Inverted Pendulum and 2D Docking. The former is a widely studied benchmark, while the latter is a safety-critical task in autonomous systems. We show that our methods significantly improve both certified robustness bounds (up to $4.6$ times) and empirical success rates under strong perturbations (up to $2.4$ times) compared to the baseline. Our results demonstrate effectiveness of training robust neural certificates for safe RL under perturbations in dynamics.

Large reasoning models (LRMs) achieve strong performance by producing long chains of thought, but their inference costs are high and often generate redundant reasoning. Small language models (SLMs) are far more efficient, yet struggle on multi-step reasoning tasks. A natural idea is to let a large model guide a small one at inference time as a mentor, yet existing collaboration methods often promote imitation, resulting in verbose reasoning without consistent error correction. We propose MentorCollab, an inference-time collaboration method in which an LRM selectively and sparsely guides an SLM, rather than taking over generation. At randomly sampled token positions, we probe for divergences between the two models and use a lightweight verifier to decide whether the SLM should follow a short lookahead segment from its mentor or continue on its own. Across 15 SLM--LRM pairs and 3 domains (math reasoning, general knowledge, and commonsense reasoning), our method improves performance in 12 settings, with average gains of 3.0% and up to 8.0%, while adopting only having 18.4% tokens generated by the expensive mentor model on average. We find that short segments and selective probing are sufficient for effective collaboration. Our results show that selective inference-time guidance restores large-model reasoning ability without substantial inference overhead.

Generating long-form content, such as minute-long videos and extended texts, is increasingly important for modern generative models. Block diffusion improves inference efficiency via KV caching and block-wise causal inference and has been widely adopted in diffusion language models and video generation. However, in long-context settings, block diffusion still incurs substantial overhead from repeatedly computing attention over a growing KV cache. We identify an underexplored property of block diffusion: cross-step redundancy of attention within a block. Our analysis shows that attention outputs from tokens outside the current block remain largely stable across diffusion steps, while block-internal attention varies significantly. Based on this observation, we propose FlashBlock, a cached block-external attention mechanism that reuses stable attention output, reducing attention computation and KV cache access without modifying the diffusion process. Moreover, FlashBlock is orthogonal to sparse attention and can be combined as a complementary residual reuse strategy, substantially improving model accuracy under aggressive sparsification. Experiments on diffusion language models and video generation demonstrate up to 1.44$\times$ higher token throughput and up to 1.6$\times$ reduction in attention time, with negligible impact on generation quality. Project page: https://caesarhhh.github.io/FlashBlock/.

Stochastic variance-reduced algorithms such as Stochastic Average Gradient (SAG) and SAGA, and their deterministic counterparts like the Incremental Aggregated Gradient (IAG) method, have been extensively studied in large-scale machine learning. Despite their popularity, existing analyses for these algorithms are disparate, relying on different proof techniques tailored to each method. Furthermore, the original proof of SAG is known to be notoriously involved, requiring computer-aided analysis. Focusing on finite-sum optimization with smooth and strongly convex objective functions, our main contribution is to develop a single unified convergence analysis that applies to all three algorithms: SAG, SAGA, and IAG. Our analysis features two key steps: (i) establishing a bound on delays due to stochastic sub-sampling using simple concentration tools, and (ii) carefully designing a novel Lyapunov function that accounts for such delays. The resulting proof is short and modular, providing the first high-probability bounds for SAG and SAGA that can be seamlessly extended to non-convex objectives and Markov sampling. As an immediate byproduct of our new analysis technique, we obtain the best known rates for the IAG algorithm, significantly improving upon prior bounds.

We present an in-depth evaluation of LLMs' ability to negotiate, a central business task that requires strategic reasoning, theory of mind, and economic value creation. To do so, we introduce PieArena, a large-scale negotiation benchmark grounded in multi-agent interactions over realistic scenarios drawn from an MBA negotiation course at an elite business school. We find systematic evidence of AGI-level performance in which a representative frontier agent (GPT-5) matches or outperforms trained business-school students, despite a semester of general negotiation instruction and targeted coaching immediately prior to the task. We further study the effects of joint-intentionality agentic scaffolding and find asymmetric gains, with large improvements for mid- and lower-tier LMs and diminishing returns for frontier LMs. Beyond deal outcomes, PieArena provides a multi-dimensional negotiation behavioral profile, revealing novel cross-model heterogeneity, masked by deal-outcome-only benchmarks, in deception, computation accuracy, instruction compliance, and perceived reputation. Overall, our results suggest that frontier language agents are already intellectually and psychologically capable of deployment in high-stakes economic settings, but deficiencies in robustness and trustworthiness remain open challenges.

In practice, the hyperparameters $(β_1, β_2)$ and weight-decay $λ$ in AdamW are typically kept at fixed values. Is there any reason to do otherwise? We show that for large-scale language model training, the answer is yes: by exploiting the power-law structure of language data, one can design time-varying schedules for $(β_1, β_2, λ)$ that deliver substantial performance gains. We study logarithmic-time scheduling, in which the optimizer's gradient memory horizon grows with training time. Although naive variants of this are unstable, we show that suitable damping mechanisms restore stability while preserving the benefits of longer memory. Based on this, we present ADANA, an AdamW-like optimizer that couples log-time schedules with explicit damping to balance stability and performance. We empirically evaluate ADANA across transformer scalings (45M to 2.6B parameters), comparing against AdamW, Muon, and AdEMAMix. When properly tuned, ADANA achieves up to 40% compute efficiency relative to a tuned AdamW, with gains that persist--and even improve--as model scale increases. We further show that similar benefits arise when applying logarithmic-time scheduling to AdEMAMix, and that logarithmic-time weight-decay alone can yield significant improvements. Finally, we present variants of ADANA that mitigate potential failure modes and improve robustness.

MOOC recommendation systems have received increasing attention to help learners navigate and select preferred learning content. Traditional methods such as collaborative filtering and content-based filtering suffer from data sparsity and over-specialization. To alleviate these limitations, graph-based approaches have been proposed; however, they still rely heavily on manually predefined metapaths, which often capture only superficial structural relationships and impose substantial burdens on domain experts as well as significant engineering costs. To overcome these limitations, we propose AMR (Aspect-aware MOOC Recommendation), a novel framework that models path-specific multiple aspects by embedding the semantic content of nodes within each metapath. AMR automatically discovers metapaths through bi-directional walks, derives aspect-aware path representations using a bi-LSTM-based encoder, and incorporates these representations as edge features in the learner-learner and KC-KC subgraphs to achieve fine-grained semantically informed KC recommendations. Extensive experiments on the large-scale MOOCCube and PEEK datasets show that AMR consistently outperforms state-of-the-art graph neural network baselines across key metrics such as HR@K and nDCG@K. Further analysis confirms that AMR effectively captures rich path-specific aspect information, allowing more accurate recommendations than those methods that rely solely on predefined metapaths. The code will be available upon accepted.

Applications are moving away from monolithic designs to microservice and serverless architectures, where fleets of lightweight and independently deployable components run on public clouds. Autoscaling serves as the primary control mechanism for balancing resource utilization and quality of service, yet existing policies are either opaque learned models that require substantial per-deployment training or brittle hand-tuned rules that fail to generalize. We investigate whether large language models can act as universal few-shot resource allocators that adapt across rapidly evolving microservice deployments. We propose ORACL, Optimized Reasoning for Autoscaling via Chain of Thought with LLMs for Microservices, a framework that leverages prior knowledge and chain-of-thought reasoning to diagnose performance regressions and recommend resource allocations. ORACL transforms runtime telemetry, including pods, replicas, CPU and memory usage, latency, service-level objectives, and fault signals, into semantic natural-language state descriptions and invokes an LLM to produce an interpretable intermediate reasoning trace. This reasoning identifies likely root causes, prunes the action space, and issues safe allocation decisions under policy constraints. Experiments on representative open-source microservice workloads show that ORACL improves root-cause identification accuracy by 15 percent, accelerates training by up to 24x, and improves quality of service by 6 percent in short-term scenarios, without deployment-specific retraining.

Recent advancements in Large Language Models (LLMs) have greatly extended the capabilities of Multi-Agent Systems (MAS), demonstrating significant effectiveness across a wide range of complex and open-ended domains. However, despite this rapid progress, the field still relies heavily on empirical trial-and-error. It lacks a unified and principled scientific framework necessary for systematic optimization and improvement. This bottleneck stems from the ambiguity of attribution: first, the absence of a structured taxonomy of factors leaves researchers restricted to unguided adjustments; second, the lack of a unified metric fails to distinguish genuine collaboration gain from mere resource accumulation. In this paper, we advocate for a transition to design science through an integrated framework. We advocate to establish the collaboration gain metric ($Γ$) as the scientific standard to isolate intrinsic gains from increased budgets. Leveraging $Γ$, we propose a factor attribution paradigm to systematically identify collaboration-driving factors. To support this, we construct a systematic MAS factor library, structuring the design space into control-level presets and information-level dynamics. Ultimately, this framework facilitates the transition from blind experimentation to rigorous science, paving the way towards a true science of Collective AI.

This position paper argues that the pursuit of "Universal Foundation Models for Time Series" rests on a fundamental category error, mistaking a structural Container for a semantic Modality. We contend that because time series hold incompatible generative processes (e.g., finance vs. fluid dynamics), monolithic models degenerate into expensive "Generic Filters" that fail to generalize under distributional drift. To address this, we introduce the "Autoregressive Blindness Bound," a theoretical limit proving that history-only models cannot predict intervention-driven regime shifts. We advocate replacing universality with a Causal Control Agent paradigm, where an agent leverages external context to orchestrate a hierarchy of specialized solvers, from frozen domain experts to lightweight Just-in-Time adaptors. We conclude by calling for a shift in benchmarks from "Zero-Shot Accuracy" to "Drift Adaptation Speed" to prioritize robust, control-theoretic systems.

Healthcare facility visit prediction is essential for optimizing healthcare resource allocation and informing public health policy. Despite advanced machine learning methods being employed for better prediction performance, existing works usually formulate this task as a time-series forecasting problem without considering the intrinsic spatial dependencies of different types of healthcare facilities, and they also fail to provide reliable predictions under abnormal situations such as public emergencies. To advance existing research, we propose HealthMamba, an uncertainty-aware spatiotemporal framework for accurate and reliable healthcare facility visit prediction. HealthMamba comprises three key components: (i) a Unified Spatiotemporal Context Encoder that fuses heterogeneous static and dynamic information, (ii) a novel Graph State Space Model called GraphMamba for hierarchical spatiotemporal modeling, and (iii) a comprehensive uncertainty quantification module integrating three uncertainty quantification mechanisms for reliable prediction. We evaluate HealthMamba on four large-scale real-world datasets from California, New York, Texas, and Florida. Results show HealthMamba achieves around 6.0% improvement in prediction accuracy and 3.5% improvement in uncertainty quantification over state-of-the-art baselines.

In many biophysical inverse problems, the goal is to generate biomolecular conformations that are both physically plausible and consistent with experimental measurements. As recent sequence-to-structure diffusion models provide powerful data-driven priors, posterior sampling has emerged as a popular framework by guiding atomic coordinates to target conformations using experimental likelihoods. However, when the target lies in a low-density region of the prior, posterior sampling requires aggressive and brittle weighting of the likelihood guidance. Motivated by this limitation, we propose EmbedOpt, an alternative inference-time approach for steering diffusion models to optimize experimental likelihoods in the conditional embedding space. As this space encodes rich sequence and coevolutionary signals, optimizing over it effectively shifts the diffusion prior to align with experimental constraints. We validate EmbedOpt on two benchmarks simulating cryo-electron microscopy map fitting and experimental distance constraints. We show that EmbedOpt outperforms the coordinate-based posterior sampling method in map fitting tasks, matches performance on distance constraint tasks, and exhibits superior engineering robustness across hyperparameters spanning two orders of magnitude. Moreover, its smooth optimization behavior enables a significant reduction in the number of diffusion steps required for inference, leading to better efficiency.

Reinforcement Learning with Verifiable Rewards (RLVR) has emerged as an indispensable paradigm for enhancing reasoning in Large Language Models (LLMs). However, standard policy optimization methods, such as Group Relative Policy Optimization (GRPO), often converge to low-entropy policies, leading to severe mode collapse and limited output diversity. We analyze this issue from the perspective of sampling probability dynamics, identifying that the standard objective disproportionately reinforces the highest-likelihood paths, thereby suppressing valid alternative reasoning chains. To address this, we propose a novel Advantage Re-weighting Mechanism (ARM) designed to equilibrate the confidence levels across all correct responses. By incorporating Prompt Perplexity and Answer Confidence into the advantage estimation, our method dynamically reshapes the reward signal to attenuate the gradient updates of over-confident reasoning paths, while redistributing probability mass toward under-explored correct solutions. Empirical results demonstrate that our approach significantly enhances generative diversity and response entropy while maintaining competitive accuracy, effectively achieving a superior trade-off between exploration and exploitation in reasoning tasks. Empirical results on Qwen2.5 and DeepSeek models across mathematical and coding benchmarks show that ProGRPO significantly mitigates entropy collapse. Specifically, on Qwen2.5-7B, our method outperforms GRPO by 5.7% in Pass@1 and, notably, by 13.9% in Pass@32, highlighting its superior capability in generating diverse correct reasoning paths.

Large language models (LLMs) are promising tools for supporting security management tasks, such as incident response planning. However, their unreliability and tendency to hallucinate remain significant challenges. In this paper, we address these challenges by introducing a principled framework for using an LLM as decision support in security management. Our framework integrates the LLM in an iterative loop where it generates candidate actions that are checked for consistency with system constraints and lookahead predictions. When consistency is low, we abstain from the generated actions and instead collect external feedback, e.g., by evaluating actions in a digital twin. This feedback is then used to refine the candidate actions through in-context learning (ICL). We prove that this design allows to control the hallucination risk by tuning the consistency threshold. Moreover, we establish a bound on the regret of ICL under certain assumptions. To evaluate our framework, we apply it to an incident response use case where the goal is to generate a response and recovery plan based on system logs. Experiments on four public datasets show that our framework reduces recovery times by up to 30% compared to frontier LLMs.

As software systems evolve, patches may unintentionally alter program behavior. Validating patches against their intended semantics is difficult due to incomplete regression tests and informal, non-executable natural language (NL) descriptions of patch intent. We present PatchGuru, the first automated technique that infers executable patch specifications from real-world pull requests (PRs). Given a PR, PatchGuru uses large language models (LLMs) to extract developer intent from NL artifacts and synthesizes patch oracles: under-approximate yet practical specifications expressed as runtime assertions in comparison programs that integrate pre- and post-patch versions. Patch oracles focus on patch-relevant behaviors, enable automated validation, and support cross-version properties. PatchGuru iteratively refines inferred oracles by comparing pre- and post-patch behaviors, identifies violations, filters inconsistencies via self-review, and generates bug reports. We evaluate PatchGuru on 400 recent PRs from four widely used open-source Python projects. PatchGuru reports 39 warnings with a precision of 0.62, yielding 24 confirmed true positives, including 12 previously unknown bugs, 11 of which were subsequently fixed by developers. Compared to the state-of-the-art technique Testora, PatchGuru detects 17 more bugs (24 vs. 7) while improving precision from 0.32 to 0.62. PatchGuru incurs an average cost of 8.9 minutes and USD 0.07 per PR. These results suggest that PatchGuru complements code review and regression testing by providing executable documentation and automated validation of patch intent.

The deployment of Large Language Models (LLMs) on edge devices is fundamentally constrained by the "Memory Wall" -- a hardware limitation where memory bandwidth, not compute, becomes the bottleneck. Recent 1.58-bit quantization techniques (e.g., BitNet b1.58) dramatically reduce memory footprint but typically incur a perplexity degradation of 20-25% compared to FP16 baselines. In this work, we introduce Hybrid Gated Flow (HGF), a dual-stream architecture that couples a 1.58-bit ternary backbone with a learnable, low-rank FP16 correction path controlled by adaptive gates. Through extensive experiments on the TinyStories dataset across two training regimes (2500 and 3500 steps), we demonstrate that HGF 5.4 achieves a validation loss of 0.9306 compared to BitNet's 1.0294, recovering approximately 55% of the quality gap between pure ternary quantization and the FP16 baseline (0.8490). This recovery is achieved with only ~12-15% memory overhead beyond the ternary backbone. Furthermore, we provide empirical evidence for an emergent phenomenon: quantization as structural regularization. While a full-precision differential attention baseline (Diff_Only) exhibited training instability with validation loss exceeding 1.68, the ternary-anchored HGF maintained robust convergence throughout training. Finally, we report preliminary results extending this architecture to 1.2B and 3B parameter models trained on SlimPajama and FineWeb-Edu. These larger-scale experiments confirm that the architectural stability and quality recovery observed in small-scale proxies scale linearly to production-grade language modeling regimes.

Cosine similarity, the standard metric for measuring semantic similarity in vector spaces, is mathematically grounded in the Cauchy-Schwarz inequality, which inherently limits it to capturing linear relationships--a constraint that fails to model the complex, nonlinear structures of real-world semantic spaces. We advance this theoretical underpinning by deriving a tighter upper bound for the dot product than the classical Cauchy-Schwarz bound. This new bound leads directly to recos, a similarity metric that normalizes the dot product by the sorted vector components. recos relaxes the condition for perfect similarity from strict linear dependence to ordinal concordance, thereby capturing a broader class of relationships. Extensive experiments across 11 embedding models--spanning static, contextualized, and universal types--demonstrate that recos consistently outperforms traditional cosine similarity, achieving higher correlation with human judgments on standard Semantic Textual Similarity (STS) benchmarks. Our work establishes recos as a mathematically principled and empirically superior alternative, offering enhanced accuracy for semantic analysis in complex embedding spaces.

Recent applications of Reinforcement Learning with Verifiable Rewards (RLVR) to Large Language Models (LLMs) and Vision-Language Models (VLMs) have demonstrated significant success in enhancing reasoning capabilities for complex tasks. During RLVR training, an increase in response length is often regarded as a key factor contributing to the growth of reasoning ability. However, the patterns of change in response length vary significantly across different RLVR algorithms during the training process. To provide a fundamental explanation for these variations, this paper conducts an in-depth analysis of the components of mainstream RLVR algorithms. We present a theoretical analysis of the factors influencing response length and validate our theory through extensive experimentation. Building upon these theoretical findings, we propose the Length-Unbiased Sequence Policy Optimization (LUSPO) algorithm. Specifically, we rectify the length bias inherent in Group Sequence Policy Optimization (GSPO), rendering its loss function unbiased with respect to response length and thereby resolving the issue of response length collapse. We conduct extensive experiments across mathematical reasoning benchmarks and multimodal reasoning scenarios, where LUSPO consistently achieves superior performance. Empirical results demonstrate that LUSPO represents a novel, state-of-the-art optimization strategy compared to existing methods such as GRPO and GSPO.

Rotary Positional Embedding (RoPE) is a key component of context scaling in Large Language Models (LLMs). While various methods have been proposed to adapt RoPE to longer contexts, their guiding principles generally fall into two categories: (1) out-of-distribution (OOD) mitigation, which scales RoPE frequencies to accommodate unseen positions, and (2) Semantic Modeling, which posits that the attention scores computed with RoPE should always prioritize semantically similar tokens. In this work, we unify these seemingly distinct objectives through a minimalist intervention, namely CoPE: soft clipping lowfrequency components of RoPE. CoPE not only eliminates OOD outliers and refines semantic signals, but also prevents spectral leakage caused by hard clipping. Extensive experiments demonstrate that simply applying our soft clipping strategy to RoPE yields significant performance gains that scale up to 256k context length, validating our theoretical analysis and establishing CoPE as a new state-of-the-art for length generalization. Our code, data, and models are available at https://github.com/hrlics/CoPE.

We present Copyright Detective, the first interactive forensic system for detecting, analyzing, and visualizing potential copyright risks in LLM outputs. The system treats copyright infringement versus compliance as an evidence discovery process rather than a static classification task due to the complex nature of copyright law. It integrates multiple detection paradigms, including content recall testing, paraphrase-level similarity analysis, persuasive jailbreak probing, and unlearning verification, within a unified and extensible framework. Through interactive prompting, response collection, and iterative workflows, our system enables systematic auditing of verbatim memorization and paraphrase-level leakage, supporting responsible deployment and transparent evaluation of LLM copyright risks even with black-box access.

Large-scale multimodal pre-trained models like CLIP rely heavily on high-quality training data, yet raw web-crawled datasets are often noisy, misaligned, and redundant, leading to inefficient training and suboptimal generalization. Existing data selection methods are either heuristic-based, suffering from bias and limited diversity, or data-driven but task-agnostic, failing to optimize for multi-task scenarios. To address these gaps, we introduce TADS (Task-Aware Data Selection), a novel framework for multi-task multimodal pre-training that integrates Intrinsic Quality, Task Relevance, and Distributional Diversity into a learnable value function. TADS employs a comprehensive quality assessment system with unimodal and cross-modal operators, quantifies task relevance via interpretable similarity vectors, and optimizes diversity through cluster-based weighting. A feedback-driven meta-learning mechanism adaptively refines the selection strategy based on proxy model performance across multiple downstream tasks. Experiments on CC12M demonstrate that TADS achieves superior zero-shot performance on benchmarks like ImageNet, CIFAR-100, MS-COCO, and Flickr30K, using only 36% of the data while outperforming baselines by an average of 1.0%. This highlights that TADS significantly enhances data efficiency by curating a high-utility subset that yields a much higher performance ceiling within the same computational constraints.

As general intelligent agents are poised for widespread deployment in diverse households, evaluation tailored to each unique unseen 3D environment has become a critical prerequisite. However, existing benchmarks suffer from severe data contamination and a lack of scene specificity, inadequate for assessing agent capabilities in unseen settings. To address this, we propose a dynamic in-situ task generation method for unseen environments inspired by human cognition. We define tasks through a structured graph representation and construct a two-stage interaction-evolution task generation system for embodied agents (TEA). In the interaction stage, the agent actively interacts with the environment, creating a loop between task execution and generation that allows for continuous task generation. In the evolution stage, task graph modeling allows us to recombine and reuse existing tasks to generate new ones without external data. Experiments across 10 unseen scenes demonstrate that TEA automatically generated 87,876 tasks in two cycles, which human verification confirmed to be physically reasonable and encompassing essential daily cognitive capabilities. Benchmarking SOTA models against humans on our in-situ tasks reveals that models, despite excelling on public benchmarks, perform surprisingly poorly on basic perception tasks, severely lack 3D interaction awareness and show high sensitivity to task types in reasoning. These sobering findings highlight the necessity of in-situ evaluation before deploying agents into real-world human environments.

Vision Transformers achieve strong accuracy but incur high compute and memory cost. Structured pruning can reduce inference cost, but most methods rely on retraining or multi-stage optimization. These requirements limit post-training deployment. We propose \textbf{CORP}, a closed-form one-shot structured pruning framework for Vision Transformers. CORP removes entire MLP hidden dimensions and attention substructures without labels, gradients, or fine-tuning. It operates under strict post-training constraints using only a small unlabeled calibration set. CORP formulates structured pruning as a representation recovery problem. It models removed activations and attention logits as affine functions of retained components and derives closed-form ridge regression solutions that fold compensation into model weights. This minimizes expected representation error under the calibration distribution. Experiments on ImageNet with DeiT models show strong redundancy in MLP and attention representations. Without compensation, one-shot structured pruning causes severe accuracy degradation. With CORP, models preserve accuracy under aggressive sparsity. On DeiT-Huge, CORP retains 82.8\% Top-1 accuracy after pruning 50\% of both MLP and attention structures. CORP completes pruning in under 20 minutes on a single GPU and delivers substantial real-world efficiency gains.

Agentic Test-Time Scaling (TTS) has delivered state-of-the-art (SOTA) performance on complex software engineering tasks such as code generation and bug fixing. However, its practical adoption remains limited due to significant computational overhead, primarily driven by two key challenges: (1) the high cost associated with deploying excessively large ensembles, and (2) the lack of a reliable mechanism for selecting the optimal candidate solution, ultimately constraining the performance gains that can be realized. To address these challenges, we propose Entropy-Guided Stepwise Scaling (EGSS), a novel TTS framework that dynamically balances efficiency and effectiveness through entropy-guided adaptive search and robust test-suite augmentation. Extensive experiments on SWE-Bench-Verified demonstrate that EGSS consistently boosts performance by 5-10% across all evaluated models. Specifically, it increases the resolved ratio of Kimi-K2-Intruct from 63.2% to 72.2%, and GLM-4.6 from 65.8% to 74.6%. Furthermore, when paired with GLM-4.6, EGSS achieves a new state-of-the-art among open-source large language models. In addition to these accuracy improvements, EGSS reduces inference-time token usage by over 28% compared to existing TTS methods, achieving simultaneous gains in both effectiveness and computational efficiency.

This study explores the integration of multiple Explainable AI (XAI) techniques to enhance the interpretability of deep learning models for brain tumour detection. A custom Convolutional Neural Network (CNN) was developed and trained on the BraTS 2021 dataset, achieving 91.24% accuracy in distinguishing between tumour and non-tumour regions. This research combines Gradient-weighted Class Activation Mapping (GRAD-CAM), Layer-wise Relevance Propagation (LRP) and SHapley Additive exPlanations (SHAP) to provide comprehensive insights into the model's decision-making process. This multi-technique approach successfully identified both full and partial tumours, offering layered explanations ranging from broad regions of interest to pixel-level details. GRAD-CAM highlighted important spatial regions, LRP provided detailed pixel-level relevance and SHAP quantified feature contributions. The integrated approach effectively explained model predictions, including cases with partial tumour visibility thus showing superior explanatory power compared to individual XAI methods. This research enhances transparency and trust in AI-driven medical imaging analysis by offering a more comprehensive perspective on the model's reasoning. The study demonstrates the potential of integrated XAI techniques in improving the reliability and interpretability of AI systems in healthcare, particularly for critical tasks like brain tumour detection.

Die casting plays a crucial role across various industries due to its ability to craft intricate shapes with high precision and smooth surfaces. However, surface defects remain a major issue that impedes die casting quality control. Recently, computer vision techniques have been explored to automate and improve defect detection. In this work, we combine local neighbor-aware patch features with a normalizing flow model and bridge the gap between the generic pretrained feature extractor and industrial product images by introducing an adapter module to increase the efficiency and accuracy of automated anomaly detection. Compared to state-of-the-art methods, our approach reduces the error rate by 20\% on the MVTec AD dataset, achieving an image-level AUROC of 99.28\%. Our approach has also enhanced performance on the VisA dataset , achieving an image-level AUROC of 96.48\%. Compared to the state-of-the-art models, this represents a 28.2\% reduction in error. Additionally, experiments on a proprietary die casting dataset yield an accuracy of 95.77\% for anomaly detection, without requiring any anomalous samples for training. Our method illustrates the potential of leveraging computer vision and deep learning techniques to advance inspection capabilities for the die casting industry

Retrieval-Augmented Generation (RAG) enhances Large Language Models (LLMs) by grounding generation in external knowledge to improve factuality and reduce hallucinations. Yet most deployments assume a centralized corpus, which is infeasible in privacy aware domains where knowledge remains siloed. This motivates federated RAG (FedRAG), where a central LLM server collaborates with distributed silos without sharing raw documents. In context RAG violates this requirement by transmitting verbatim documents, whereas parametric RAG encodes documents into lightweight adapters that merge with a frozen LLM at inference, avoiding raw-text exchange. We adopt the parametric approach but face two unique challenges induced by FedRAG: high storage and communication from per-document adapters, and destructive aggregation caused by indiscriminately merging multiple adapters. We present FedMosaic, the first federated RAG framework built on parametric adapters. FedMosaic clusters semantically related documents into multi-document adapters with document-specific masks to reduce overhead while preserving specificity, and performs selective adapter aggregation to combine only relevance-aligned, nonconflicting adapters. Experiments show that FedMosaic achieves an average 10.9% higher accuracy than state-of-the-art methods in four categories, while lowering storage costs by 78.8% to 86.3% and communication costs by 91.4%, and never sharing raw documents.

Intervention-based model steering offers a lightweight and interpretable alternative to prompting and fine-tuning. However, by adapting strong optimization objectives from fine-tuning, current methods are susceptible to overfitting and often underperform, sometimes generating unnatural outputs. We hypothesize that this is because effective steering requires the faithful identification of internal model mechanisms, not the enforcement of external preferences. To this end, we build on the principles of distributed alignment search (DAS), the standard for causal variable localization, to propose a new steering method: Concept DAS (CDAS). While we adopt the core mechanism of DAS, distributed interchange intervention (DII), we introduce a novel distribution matching objective tailored for the steering task by aligning intervened output distributions with counterfactual distributions. CDAS differs from prior work in two main ways: first, it learns interventions via weak-supervised distribution matching rather than probability maximization; second, it uses DIIs that naturally enable bi-directional steering and allow steering factors to be derived from data, reducing the effort required for hyperparameter tuning and resulting in more faithful and stable control. On AxBench, a large-scale model steering benchmark, we show that CDAS does not always outperform preference-optimization methods but may benefit more from increased model scale. In two safety-related case studies, overriding refusal behaviors of safety-aligned models and neutralizing a chain-of-thought backdoor, CDAS achieves systematic steering while maintaining general model utility. These results indicate that CDAS is complementary to preference-optimization approaches and conditionally constitutes a robust approach to intervention-based model steering. Our code is available at https://github.com/colored-dye/concept_das.

Graph anomaly detection (GAD) is crucial in applications like fraud detection and cybersecurity. Despite recent advancements using graph neural networks (GNNs), two major challenges persist. At the model level, most methods adopt a transductive learning paradigm, which assumes static graph structures, making them unsuitable for dynamic, evolving networks. At the data level, the extreme class imbalance, where anomalous nodes are rare, leads to biased models that fail to generalize to unseen anomalies. These challenges are interdependent: static transductive frameworks limit effective data augmentation, while imbalance exacerbates model distortion in inductive learning settings. To address these challenges, we propose a novel data-centric framework that integrates dynamic graph modeling with balanced anomaly synthesis. Our framework features: (1) a discrete ego-graph diffusion model, which captures the local topology of anomalies to generate ego-graphs aligned with anomalous structural distribution, and (2) a curriculum anomaly augmentation mechanism, which dynamically adjusts synthetic data generation during training, focusing on underrepresented anomaly patterns to improve detection and generalization. Experiments on five datasets demonstrate that the effectiveness of our framework.

Linear attention methods offer Transformers $O(N)$ complexity but typically underperform standard softmax attention. We identify two fundamental limitations affecting these approaches: the restriction to convex combinations that only permits additive information blending, and uniform accumulated weight bias that dilutes attention in long contexts. We propose Zero-Sum Linear Attention (ZeroS), which addresses these limitations by removing the constant zero-order term $1/t$ and reweighting the remaining zero-sum softmax residuals. This modification creates mathematically stable weights, enabling both positive and negative values and allowing a single attention layer to perform contrastive operations. While maintaining $O(N)$ complexity, ZeroS theoretically expands the set of representable functions compared to convex combinations. Empirically, it matches or exceeds standard softmax attention across various sequence modeling benchmarks.

Harmful fine-tuning can invalidate safety alignment of large language models, exposing significant safety risks. In this paper, we utilize the attention sink mechanism to mitigate harmful fine-tuning. Specifically, we first measure a statistic named \emph{sink divergence} for each attention head and observe that \emph{different attention heads exhibit two different signs of sink divergence}. To understand its safety implications, we conduct experiments and find that the number of attention heads of positive sink divergence increases along with the increase of the model's harmfulness when undergoing harmful fine-tuning. Based on this finding, we propose a separable sink divergence hypothesis -- \emph{attention heads associating with learning harmful patterns during fine-tuning are separable by their sign of sink divergence}. Based on the hypothesis, we propose a fine-tuning-stage defense, dubbed Surgery. Surgery utilizes a regularizer for sink divergence suppression, which steers attention heads toward the negative sink divergence group, thereby reducing the model's tendency to learn and amplify harmful patterns. Extensive experiments demonstrate that Surgery improves defense performance by 5.90\%, 11.25\%, and 9.55\% on the BeaverTails, HarmBench, and SorryBench benchmarks, respectively. Source code is available on https://github.com/Lslland/Surgery.

Gradient flows of the Kullback--Leibler (KL) divergence, such as the Fokker--Planck equation and Stein Variational Gradient Descent, evolve a distribution toward a target density known only up to a normalizing constant. We introduce new gradient flows of the KL divergence with a remarkable combination of properties: they admit accurate interacting-particle approximations in high dimensions, and the per-step cost scales linearly in both the number of particles and the dimension. These gradient flows are based on new transportation-based Riemannian geometries on the space of probability measures: the Radon--Wasserstein geometry and the related Regularized Radon--Wasserstein (RRW) geometry. We define these geometries using the Radon transform so that the gradient-flow velocities depend only on one-dimensional projections. This yields interacting-particle-based algorithms whose per-step cost follows from efficient Fast Fourier Transform-based evaluation of the required 1D convolutions. We additionally provide numerical experiments that study the performance of the proposed algorithms and compare convergence behavior and quantization. Finally, we prove some theoretical results including well-posedness of the flows and long-time convergence guarantees for the RRW flow.

Current audio foundation models typically rely on rigid, task-specific supervision, addressing isolated factors of audio rather than the whole. In contrast, human intelligence processes audio holistically, seamlessly bridging physical signals with abstract cognitive concepts to execute complex tasks. Grounded in this philosophy, we introduce Bagpiper, an 8B audio foundation model that interprets physical audio via rich captions, i.e., comprehensive natural language descriptions that encapsulate the critical cognitive concepts inherent in the signal (e.g., transcription, audio events). By pre-training on a massive corpus of 600B tokens, the model establishes a robust bidirectional mapping between raw audio and this high-level conceptual space. During fine-tuning, Bagpiper adopts a caption-then-process workflow, simulating an intermediate cognitive reasoning step to solve diverse tasks without task-specific priors. Experimentally, Bagpiper outperforms Qwen-2.5-Omni on MMAU and AIRBench for audio understanding and surpasses CosyVoice3 and TangoFlux in generation quality, capable of synthesizing arbitrary compositions of speech, music, and sound effects. To the best of our knowledge, Bagpiper is among the first works that achieve unified understanding generation for general audio. Model, data, and code are available at Bagpiper Home Page.

We study differentially private prediction introduced by Dwork and Feldman (COLT 2018): an algorithm receives one labeled sample set $S$ and then answers a stream of unlabeled queries while the output transcript remains $(\varepsilon,δ)$-differentially private with respect to $S$. Standard composition yields a $\sqrt{T}$ dependence for $T$ queries. We show that this dependence can be reduced to polylogarithmic in $T$ in streaming settings. For an oblivious online adversary and any concept class $\mathcal{C}$, we give a private predictor that answers $T$ queries with $|S|= \tilde{O}(VC(\mathcal{C})^{3.5}\log^{3.5}T)$ labeled examples. For an adaptive online adversary and halfspaces over $\mathbb{R}^d$, we obtain $|S|=\tilde{O}\left(d^{5.5}\log T\right)$.

Searching for mathematical results remains difficult: most existing tools retrieve entire papers, while mathematicians and theorem-proving agents often seek a specific theorem, lemma, or proposition that answers a query. While semantic search has seen rapid progress, its behavior on large, highly technical corpora such as research-level mathematical theorems remains poorly understood. In this work, we introduce and study semantic theorem retrieval at scale over a unified corpus of $9.2$ million theorem statements extracted from arXiv and seven other sources, representing the largest publicly available corpus of human-authored, research-level theorems. We represent each theorem with a short natural-language description as a retrieval representation and systematically analyze how representation context, language model choice, embedding model, and prompting strategy affect retrieval quality. On a curated evaluation set of theorem-search queries written by professional mathematicians, our approach substantially improves both theorem-level and paper-level retrieval compared to existing baselines, demonstrating that semantic theorem search is feasible and effective at web scale. The theorem search tool is available at \href{https://huggingface.co/spaces/uw-math-ai/theorem-search}{this link}, and the dataset is available at \href{https://huggingface.co/datasets/uw-math-ai/TheoremSearch}{this link}.

Disentangled representation learning aims to capture the underlying explanatory factors of observed data, enabling a principled understanding of the data-generating process. Recent advances in generative modeling have introduced new paradigms for learning such representations. However, existing diffusion-based methods encourage factor independence via inductive biases, yet frequently lack strong semantic alignment. In this work, we propose a flow matching-based framework for disentangled representation learning, which casts disentanglement as learning factor-conditioned flows in a compact latent space. To enforce explicit semantic alignment, we introduce a non-overlap (orthogonality) regularizer that suppresses cross-factor interference and reduces information leakage between factors. Extensive experiments across multiple datasets demonstrate consistent improvements over representative baselines, yielding higher disentanglement scores as well as improved controllability and sample fidelity.

The academia and industry are characterized by a reciprocal shaping and dynamic feedback mechanism. Despite distinct institutional logics, they have adapted closely in collaborative publishing and talent mobility, demonstrating tension between institutional divergence and intensive collaboration. Existing studies on their knowledge proximity mainly rely on macro indicators such as the number of collaborative papers or patents, lacking an analysis of knowledge units in the literature. This has led to an insufficient grasp of fine-grained knowledge proximity between industry and academia, potentially undermining collaboration frameworks and resource allocation efficiency. To remedy the limitation, this study quantifies the trajectory of academia-industry co-evolution through fine-grained entities and semantic space. In the entity measurement part, we extract fine-grained knowledge entities via pre-trained models, measure sequence overlaps using cosine similarity, and analyze topological features through complex network analysis. At the semantic level, we employ unsupervised contrastive learning to quantify convergence in semantic spaces by measuring cross-institutional textual similarities. Finally, we use citation distribution patterns to examine correlations between bidirectional knowledge flows and similarity. Analysis reveals that knowledge proximity between academia and industry rises, particularly following technological change. This provides textual evidence of bidirectional adaptation in co-evolution. Additionally, academia's knowledge dominance weakens during technological paradigm shifts. The dataset and code for this paper can be accessed at https://github.com/tinierZhao/Academic-Industrial-associations.

Although diffusion-based, non-autoregressive text-to-speech (TTS) systems have demonstrated impressive zero-shot synthesis capabilities, their efficacy is still hindered by two key challenges: the difficulty of text-speech alignment modeling and the high computational overhead of the iterative denoising process. To address these limitations, we propose ARCHI-TTS that features a dedicated semantic aligner to ensure robust temporal and semantic consistency between text and audio. To overcome high computational inference costs, ARCHI-TTS employs an efficient inference strategy that reuses encoder features across denoising steps, drastically accelerating synthesis without performance degradation. An auxiliary CTC loss applied to the condition encoder further enhances the semantic understanding. Experimental results demonstrate that ARCHI-TTS achieves a WER of 1.98% on LibriSpeech-PC test-clean, and 1.47%/1.42% on SeedTTS test-en/test-zh with a high inference efficiency, consistently outperforming recent state-of-the-art TTS systems.

Most services built on powerful large-scale language models (LLMs) add citations to their output to enhance credibility. Recent research has paid increasing attention to the question of what reference documents to link to outputs. However, how LLMs recognize cite-worthiness and how this process should be controlled remains underexplored. In this study, we focus on what kinds of content LLMs currently tend to cite and how well that behavior aligns with human preferences. We construct a dataset to characterize the relationship between human citation preferences and LLM behavior. Web-derived texts are categorized into eight citation-motivation types, and pairwise citation preferences are exhaustively evaluated across all type combinations to capture fine-grained contrasts. Our results show that humans most frequently seek citations for medical text, and stronger models display a similar tendency. We also find that current models are as much as $27\%$ more likely than humans to add citations to text that is explicitly marked as needing citations on sources such as Wikipedia, and this overemphasis reduces alignment accuracy. Conversely, models systematically underselect numeric sentences (by $-22.6\%$ relative to humans) and sentences containing personal names (by $-20.1\%$), categories for which humans typically demand citations. Furthermore, experiments with Direct Preference Optimization demonstrate that model behavior can be calibrated to better match human citation preferences. We expect this study to provide a foundation for more fine-grained investigations into LLM citation preferences.

Accurate forecasting of extreme weather events such as heavy rainfall or storms is critical for risk management and disaster mitigation. Although high-resolution radar observations have spurred extensive research on nowcasting models, precipitation nowcasting remains particularly challenging due to pronounced spatial locality, intricate fine-scale rainfall structures, and variability in forecasting horizons. While recent diffusion-based generative ensembles show promising results, they are computationally expensive and unsuitable for real-time applications. In contrast, deterministic models are computationally efficient but remain biased toward normal rainfall. Furthermore, the benchmark datasets commonly used in prior studies are themselves skewed--either dominated by ordinary rainfall events or restricted to extreme rainfall episodes--thereby hindering general applicability in real-world settings. In this paper, we propose exPreCast, an efficient deterministic framework for generating finely detailed radar forecasts, and introduce a newly constructed balanced radar dataset from the Korea Meteorological Administration (KMA), which encompasses both ordinary precipitation and extreme events. Our model integrates local spatiotemporal attention, a texture-preserving cubic dual upsampling decoder, and a temporal extractor to flexibly adjust forecasting horizons. Experiments on established benchmarks (SEVIR and MeteoNet) as well as on the balanced KMA dataset demonstrate that our approach achieves state-of-the-art performance, delivering accurate and reliable nowcasts across both normal and extreme rainfall regimes.

Retrieval-augmented generation (RAG) promises grounded question answering, yet domain settings with multiple heterogeneous knowledge bases (KBs) remain challenging. In Chinese Tibetan medicine, encyclopedia entries are often dense and easy to match, which can dominate retrieval even when classics or clinical papers provide more authoritative evidence. We study a practical setting with three KBs (encyclopedia, classics, and clinical papers) and a 500-query benchmark (cutoff $K{=}5$) covering both single-KB and cross-KB questions. We propose two complementary methods to improve traceability, reduce hallucinations, and enable cross-KB verification. First, DAKS performs KB routing and budgeted retrieval to mitigate density-driven bias and to prioritize authoritative sources when appropriate. Second, we use an alignment graph to guide evidence fusion and coverage-aware packing, improving cross-KB evidence coverage without relying on naive concatenation. All answers are generated by a lightweight generator, \textsc{openPangu-Embedded-7B}. Experiments show consistent gains in routing quality and cross-KB evidence coverage, with the full system achieving the best CrossEv@5 while maintaining strong faithfulness and citation correctness.

To assess the ability of current AI systems to correctly answer research-level mathematics questions, we share a set of ten math questions which have arisen naturally in the research process of the authors. The questions had not been shared publicly until now; the answers are known to the authors of the questions but will remain encrypted for a short time.

As long-context inference becomes central to large language models (LLMs), attention over growing key-value caches emerges as a dominant decoding bottleneck, motivating sparse attention for scalable inference. Fixed-budget top-k sparse attention cannot adapt to heterogeneous attention distributions across heads and layers, whereas top-p sparse attention directly preserves attention mass and provides stronger accuracy guarantees. Existing top-p methods, however, fail to jointly optimize top-p accuracy, selection overhead, and sparse attention cost, which limits their overall efficiency. We present Double-P, a hierarchical sparse attention framework that optimizes all three stages. Double-P first performs coarse-grained top-p estimation at the cluster level using size-weighted centroids, then adaptively refines computation through a second top-p stage that allocates token-level attention only when needed. Across long-context benchmarks, Double-P consistently achieves near-zero accuracy drop, reducing attention computation overhead by up to 1.8x and delivers up to 1.3x end-to-end decoding speedup over state-of-the-art fixed-budget sparse attention methods.

As internet access expands, so does exposure to harmful content, increasing the need for effective moderation. Research has demonstrated that large language models (LLMs) can be effectively utilized for social media moderation tasks, including harmful content detection. While proprietary LLMs have been shown to zero-shot outperform traditional machine learning models, the out-of-the-box capability of open-weight LLMs remains an open question. Motivated by recent developments of reasoning LLMs, we evaluate seven state-of-the-art models: four proprietary and three open-weight. Testing with real-world posts on Bluesky, moderation decisions by Bluesky Moderation Service, and annotations by two authors, we find a considerable degree of overlap between the sensitivity (81%--97%) and specificity (91%--100%) of the open-weight LLMs and those (72%--98%, and 93%--99%) of the proprietary ones. Additionally, our analysis reveals that specificity exceeds sensitivity for rudeness detection, but the opposite holds for intolerance and threats. Lastly, we identify inter-rater agreement across human moderators and the LLMs, highlighting considerations for deploying LLMs in both platform-scale and personalized moderation contexts. These findings show open-weight LLMs can support privacy-preserving moderation on consumer-grade hardware and suggest new directions for designing moderation systems that balance community values with individual user preferences.

Spectral neural operators, particularly Fourier Neural Operators (FNO), are a powerful framework for learning solution operators of partial differential equations (PDEs) due to their efficient global mixing in the frequency domain. However, existing spectral operators rely on static Fourier kernels applied uniformly across inputs, limiting their ability to capture multi-scale, regime-dependent, and anisotropic dynamics governed by the global state of the system. We introduce SpectraKAN, a neural operator that conditions the spectral operator on the input itself, turning static spectral convolution into an input-conditioned integral operator. This is achieved by extracting a compact global representation from spatio-temporal history and using it to modulate a multi-scale Fourier trunk via single-query cross-attention, enabling the operator to adapt its behaviour while retaining the efficiency of spectral mixing. We provide theoretical justification showing that this modulation converges to a resolution-independent continuous operator under mesh refinement and KAN gives smooth, Lipschitz-controlled global modulation. Across diverse PDE benchmarks, SpectraKAN achieves state-of-the-art performance, reducing RMSE by up to 49% over strong baselines, with particularly large gains on challenging spatio-temporal prediction tasks.

Neural networks organize information according to the hierarchical, multi-scale structure of natural data. Methods to interpret model internals should be similarly scale-aware, explicitly tracking how features compose across resolutions and guaranteeing bounds on the influence of fine-grained structure that is discarded as irrelevant noise. We posit that the renormalisation framework from physics can meet this need by offering technical tools that can overcome limitations of current methods. Moreover, relevant work from adjacent fields has now matured to a point where scattered research threads can be synthesized into practical, theory-informed tools. To combine these threads in an AI safety context, we propose a unifying research agenda -- \emph{scale-aware interpretability} -- to develop formal machinery and interpretability tools that have robustness and faithfulness properties supported by statistical physics.

Large language models (LLMs) are increasingly trained in complex Reinforcement Learning, multi-agent environments, making it difficult to understand how behavior changes over training. Sparse Autoencoders (SAEs) have recently shown to be useful for data-centric interpretability. In this work, we analyze large-scale reinforcement learning training runs from the sophisticated environment of Full-Press Diplomacy by applying pretrained SAEs, alongside LLM-summarizer methods. We introduce Meta-Autointerp, a method for grouping SAE features into interpretable hypotheses about training dynamics. We discover fine-grained behaviors including role-playing patterns, degenerate outputs, language switching, alongside high-level strategic behaviors and environment-specific bugs. Through automated evaluation, we validate that 90% of discovered SAE Meta-Features are significant, and find a surprising reward hacking behavior. However, through two user studies, we find that even subjectively interesting and seemingly helpful SAE features may be worse than useless to humans, along with most LLM generated hypotheses. However, a subset of SAE-derived hypotheses are predictively useful for downstream tasks. We further provide validation by augmenting an untrained agent's system prompt, improving the score by +14.2%. Overall, we show that SAEs and LLM-summarizer provide complementary views into agent behavior, and together our framework forms a practical starting point for future data-centric interpretability work on ensuring trustworthy LLM behavior throughout training.

Model collaboration -- systems where multiple language models (LMs) collaborate -- combines the strengths of diverse models with cost in loading multiple LMs. We improve efficiency while preserving the strengths of collaboration by distilling collaborative patterns into a single model, where the model is trained on the outputs of the model collaboration system. At inference time, only the distilled model is employed: it imitates the collaboration while only incurring the cost of a single model. Furthermore, we propose the single-multi evolution loop: multiple LMs collaborate, each distills from the collaborative outputs, and these post-distillation improved LMs collaborate again, forming a collective evolution ecosystem where models evolve and self-improve by interacting with an environment of other models. Extensive experiments with 7 collaboration strategies and 15 tasks (QA, reasoning, factuality, etc.) demonstrate that: 1) individual models improve by 8.0% on average, absorbing the strengths of collaboration while reducing the cost to a single model; 2) the collaboration also benefits from the stronger and more synergistic LMs after distillation, improving over initial systems without evolution by 14.9% on average. Analysis reveals that the single-multi evolution loop outperforms various existing evolutionary AI methods, is compatible with diverse model/collaboration/distillation settings, and helps solve problems where the initial model/system struggles to.

A major bottleneck in scenario-based Sample Average Approximation (SAA) for stochastic programming (SP) is the cost of solving an exact second-stage problem for every scenario, especially when each scenario contains an NP-hard combinatorial structure. This has led much of the SP literature to restrict the second stage to linear or simplified models. We develop a GPU-based framework that makes full-fidelity integer second-stage models tractable at scale. The key innovation is a set of hardware-aware, scenario-batched GPU kernels that expose parallelism across scenarios, dynamic-programming (DP) layers, and route or action options, enabling Bellman updates to be executed in a single pass over more than 1,000,000 realizations. We evaluate the approach in two representative SP settings: a vectorized split operator for stochastic vehicle routing and a DP for inventory reinsertion. Implementation scales nearly linearly in the number of scenarios and achieves a one-two to four-five orders of magnitude speedup, allowing far larger scenario sets and reliably stronger first-stage decisions. The computational leverage directly improves decision quality: much larger scenario sets and many more first-stage candidates can be evaluated within fixed time budgets, consistently yielding stronger SAA solutions. Our results show that full-fidelity integer second-stage models are tractable at scales previously considered impossible, providing a practical path to large-scale, realistic stochastic discrete optimization.

Coastal hypoxia, especially in the northern part of Gulf of Mexico, presents a persistent ecological and economic concern. Seasonal models offer coarse forecasts that miss the fine-scale variability needed for daily, responsive ecosystem management. We present study that compares four deep learning architectures for daily hypoxia classification: Bidirectional Long Short-Term Memory (BiLSTM), Medformer (Medical Transformer), Spatio-Temporal Transformer (ST-Transformer), and Temporal Convolutional Network (TCN). We trained our models with twelve years of daily hindcast data from 2009-2020 Our training data consists of 2009-2020 hindcast data from a coupled hydrodynamic-biogeochemical model. Similarly, we use hindcast data from 2020 through 2024 as a test data. We constructed classification models incorporating water column stratification, sediment oxygen consumption, and temperature-dependent decomposition rates. We evaluated each architectures using the same data preprocessing, input/output formulation, and validation protocols. Each model achieved high classification accuracy and strong discriminative ability with ST-Transformer achieving the highest performance across all metrics and tests periods (AUC-ROC: 0.982-0.992). We also employed McNemar's method to identify statistically significant differences in model predictions. Our contribution is a reproducible framework for operational real-time hypoxia prediction that can support broader efforts in the environmental and ocean modeling systems community and in ecosystem resilience. The source code is available https://github.com/rmagesh148/hypoxia-ai/

Language models (LMs) are increasingly used in collaboration: multiple LMs trained by different parties collaborate through routing systems, multi-agent debate, model merging, and more. Critical safety risks remain in this decentralized paradigm: what if some of the models in multi-LLM systems are compromised or malicious? We first quantify the impact of malicious models by engineering four categories of malicious LMs, plug them into four types of popular model collaboration systems, and evaluate the compromised system across 10 datasets. We find that malicious models have a severe impact on the multi-LLM systems, especially for reasoning and safety domains where performance is lowered by 7.12% and 7.94% on average. We then propose mitigation strategies to alleviate the impact of malicious components, by employing external supervisors that oversee model collaboration to disable/mask them out to reduce their influence. On average, these strategies recover 95.31% of the initial performance, while making model collaboration systems fully resistant to malicious models remains an open research question.

Riemannian flow matching (RFM) extends flow-based generative modeling to data supported on manifolds by learning a time-dependent tangent vector field whose flow-ODE transports a simple base distribution to the data law. We develop a nonasymptotic Total Variation (TV) convergence analysis for RFM samplers that use a learned vector field together with Euler discretization on manifolds. Our key technical ingredient is a differential inequality governing the evolution of TV between two manifold ODE flows, which expresses the time-derivative of TV through the divergence of the vector-field mismatch and the score of the reference flow; controlling these terms requires establishing new bounds that explicitly account for parallel transport and curvature. Under smoothness assumptions on the population flow-matching field and either uniform (compact manifolds) or mean-square (Hadamard manifolds) approximation guarantees for the learned field, we obtain explicit bounds of the form $\mathrm{TV}\le C_{\mathrm{Lip}}\,h + C_{\varepsilon}\,\varepsilon$ (with an additional higher-order $\varepsilon^2$ term on compact manifolds), cleanly separating numerical discretization and learning errors. Here, $h$ is the step-size and $\varepsilon$ is the target accuracy. Instantiations yield \emph{explicit} polynomial iteration complexities on the hypersphere $S^d$, and on the SPD$(n)$ manifolds under mild moment conditions.

We derive finite-particle rates for the regularized Stein variational gradient descent (R-SVGD) algorithm introduced by He et al. (2024) that corrects the constant-order bias of the SVGD by applying a resolvent-type preconditioner to the kernelized Wasserstein gradient. For the resulting interacting $N$-particle system, we establish explicit non-asymptotic bounds for time-averaged (annealed) empirical measures, illustrating convergence in the \emph{true} (non-kernelized) Fisher information and, under a $\mathrm{W}_1\mathrm{I}$ condition on the target, corresponding $\mathrm{W}_1$ convergence for a large class of smooth kernels. Our analysis covers both continuous- and discrete-time dynamics and yields principled tuning rules for the regularization parameter, step size, and averaging horizon that quantify the trade-off between approximating the Wasserstein gradient flow and controlling finite-particle estimation error.

The study by Jung et al. (Jung H, Covino R, Arjun A, et al., Nat Comput Sci. 3:334-345 (2023)) introduced Artificial Intelligence for Molecular Mechanism Discovery (AIMMD), a novel sampling algorithm that integrates machine learning to enhance the efficiency of transition path sampling (TPS). By enabling on-the-fly estimation of the committor probability and simultaneously deriving a human-interpretable reaction coordinate, AIMMD offers a robust framework for elucidating the mechanistic pathways of complex molecular processes. This commentary provides a discussion and critical analysis of the core AIMMD framework, explores its recent extensions, and offers an assessment of the method's potential impact and limitations.

Reinforcement Learning with Verifiable Rewards (RLVR) has proven effective for enhancing the reasoning capabilities of Large Language Models (LLMs). However, dominant approaches like Group Relative Policy Optimization (GRPO) face critical stability challenges: they suffer from high estimator variance under computational constraints (small group sizes) and vanishing gradient signals in saturated failure regimes where all responses yield identical zero rewards. To address this, we propose Empirical Bayes Policy Optimization (EBPO), a novel framework that regularizes local group-based baselines by borrowing strength from the policy's accumulated global statistics. Instead of estimating baselines in isolation, EBPO employs a shrinkage estimator that dynamically balances local group statistics with a global prior updated via Welford's online algorithm. Theoretically, we demonstrate that EBPO guarantees strictly lower Mean Squared Error (MSE), bounded entropy decay, and non-vanishing penalty signals in failure scenarios compared to GRPO. Empirically, EBPO consistently outperforms GRPO and other established baselines across diverse benchmarks, including AIME and OlympiadBench. Notably, EBPO exhibits superior training stability, achieving high-performance gains even with small group sizes, and benefits significantly from difficulty-stratified curriculum learning.

Foundation models are trained on broad data distributions, yielding generalist capabilities that enable many downstream applications but also expand the space of potential misuse and failures. This position paper argues that capability control -- imposing restrictions on permissible model behavior -- should be treated as a distinct goal from alignment. While alignment is often context and preference-driven, capability control aims to impose hard operational limits on permissible behaviors, including under adversarial elicitation. We organize capability control mechanisms across the model lifecycle into three layers: (i) data-based control of the training distribution, (ii) learning-based control via weight- or representation-level interventions, and (iii) system-based control via post-deployment guardrails over inputs, outputs, and actions. Because each layer has characteristic failure modes when used in isolation, we advocate for a defense-in-depth approach that composes complementary controls across the full stack. We further outline key open challenges in achieving such control, including the dual-use nature of knowledge and compositional generalization.

Deep neural networks often inherit social and demographic biases from annotated data during model training, leading to unfair predictions, especially in the presence of sensitive attributes like race, age, gender etc. Existing methods fall prey to the inherent data imbalance between attribute groups and inadvertently emphasize on sensitive attributes, worsening unfairness and performance. To surmount these challenges, we propose SHaSaM (Submodular Hard Sample Mining), a novel combinatorial approach that models fairness-driven representation learning as a submodular hard-sample mining problem. Our two-stage approach comprises of SHaSaM-MINE, which introduces a submodular subset selection strategy to mine hard positives and negatives - effectively mitigating data imbalance, and SHaSaM-LEARN, which introduces a family of combinatorial loss functions based on Submodular Conditional Mutual Information to maximize the decision boundary between target classes while minimizing the influence of sensitive attributes. This unified formulation restricts the model from learning features tied to sensitive attributes, significantly enhancing fairness without sacrificing performance. Experiments on CelebA and UTKFace demonstrate that SHaSaM achieves state-of-the-art results, with up to 2.7 points improvement in model fairness (Equalized Odds) and a 3.5% gain in Accuracy, within fewer epochs as compared to existing methods.

The intersection of Safety of Intended Functionality (SOTIF) and Functional Safety (FuSa) analysis of driving automation features has traditionally excluded Quality Management (QM) components from rigorous safety impact evaluations. While QM components are not typically classified as safety-relevant, recent developments in artificial intelligence (AI) integration reveal that such components can contribute to SOTIF-related hazardous risks. Compliance with emerging AI safety standards, such as ISO/PAS 8800, necessitates re-evaluating safety considerations for these components. This paper examines the necessity of conducting holistic safety analysis and risk assessment on AI components, emphasizing their potential to introduce hazards with the capacity to violate risk acceptance criteria when deployed in safety-critical driving systems, particularly in perception algorithms. Using case studies, we demonstrate how deficiencies in AI-driven perception systems can emerge even in QM-classified components, leading to unintended functional behaviors with critical safety implications. By bridging theoretical analysis with practical examples, this paper argues for the adoption of comprehensive FuSa, SOTIF, and AI standards-driven methodologies to identify and mitigate risks in AI components. The findings demonstrate the importance of revising existing safety frameworks to address the evolving challenges posed by AI, ensuring comprehensive safety assurance across all component classifications spanning multiple safety standards.

Large Language Models (LLMs) are commonly trained on multilingual corpora that include Greek, yet reliable evaluation benchmarks for Greek-particularly those based on authentic, native-sourced content-remain limited. Existing datasets are often machine-translated from English, failing to capture Greek linguistic and cultural characteristics. We introduce GreekMMLU, a native-sourced benchmark for massive multitask language understanding in Greek, comprising 21,805 multiple-choice questions across 45 subject areas, organized under a newly defined subject taxonomy and annotated with educational difficulty levels spanning primary to professional examinations. All questions are sourced or authored in Greek from academic, professional, and governmental exams. We publicly release 16,857 samples and reserve 4,948 samples for a private leaderboard to enable robust and contamination-resistant evaluation. Evaluations of over 80 open- and closed-source LLMs reveal substantial performance gaps between frontier and open-weight models, as well as between Greek-adapted models and general multilingual ones. Finally, we provide a systematic analysis of factors influencing performance-including model scale, adaptation, and prompting-and derive insights for improving LLM capabilities in Greek.

Parameter-Efficient Fine-Tuning (PEFT) has emerged as a practical paradigm for adapting large language models (LLMs) without updating all parameters. Most existing approaches, such as LoRA and PiSSA, rely on low-rank decompositions of weight updates. However, the low-rank assumption may restrict expressivity, particularly in task-specific adaptation scenarios where singular values are distributed relatively uniformly. To address this limitation, we propose CoSA (Compressed Sensing-Based Adaptation), a new PEFT method extended from compressed sensing theory. Instead of constraining weight updates to a low-rank subspace, CoSA expresses them through fixed random projection matrices and a compact learnable core. We provide a formal theoretical analysis of CoSA as a synthesis process, proving that weight updates can be compactly encoded into a low-dimensional space and mapped back through random projections. Extensive experimental results show that CoSA provides a principled perspective for efficient and expressive multi-scale model adaptation. Specifically, we evaluate CoSA on 10 diverse tasks, including natural language understanding and generation, employing 5 models of different scales from RoBERTa, Llama, and Qwen families. Across these settings, CoSA consistently matches or outperforms state-of-the-art PEFT methods.

Machine systems inherently generate signals in which fault conditions and various physical variables are physically coupled. Although many existing fault classification studies rely solely on direct fault labels, the aforementioned signals naturally embed additional information shaped by other physically coupled information. Herein, we leverage this coupling through a multi-task learning (MTL) framework that jointly learns fault conditions and the related physical variables. Among MTL architectures, crosstalk structures have distinct advantages because they allow for controlled information exchange between tasks through the cross-talk layer while preventing negative transfer, in contrast to shared trunk architectures that often mix incompatible features. We build on our previously introduced residual neural dimension reductor model, and extend its application to two benchmarks where physical coupling is prominent. The first benchmark is a drone fault dataset, in which machine type and maneuvering direction significantly alter the frequency components of measured signals even under the same nominal condition. By learning fault classification together with these physical attributes, the cross-talk architecture can better classify faults. The second benchmark dataset is the motor compound fault dataset. In this system, each fault component, inner race fault, outer race fault, misalignment, and unbalance is coupled to the other. For motor compound fault, we also test classification performance when we use single-channel data or multi-channel data as input to the classifier. Across both benchmarks, our residual neural dimension reductor, consistently outperformed single-task models, multi-class models that merge all label combinations, and shared trunk multi-task models.

Speculative decoding can substantially accelerate LLM inference, but realizing its benefits in practice is challenging due to evolving workloads and system-level constraints. We present TIDE (Temporal Incremental Draft Engine), a serving-engine-native framework that integrates online draft adaptation directly into high-performance LLM inference systems. TIDE reuses target model hidden states generated during inference as training signals, enabling zero-overhead draft adaptation without reloading the target model, and employs adaptive runtime control to activate speculation and training only when beneficial. TIDE exploits heterogeneous clusters by mapping decoupled inference and training to appropriate GPU classes. Across diverse real-world workloads, TIDE achieves up to 1.15x throughput improvement over static speculative decoding while reducing draft training time by 1.67x compared to approaches that recompute training signals.

Machine learning systems are often trained and evaluated for fairness on historical data, yet deployed in environments where conditions have shifted. A particularly common form of shift occurs when the prevalence of positive outcomes changes differently across demographic groups--for example, when disease rates rise faster in one population than another, or when economic conditions affect loan default rates unequally. We study group-conditional prior probability shift (GPPS), where the label prevalence $P(Y=1\mid A=a)$ may change between training and deployment while the feature-generation process $P(X\mid Y,A)$ remains stable. Our analysis yields three main contributions. First, we prove a fundamental dichotomy: fairness criteria based on error rates (equalized odds) are structurally invariant under GPPS, while acceptance-rate criteria (demographic parity) can drift--and we prove this drift is unavoidable for non-trivial classifiers (shift-robust impossibility). Second, we show that target-domain risk and fairness metrics are identifiable without target labels: the invariance of ROC quantities under GPPS enables consistent estimation from source labels and unlabeled target data alone, with finite-sample guarantees. Third, we propose TAP-GPPS, a label-free post-processing algorithm that estimates prevalences from unlabeled data, corrects posteriors, and selects thresholds to satisfy demographic parity in the target domain. Experiments validate our theoretical predictions and demonstrate that TAP-GPPS achieves target fairness with minimal utility loss.

Retrieval augmented generation (RAG) has enhanced large language models by enabling access to external knowledge, with graph-based RAG emerging as a powerful paradigm for structured retrieval and reasoning. However, existing graph-based methods often over-rely on surface-level node matching and lack explicit causal modeling, leading to unfaithful or spurious answers. Prior attempts to incorporate causality are typically limited to local or single-document contexts and also suffer from information isolation that arises from modular graph structures, which hinders scalability and cross-module causal reasoning. To address these challenges, we propose HugRAG, a framework that rethinks knowledge organization for graph-based RAG through causal gating across hierarchical modules. HugRAG explicitly models causal relationships to suppress spurious correlations while enabling scalable reasoning over large-scale knowledge graphs. Extensive experiments demonstrate that HugRAG consistently outperforms competitive graph-based RAG baselines across multiple datasets and evaluation metrics. Our work establishes a principled foundation for structured, scalable, and causally grounded RAG systems.

The multi-armed bandit problem is a core framework for sequential decision-making under uncertainty, but classical algorithms often fail in environments with hidden, time-varying states that confound reward estimation and optimal action selection. We address key challenges arising from unobserved confounders, such as biased reward estimates and limited state information, by introducing a family of state-model-free bandit algorithms that leverage lagged contextual features and coordinated probing strategies. These implicitly track latent states and disambiguate state-dependent reward patterns. Our methods and their adaptive variants can learn optimal policies without explicit state modeling, combining computational efficiency with robust adaptation to non-stationary rewards. Empirical results across diverse settings demonstrate superior performance over classical approaches, and we provide practical recommendations for algorithm selection in real-world applications.

Is the standard weight decay in AdamW truly optimal? Although AdamW decouples weight decay from adaptive gradient scaling, a fundamental conflict remains: the Radial Tug-of-War. In deep learning, gradients tend to increase parameter norms to expand effective capacity while steering directions to learn features, whereas weight decay indiscriminately suppresses norm growth. This push--pull interaction induces radial oscillations, injecting noise into Adam's second-moment estimates and potentially degrading delicate tangential feature learning. We argue that magnitude and direction play distinct roles and should be decoupled in optimizer dynamics. We propose Orthogonal Dynamics Decoupling and instantiate it as AdamO: an SGD-style update handles the one-dimensional norm control, while Adam's adaptive preconditioning is confined to the tangential subspace. AdamO further incorporates curvature-adaptive radial step sizing and architecture-aware rules and projections for scale-invariant layers and low-dimensional parameters. Experiments on vision and language tasks show that AdamO improves generalization and stability over AdamW without introducing additional complex constraints.

Real-world machine learning on tabular data relies on complex data preparation pipelines for prediction, data integration, augmentation, and debugging. Designing these pipelines requires substantial domain expertise and engineering effort, motivating the question of how large language models (LLMs) can support tabular ML through code synthesis. We introduce SemPipes, a novel declarative programming model that integrates LLM-powered semantic data operators into tabular ML pipelines. Semantic operators specify data transformations in natural language while delegating execution to a runtime system. During training, SemPipes synthesizes custom operator implementations based on data characteristics, operator instructions, and pipeline context. This design enables the automatic optimization of data operations in a pipeline via LLM-based code synthesis guided by evolutionary search. We evaluate SemPipes across diverse tabular ML tasks and show that semantic operators substantially improve end-to-end predictive performance for both expert-designed and agent-generated pipelines, while reducing pipeline complexity. We implement SemPipes in Python and release it at https://github.com/deem-data/sempipes/tree/v1.

Traffic prediction in data-scarce, cross-city settings is challenging due to complex nonlinear dynamics and domain shifts. Existing methods often fail to capture traffic's inherent chaotic nature for effective few-shot learning. We propose CAST-CKT, a novel Chaos-Aware Spatio-Temporal and Cross-City Knowledge Transfer framework. It employs an efficient chaotic analyser to quantify traffic predictability regimes, driving several key innovations: chaos-aware attention for regime-adaptive temporal modelling; adaptive topology learning for dynamic spatial dependencies; and chaotic consistency-based cross-city alignment for knowledge transfer. The framework also provides horizon-specific predictions with uncertainty quantification. Theoretical analysis shows improved generalisation bounds. Extensive experiments on four benchmarks in cross-city few-shot settings show CAST-CKT outperforms state-of-the-art methods by significant margins in MAE and RMSE, while offering interpretable regime analysis. Code is available at https://github.com/afofanah/CAST-CKT.

Scientific workflows have become essential for orchestrating complex computational processes across distributed resources, managing large datasets, and ensuring reproducibility in modern research. The Workflows Community Summit 2025, held in Amsterdam on June 6th, 2025, convened international experts to examine emerging challenges and opportunities in this domain. Participants identified key barriers to workflow adoption, including tensions between system generality and domain-specific utility, concerns over long-term sustainability of workflow systems and services, insufficient recognition for those who develop and maintain reproducible workflows, and gaps in standardization, funding, training, and cross-disciplinary collaboration. To address these challenges, the summit proposed action lines spanning technology, policy, and community dimensions: shifting evaluation metrics from raw computational performance toward measuring genuine scientific impact; formalizing workflow patterns and community-driven benchmarks to improve transparency, reproducibility, and usability; cultivating a cohesive international workflows community that engages funding bodies and research stakeholders; and investing in human capital through dedicated workflow engineering roles, career pathways, and integration of workflow concepts into educational curricula and long-term training initiatives. This document presents the summit's findings, beginning with an overview of the current computing ecosystem and the rationale for workflow-centric approaches, followed by a discussion of identified challenges and recommended action lines for advancing scientific discovery through workflows.

Recently, rubrics have been used to guide LLM judges in capturing subjective, nuanced, multi-dimensional human preferences, and have been extended from evaluation to reward signals for reinforcement fine-tuning (RFT). However, rubric generation remains hard to control: rubrics often lack coverage, conflate dimensions, misalign preference direction, and contain redundant or highly correlated criteria, degrading judge accuracy and producing suboptimal rewards during RFT. We propose RRD, a principled framework for rubric refinement built on a recursive decompose-filter cycle. RRD decomposes coarse rubrics into fine-grained, discriminative criteria, expanding coverage while sharpening separation between responses. A complementary filtering mechanism removes misaligned and redundant rubrics, and a correlation-aware weighting scheme prevents over-representing highly correlated criteria, yielding rubric sets that are informative, comprehensive, and non-redundant. Empirically, RRD delivers large, consistent gains across both evaluation and training: it improves preference-judgment accuracy on JudgeBench and PPE for both GPT-4o and Llama3.1-405B judges, achieving top performance in all settings with up to +17.7 points on JudgeBench. When used as the reward source for RFT on WildChat, it yields substantially stronger and more stable learning signals, boosting reward by up to 160% (Qwen3-4B) and 60% (Llama3.1-8B) versus 10-20% for prior rubric baselines, with gains that transfer to HealthBench-Hard and BiGGen Bench. Overall, RRD establishes recursive rubric refinement as a scalable and interpretable foundation for LLM judging and reward modeling in open-ended domains.

Exceptions allow developers to handle error cases expected to occur infrequently. Ideally, good test suites should test both normal and exceptional behaviors to catch more bugs and avoid regressions. While current research analyzes exceptions that propagate to tests, it does not explore other exceptions that do not reach the tests. In this paper, we provide an empirical study to explore how frequently exceptional behaviors are tested in real-world systems. We consider both exceptions that propagate to tests and the ones that do not reach the tests. For this purpose, we run an instrumented version of test suites, monitor their execution, and collect information about the exceptions raised at runtime. We analyze the test suites of 25 Python systems, covering 5,372 executed methods, 17.9M calls, and 1.4M raised exceptions. We find that 21.4% of the executed methods do raise exceptions at runtime. In methods that raise exceptions, on the median, 1 in 10 calls exercise exceptional behaviors. Close to 80% of the methods that raise exceptions do so infrequently, but about 20% raise exceptions more frequently. Finally, we provide implications for researchers and practitioners. We suggest developing novel tools to support exercising exceptional behaviors and refactoring expensive try/except blocks. We also call attention to the fact that exception-raising behaviors are not necessarily "abnormal" or rare.

Unittest and pytest are the most popular testing frameworks in Python. Overall, pytest provides some advantages, including simpler assertion, reuse of fixtures, and interoperability. Due to such benefits, multiple projects in the Python ecosystem have migrated from unittest to pytest. To facilitate the migration, pytest can also run unittest tests, thus, the migration can happen gradually over time. However, the migration can be time-consuming and take a long time to conclude. In this context, projects would benefit from automated solutions to support the migration process. In this paper, we propose TestMigrationsInPy, a dataset of test migrations from unittest to pytest. TestMigrationsInPy contains 923 real-world migrations performed by developers. Future research proposing novel solutions to migrate frameworks in Python can rely on TestMigrationsInPy as a ground truth. Moreover, as TestMigrationsInPy includes information about the migration type (e.g., changes in assertions or fixtures), our dataset enables novel solutions to be verified effectively, for instance, from simpler assertion migrations to more complex fixture migrations. TestMigrationsInPy is publicly available at: https://github.com/altinoalvesjunior/TestMigrationsInPy.

The proliferation of agentic systems has thrust the reasoning capabilities of AI into the forefront of contemporary machine learning. While it is known that there \emph{exist} neural networks which can reason through any Boolean task $f:\{0,1\}^B\to\{0,1\}$, in the sense that they emulate Boolean circuits with fan-in $2$ and fan-out $1$ gates, trained models have been repeatedly demonstrated to fall short of these theoretical ideals. This raises the question: \textit{Can one exhibit a deep learning model which \textbf{certifiably} always reasons and can \textbf{universally} reason through any Boolean task?} Moreover, such a model should ideally require few parameters to solve simple Boolean tasks. We answer this question affirmatively by exhibiting a deep learning architecture which parameterizes distributions over Boolean circuits with the guarantee that, for every parameter configuration, a sample is almost surely a valid Boolean circuit (and hence admits an intrinsic circuit-level certificate). We then prove a universality theorem: for any Boolean $f:\{0,1\}^B\to\{0,1\}$, there exists a parameter configuration under which the sampled circuit computes $f$ with arbitrarily high probability. When $f$ is an $\mathcal{O}(\log B)$-junta, the required number of parameters scales linearly with the input dimension $B$. Empirically, on a controlled truth-table completion benchmark aligned with our setting, the proposed architecture trains reliably and achieves high exact-match accuracy while preserving the predicted structure: every internal unit is Boolean-valued on $\{0,1\}^B$. Matched MLP baselines reach comparable accuracy, but only about $10\%$ of hidden units admit a Boolean representation; i.e.\ are two-valued over the Boolean cube.

In a probabilistic reformulation of a combinatorial problem, we often face an optimization over a hypercube, which corresponds to the Bernoulli probability parameter for each binary variable in the primal problem. The combinatorial nature suggests that an exact gradient computation requires multiple queries. We propose a stochastic gradient that is unbiased and requires only a single query of the combinatorial function. This method encompasses a well-established REINFORCE (through an importance sampling), as well as including a class of new stochastic gradients.

Large language models (LLMs) are increasingly evaluated in interactive environments to test their social intelligence. However, existing benchmarks often assume idealized communication between agents, limiting our ability to diagnose whether LLMs can maintain and repair interactions in more realistic, imperfect settings. To close this gap, we present \textsc{SocialVeil}, a social learning environment that can simulate social interaction under cognitive-difference-induced communication barriers. Grounded in a systematic literature review of communication challenges in human interaction, \textsc{SocialVeil} introduces three representative types of such disruption, \emph{semantic vagueness}, \emph{sociocultural mismatch}, and \emph{emotional interference}. We also introduce two barrier-aware evaluation metrics, \emph{unresolved confusion} and \emph{mutual understanding}, to evaluate interaction quality under impaired communication. Experiments across 720 scenarios and four frontier LLMs show that barriers consistently impair performance, with mutual understanding reduced by over 45\% on average, and confusion elevated by nearly 50\%. Human evaluations validate the fidelity of these simulated barriers (ICC$\approx$0.78, Pearson r$\approx$0.80). We further demonstrate that adaptation strategies (Repair Instruction and Interactive learning) only have a modest effect far from barrier-free performance. This work takes a step toward bringing social interaction environments closer to real-world communication, opening opportunities for exploring the social intelligence of LLM agents.

Whose values should AI systems learn? Preference based alignment methods like RLHF derive their training signal from human raters, yet these rater pools are typically convenience samples that systematically over represent some demographics and under represent others. We introduce Democratic Preference Optimization, or DemPO, a framework that applies algorithmic sortition, the same mechanism used to construct citizen assemblies, to preference based fine tuning. DemPO offers two training schemes. Hard Panel trains exclusively on preferences from a quota satisfying mini public sampled via sortition. Soft Panel retains all data but reweights each rater by their inclusion probability under the sortition lottery. We prove that Soft Panel weighting recovers the expected Hard Panel objective in closed form. Using a public preference dataset that pairs human judgments with rater demographics and a seventy five clause constitution independently elicited from a representative United States panel, we evaluate Llama models from one billion to eight billion parameters fine tuned under each scheme. Across six aggregation methods, the Hard Panel consistently ranks first and the Soft Panel consistently outperforms the unweighted baseline, with effect sizes growing as model capacity increases. These results demonstrate that enforcing demographic representativeness at the preference collection stage, rather than post hoc correction, yields models whose behavior better reflects values elicited from representative publics.

Large Language Models (LLMs) are increasingly used as evaluators of reasoning quality, yet their reliability and bias in payments-risk settings remain poorly understood. We introduce a structured multi-evaluator framework for assessing LLM reasoning in Merchant Category Code (MCC)-based merchant risk assessment, combining a five-criterion rubric with Monte-Carlo scoring to evaluate rationale quality and evaluator stability. Five frontier LLMs generate and cross-evaluate MCC risk rationales under attributed and anonymized conditions. To establish a judge-independent reference, we introduce a consensus-deviation metric that eliminates circularity by comparing each judge's score to the mean of all other judges, yielding a theoretically grounded measure of self-evaluation and cross-model deviation. Results reveal substantial heterogeneity: GPT-5.1 and Claude 4.5 Sonnet show negative self-evaluation bias (-0.33, -0.31), while Gemini-2.5 Pro and Grok 4 display positive bias (+0.77, +0.71), with bias attenuating by 25.8 percent under anonymization. Evaluation by 26 payment-industry experts shows LLM judges assign scores averaging +0.46 points above human consensus, and that the negative bias of GPT-5.1 and Claude 4.5 Sonnet reflects closer alignment with human judgment. Ground-truth validation using payment-network data shows four models exhibit statistically significant alignment (Spearman rho = 0.56 to 0.77), confirming that the framework captures genuine quality. Overall, the framework provides a replicable basis for evaluating LLM-as-a-judge systems in payment-risk workflows and highlights the need for bias-aware protocols in operational financial settings.

Implicit discourse relation classification is a challenging task, as it requires inferring meaning from context. While contextual cues can be distributed across modalities and vary across languages, they are not always captured by text alone. To address this, we introduce an automatic method for distantly related and unrelated language pairs to construct a multilingual and multimodal dataset for implicit discourse relations in English, French, and Spanish. For classification, we propose a multimodal approach that integrates textual and acoustic information through Qwen2-Audio, allowing joint modeling of text and audio for implicit discourse relation classification across languages. We find that while text-based models outperform audio-based models, integrating both modalities can enhance performance, and cross-lingual transfer can provide substantial improvements for low-resource languages.

Scarcity of labeled training data remains the long pole in the tent for building performant language technology and generative AI models. Transformer models -- particularly LLMs -- are increasingly being used to mitigate the data scarcity problem via synthetic data generation. However, because the models are black boxes, the properties of the synthetic data are difficult to predict. In practice it is common for language technology engineers to 'fiddle' with the LLM temperature setting and hope that what comes out the other end improves the downstream model. Faced with this uncertainty, here we propose Data Kernel Perspective Space (DKPS) to provide the foundation for mathematical analysis yielding concrete statistical guarantees for the quality of the outputs of transformer models. We first show the mathematical derivation of DKPS and how it provides performance guarantees. Next we show how DKPS performance guarantees can elucidate performance of a downstream task, such as neural machine translation models or LLMs trained using Contrastive Preference Optimization (CPO). Limitations of the current work and future research are also discussed.

As intelligent systems and multi-agent coordination become increasingly central to real-world applications, there is a growing need for simulation tools that are both scalable and accessible. Existing high-fidelity simulators, while powerful, are often computationally expensive and ill-suited for rapid prototyping or large-scale agent deployments. We present GAMMS (Graph based Adversarial Multiagent Modeling Simulator), a lightweight yet extensible simulation framework designed to support fast development and evaluation of agent behavior in environments that can be represented as graphs. GAMMS emphasizes five core objectives: scalability, ease of use, integration-first architecture, fast visualization feedback, and real-world grounding. It enables efficient simulation of complex domains such as urban road networks and communication systems, supports integration with external tools (e.g., machine learning libraries, planning solvers), and provides built-in visualization with minimal configuration. GAMMS is agnostic to policy type, supporting heuristic, optimization-based, and learning-based agents, including those using large language models. By lowering the barrier to entry for researchers and enabling high-performance simulations on standard hardware, GAMMS facilitates experimentation and innovation in multi-agent systems, autonomous planning, and adversarial modeling. The framework is open-source and available at https://github.com/GAMMSim/GAMMS/

Ensuring the reliability of machine learning models in safety-critical domains such as healthcare requires auditing methods that can uncover model shortcomings. We introduce a method for identifying important visual concepts within large multimodal models (LMMs) and use it to investigate the behaviors these models exhibit when prompted with medical tasks. We primarily focus on the task of classifying malignant skin lesions from clinical dermatology images, with supplemental experiments including both chest radiographs and natural images. After showing how LMMs display unexpected gaps in performance between different demographic subgroups when prompted with demonstrating examples, we apply our method, Visual Concept Ranking (VCR), to these models and prompts. VCR generates hypotheses related to different visual feature dependencies, which we are then able to validate with manual interventions.

Autonomous mission planning for Active Debris Removal (ADR) must balance efficiency, adaptability, and strict feasibility constraints on fuel and mission duration. This work compares three planners for the constrained multi-debris rendezvous problem in Low Earth Orbit: a nominal Masked Proximal Policy Optimization (PPO) policy trained under fixed mission parameters, a domain-randomized Masked PPO policy trained across varying mission constraints for improved robustness, and a plain Monte Carlo Tree Search (MCTS) baseline. Evaluations are conducted in a high-fidelity orbital simulation with refueling, realistic transfer dynamics, and randomized debris fields across 300 test cases in nominal, reduced fuel, and reduced mission time scenarios. Results show that nominal PPO achieves top performance when conditions match training but degrades sharply under distributional shift, while domain-randomized PPO exhibits improved adaptability with only moderate loss in nominal performance. MCTS consistently handles constraint changes best due to online replanning but incurs orders-of-magnitude higher computation time. The findings underline a trade-off between the speed of learned policies and the adaptability of search-based methods, and suggest that combining training-time diversity with online planning could be a promising path for future resilient ADR mission planners.

Simulated environments are a key piece in the success of Reinforcement Learning (RL), allowing practitioners and researchers to train decision making agents without running expensive experiments on real hardware. Simulators remain a security blind spot, however, enabling adversarial developers to alter the dynamics of their released simulators for malicious purposes. Therefore, in this work we highlight a novel threat, demonstrating how simulator dynamics can be exploited to stealthily implant action-level backdoors into RL agents. The backdoor then allows an adversary to reliably activate targeted actions in an agent upon observing a predefined ``trigger'', leading to potentially dangerous consequences. Traditional backdoor attacks are limited in their strong threat models, assuming the adversary has near full control over an agent's training pipeline, enabling them to both alter and observe agent's rewards. As these assumptions are infeasible to implement within a simulator, we propose a new attack ``Daze'' which is able to reliably and stealthily implant backdoors into RL agents trained for real world tasks without altering or even observing their rewards. We provide formal proof of Daze's effectiveness in guaranteeing attack success across general RL tasks along with extensive empirical evaluations on both discrete and continuous action space domains. We additionally provide the first example of RL backdoor attacks transferring to real, robotic hardware. These developments motivate further research into securing all components of the RL training pipeline to prevent malicious attacks.

Millions now use leading generative AI chatbots for psychological support. Despite the promise related to availability and scale, the single most pressing question in AI for mental health is whether these tools are safe. The Validation of Ethical and Responsible AI in Mental Health (VERA-MH) evaluation was recently proposed to meet the urgent need for an evidence-based automated safety benchmark. This study aimed to examine the clinical validity and reliability of the VERA-MH evaluation for AI safety in suicide risk detection and response. We first simulated a large set of conversations between large language model (LLM)-based users (user-agents) and general-purpose AI chatbots. Licensed mental health clinicians used a rubric (scoring guide) to independently rate the simulated conversations for safe and unsafe chatbot behaviors, as well as user-agent realism. An LLM-based judge used the same scoring rubric to evaluate the same set of simulated conversations. We then compared rating alignment across (a) individual clinicians and (b) clinician consensus and the LLM judge, and (c) examined clinicians' ratings of user-agent realism. Individual clinicians were generally consistent with one another in their safety ratings (chance-corrected inter-rater reliability [IRR]: 0.77), thus establishing a gold-standard clinical reference. The LLM judge was strongly aligned with this clinical consensus (IRR: 0.81) overall and within key conditions. Clinician raters generally perceived the user-agents to be realistic. For the potential mental health benefits of AI chatbots to be realized, attention to safety is paramount. Findings from this human evaluation study support the clinical validity and reliability of VERA-MH: an open-source, fully automated AI safety evaluation for mental health. Further research will address VERA-MH generalizability and robustness.

Systematic literature reviews (SLRs) are fundamental to evidence-based research, but manual screening is an increasing bottleneck as scientific output grows. Screening features low prevalence of relevant studies and scarce, costly expert decisions. Traditional active learning (AL) systems help, yet typically rely on fixed query strategies for selecting the next unlabeled documents. These static strategies do not adapt over time and ignore the relational structure of scientific literature networks. This thesis introduces AutoDiscover, a framework that reframes AL as an online decision-making problem driven by an adaptive agent. Literature is modeled as a heterogeneous graph capturing relationships among documents, authors, and metadata. A Heterogeneous Graph Attention Network (HAN) learns node representations, which a Discounted Thompson Sampling (DTS) agent uses to dynamically manage a portfolio of query strategies. With real-time human-in-the-loop labels, the agent balances exploration and exploitation under non-stationary review dynamics, where strategy utility changes over time. On the 26-dataset SYNERGY benchmark, AutoDiscover achieves higher screening efficiency than static AL baselines. Crucially, the agent mitigates cold start by bootstrapping discovery from minimal initial labels where static approaches fail. We also introduce TS-Insight, an open-source visual analytics dashboard to interpret, verify, and diagnose the agent's decisions. Together, these contributions accelerate SLR screening under scarce expert labels and low prevalence of relevant studies.

In this paper, we aim to bridge test-time-training with a new type of parametric memory that can be flexibly offloaded from or merged into model parameters. We present Locas, a Locally-Supported parametric memory that shares the design of FFN blocks in modern transformers, allowing it to be flexibly permanentized into the model parameters while supporting efficient continual learning. We discuss two major variants of Locas: one with a conventional two-layer MLP design that has a clearer theoretical guarantee; the other one shares the same GLU-FFN structure with SOTA LLMs, and can be easily attached to existing models for both parameter-efficient and computation-efficient continual learning. Crucially, we show that proper initialization of such low-rank sideway-FFN-style memories -- performed in a principled way by reusing model parameters, activations and/or gradients -- is essential for fast convergence, improved generalization, and catastrophic forgetting prevention. We validate the proposed memory mechanism on the PG-19 whole-book language modeling and LoCoMo long-context dialogue question answering tasks. With only 0.02\% additional parameters in the lowest case, Locas-GLU is capable of storing the information from past context while maintaining a much smaller context window. In addition, we also test the model's general capability loss after memorizing the whole book with Locas, through comparative MMLU evaluation. Results show the promising ability of Locas to permanentize past context into parametric knowledge with minimized catastrophic forgetting of the model's existing internal knowledge.

Strategic classification, where individuals modify their features to influence machine learning (ML) decisions, presents critical fairness challenges. While group fairness in this setting has been widely studied, individual fairness remains underexplored. We analyze threshold-based classifiers and prove that deterministic thresholds violate individual fairness. Then, we investigate the possibility of using a randomized classifier to achieve individual fairness. We introduce conditions under which a randomized classifier ensures individual fairness and leverage these conditions to find an optimal and individually fair randomized classifier through a linear programming problem. Additionally, we demonstrate that our approach can be extended to group fairness notions. Experiments on real-world datasets confirm that our method effectively mitigates unfairness and improves the fairness-accuracy trade-off.

Atmospheric blocking events drive persistent weather extremes in midlatitudes, but isolating the influence of sea surface temperature (SST) from chaotic internal atmospheric variability on these events remains a challenge. We address this challenge using century-long (1900-2010), large-ensemble simulations with two computationally efficient deep-learning general circulation models. We find these models skillfully reproduce the observed blocking climatology, matching or exceeding the performance of a traditional high-resolution model and representative CMIP6 models. Averaging the large ensembles filters internal atmospheric noise to isolate the SST-forced component of blocking variability, yielding substantially higher correlations with reanalysis than for individual ensemble members. We identify robust teleconnections linking Greenland blocking frequency to North Atlantic SST and El Niño-like patterns. Furthermore, SST-forced trends in blocking frequency show a consistent decline in winter over Greenland, and an increase over Europe. These results demonstrate that SST variability exerts a significant and physically interpretable influence on blocking frequency and establishes large ensembles from deep learning models as a powerful tool for separating forced SST signals from internal noise.

In recent years, explaining decisions made by complex machine learning models has become essential in high-stakes domains such as energy systems, healthcare, finance, and autonomous systems. However, the reliability of these explanations, namely, whether they remain stable and consistent under realistic, non-adversarial changes, remains largely unmeasured. Widely used methods such as SHAP and Integrated Gradients (IG) are well-motivated by axiomatic notions of attribution, yet their explanations can vary substantially even under system-level conditions, including small input perturbations, correlated representations, and minor model updates. Such variability undermines explanation reliability, as reliable explanations should remain consistent across equivalent input representations and small, performance-preserving model changes. We introduce the Explanation Reliability Index (ERI), a family of metrics that quantifies explanation stability under four reliability axioms: robustness to small input perturbations, consistency under feature redundancy, smoothness across model evolution, and resilience to mild distributional shifts. For each axiom, we derive formal guarantees, including Lipschitz-type bounds and temporal stability results. We further propose ERI-T, a dedicated measure of temporal reliability for sequential models, and introduce ERI-Bench, a benchmark designed to systematically stress-test explanation reliability across synthetic and real-world datasets. Experimental results reveal widespread reliability failures in popular explanation methods, showing that explanations can be unstable under realistic deployment conditions. By exposing and quantifying these instabilities, ERI enables principled assessment of explanation reliability and supports more trustworthy explainable AI (XAI) systems.

The problem with existing camera-based Deep Reinforcement Learning approaches is twofold: they rarely integrate high-level scene context into the feature representation, and they rely on rigid, fixed reward functions. To address these challenges, this paper proposes a novel pipeline that produces a neuro-symbolic feature representation that encompasses semantic, spatial, and shape information, as well as spatially boosted features of dynamic entities in the scene, with an emphasis on safety-critical road users. It also proposes a Soft First-Order Logic (SFOL) reward function that balances human values via a symbolic reasoning module. Here, semantic and spatial predicates are extracted from segmentation maps and applied to linguistic rules to obtain reward weights. Quantitative experiments in the CARLA simulation environment show that the proposed neuro-symbolic representation and SFOL reward function improved policy robustness and safety-related performance metrics compared to baseline representations and reward formulations across varying traffic densities and occlusion levels. The findings demonstrate that integrating holistic representations and soft reasoning into Reinforcement Learning can support more context-aware and value-aligned decision-making for autonomous driving.

Reliance on images for dietary assessment is an important strategy to accurately and conveniently monitor an individual's health, making it a vital mechanism in the prevention and care of chronic diseases and obesity. However, image-based dietary assessment suffers from estimating the three dimensional size of food from 2D image inputs. Many strategies have been devised to overcome this critical limitation such as the use of auxiliary inputs like depth maps, multi-view inputs, or model-based approaches such as template matching. Deep learning also helps bridge the gap by either using monocular images or combinations of the image and the auxillary inputs to precisely predict the output portion from the image input. In this paper, we explore the different strategies employed for accurate portion estimation.

As the orbital environment around Earth becomes increasingly crowded with debris, active debris removal (ADR) missions face significant challenges in ensuring safe operations while minimizing the risk of in-orbit collisions. This study presents a reinforcement learning (RL) based framework to enhance adaptive collision avoidance in ADR missions, specifically for multi-debris removal using small satellites. Small satellites are increasingly adopted due to their flexibility, cost effectiveness, and maneuverability, making them well suited for dynamic missions such as ADR. Building on existing work in multi-debris rendezvous, the framework integrates refueling strategies, efficient mission planning, and adaptive collision avoidance to optimize spacecraft rendezvous operations. The proposed approach employs a masked Proximal Policy Optimization (PPO) algorithm, enabling the RL agent to dynamically adjust maneuvers in response to real-time orbital conditions. Key considerations include fuel efficiency, avoidance of active collision zones, and optimization of dynamic orbital parameters. The RL agent learns to determine efficient sequences for rendezvousing with multiple debris targets, optimizing fuel usage and mission time while incorporating necessary refueling stops. Simulated ADR scenarios derived from the Iridium 33 debris dataset are used for evaluation, covering diverse orbital configurations and debris distributions to demonstrate robustness and adaptability. Results show that the proposed RL framework reduces collision risk while improving mission efficiency compared to traditional heuristic approaches. This work provides a scalable solution for planning complex multi-debris ADR missions and is applicable to other multi-target rendezvous problems in autonomous space mission planning.

Uncertainty quantification (UQ) for large language models (LLMs) is a key building block for safety guardrails of daily LLM applications. Yet, even as LLM agents are increasingly deployed in highly complex tasks, most UQ research still centers on single-turn question-answering. We argue that UQ research must shift to realistic settings with interactive agents, and that a new principled framework for agent UQ is needed. This paper presents the first general formulation of agent UQ that subsumes broad classes of existing UQ setups. Under this formulation, we show that prior works implicitly treat LLM UQ as an uncertainty accumulation process, a viewpoint that breaks down for interactive agents in an open world. In contrast, we propose a novel perspective, a conditional uncertainty reduction process, that explicitly models reducible uncertainty over an agent's trajectory by highlighting "interactivity" of actions. From this perspective, we outline a conceptual framework to provide actionable guidance for designing UQ in LLM agent setups. Finally, we conclude with practical implications of the agent UQ in frontier LLM development and domain-specific applications, as well as open remaining problems.

Neural networks achieve strong empirical performance, but robustness concerns still hinder deployment in safety-critical applications. Formal verification provides robustness guarantees, but current methods face a scalability-completeness trade-off. We propose a hybrid verifier in a branch-and-bound (BaB) framework that efficiently tightens both upper and lower bounds until an $ε-$global optimum is reached or early stop is triggered. The key is an exact nonlinear program with complementarity constraints (NLP-CC) for upper bounding that preserves the ReLU input-output graph, so any feasible solution yields a valid counterexample and enables rapid pruning of unsafe subproblems. We further accelerate verification with (i) warm-started NLP solves requiring minimal constraint-matrix updates and (ii) pattern-aligned strong branching that prioritizes splits most effective at tightening relaxations. We also provide conditions under which NLP-CC upper bounds are tight. Experiments on MNIST and CIFAR-10 show markedly tighter upper bounds than PGD across perturbation radii spanning up to three orders of magnitude, fast per-node solves in practice, and substantial end-to-end speedups over MIP-based verification, amplified by warm-starting, GPU batching, and pattern-aligned branching.

As AI agents automate critical workloads, they remain vulnerable to indirect prompt injection (IPI) attacks. Current defenses rely on monitoring protocols that jointly evaluate an agent's Chain-of-Thought (CoT) and tool-use actions to ensure alignment with user intent. We demonstrate that these monitoring-based defenses can be bypassed via a novel Agent-as-a-Proxy attack, where prompt injection attacks treat the agent as a delivery mechanism, bypassing both agent and monitor simultaneously. While prior work on scalable oversight has focused on whether small monitors can supervise large agents, we show that even frontier-scale monitors are vulnerable. Large-scale monitoring models like Qwen2.5-72B can be bypassed by agents with similar capabilities, such as GPT-4o mini and Llama-3.1-70B. On the AgentDojo benchmark, we achieve a high attack success rate against AlignmentCheck and Extract-and-Evaluate monitors under diverse monitoring LLMs. Our findings suggest current monitoring-based agentic defenses are fundamentally fragile regardless of model scale.

A large body of theory and empirical work hypothesizes a connection between the flatness of a neural network's loss landscape during training and its performance. However, there have been conceptually opposite pieces of evidence regarding when SGD prefers flatter or sharper solutions during training. In this work, we partially but causally clarify the flatness-seeking behavior of SGD by identifying and exactly solving an analytically solvable model that exhibits both flattening and sharpening behavior during training. In this model, the SGD training has no \textit{a priori} preference for flatness, but only a preference for minimal gradient fluctuations. This leads to the insight that, at least within this model, it is data distribution that uniquely determines the sharpness at convergence, and that a flat minimum is preferred if and only if the noise in the labels is isotropic across all output dimensions. When the noise in the labels is anisotropic, the model instead prefers sharpness and can converge to an arbitrarily sharp solution, depending on the imbalance in the noise in the labels spectrum. We reproduce this key insight in controlled settings with different model architectures such as MLP, RNN, and transformers.

Dense retrieval, which encodes queries and documents into a single dense vector, has become the dominant neural retrieval approach due to its simplicity and compatibility with fast approximate nearest neighbor algorithms. As the tasks dense retrieval performs grow in complexity, the fundamental limitations of the underlying data structure and similarity metric -- namely vectors and inner-products -- become more apparent. Prior recent work has shown theoretical limitations inherent to single vectors and inner-products that are generally tied to the embedding dimension. Given the importance of embedding dimension for retrieval capacity, understanding how dense retrieval performance changes as embedding dimension is scaled is fundamental to building next generation retrieval models that balance effectiveness and efficiency. In this work, we conduct a comprehensive analysis of the relationship between embedding dimension and retrieval performance. Our experiments include two model families and a range of model sizes from each to construct a detailed picture of embedding scaling behavior. We find that the scaling behavior fits a power law, allowing us to derive scaling laws for performance given only embedding dimension, as well as a joint law accounting for embedding dimension and model size. Our analysis shows that for evaluation tasks aligned with the training task, performance continues to improve as embedding size increases, though with diminishing returns. For evaluation data that is less aligned with the training task, we find that performance is less predictable, with performance degrading with larger embedding dimensions for certain tasks. We hope our work provides additional insight into the limitations of embeddings and their behavior as well as offers a practical guide for selecting model and embedding dimension to achieve optimal performance with reduced storage and compute costs.

Cybergrooming is an evolving threat to youth, necessitating proactive educational interventions. We propose StagePilot, an offline RL-based dialogue agent that simulates the stage-wise progression of grooming behaviors for prevention training. StagePilot selects conversational stages using a composite reward that balances user sentiment and goal proximity, with transitions constrained to adjacent stages for realism and interpretability. We evaluate StagePilot through LLM-based simulations, measuring stage completion, dialogue efficiency, and emotional engagement. Results show that StagePilot generates realistic and coherent conversations aligned with grooming dynamics. Among tested methods, the IQL+AWAC agent achieves the best balance between strategic planning and emotional coherence, reaching the final stage up to 43% more frequently than baselines while maintaining over 70% sentiment alignment.

Large Language Models are increasingly used by students to explore advanced material in computer science, including graph theory. As these tools become integrated into undergraduate and graduate coursework, it is important to understand how reliably they support mathematically rigorous thinking. This study examines the performance of a LLM on two related graph theoretic problems: a solved problem concerning the gracefulness of line graphs and an open problem for which no solution is currently known. We use an eight stage evaluation protocol that reflects authentic mathematical inquiry, including interpretation, exploration, strategy formation, and proof construction. The model performed strongly on the solved problem, producing correct definitions, identifying relevant structures, recalling appropriate results without hallucination, and constructing a valid proof confirmed by a graph theory expert. For the open problem, the model generated coherent interpretations and plausible exploratory strategies but did not advance toward a solution. It did not fabricate results and instead acknowledged uncertainty, which is consistent with the explicit prompting instructions that directed the model to avoid inventing theorems or unsupported claims. These findings indicate that LLMs can support exploration of established material but remain limited in tasks requiring novel mathematical insight or critical structural reasoning. For computing education, this distinction highlights the importance of guiding students to use LLMs for conceptual exploration while relying on independent verification and rigorous argumentation for formal problem solving.

We revisit the problem of learning fermionic linear optics (FLO), also known as fermionic Gaussian unitaries. Given black-box query access to an unknown FLO, previous proposals required $\widetilde{\mathcal{O}}(n^5 / \varepsilon^2)$ queries, where $n$ is the system size and $\varepsilon$ is the error in diamond distance. These algorithms also use unphysical operations (i.e., violating fermionic superselection rules) and/or $n$ auxiliary modes to prepare Choi states of the FLO. In this work, we establish efficient and experimentally friendly protocols that obey superselection, use minimal ancilla (at most $1$ extra mode), and exhibit improved dependence on both parameters $n$ and $\varepsilon$. For arbitrary (active) FLOs this algorithm makes at most $\widetilde{\mathcal{O}}(n^4 / \varepsilon)$ queries, while for number-conserving (passive) FLOs we show that $\mathcal{O}(n^3 / \varepsilon)$ queries suffice. The complexity of the active case can be further reduced to $\widetilde{\mathcal{O}}(n^3 / \varepsilon)$ at the cost of using $n$ ancilla. This marks the first FLO learning algorithm that attains Heisenberg scaling in precision. As a side result, we also demonstrate an improved copy complexity of $\widetilde{\mathcal{O}}(n η^2 / \varepsilon^2)$ for time-efficient state tomography of $η$-particle Slater determinants in $\varepsilon$ trace distance, which may be of independent interest.

Online scams across email, short message services, and social media increasingly challenge everyday risk assessment, particularly as generative AI enables more fluent and context-aware deception. Although transformer-based detectors achieve strong predictive performance, their explanations are often opaque to non-experts or misaligned with model decisions. We propose VEXA, an evidence-grounded and persona-adaptive framework for generating learner-facing scam explanations by integrating GradientSHAP-based attribution with theory-informed vulnerability personas. Evaluation across multi-channel datasets shows that grounding explanations in detector-derived evidence improves semantic reliability without increasing linguistic complexity, while persona conditioning introduces interpretable stylistic variation without disrupting evidential alignment. These results reveal a key design insight: evidential grounding governs semantic correctness, whereas persona-based adaptation operates at the level of presentation under constraints of faithfulness. Together, VEXA demonstrates the feasibility of persona-adaptive, evidence-grounded explanations and provides design guidance for trustworthy, learner-facing security explanations in non-formal contexts.

Inclement weather conditions can significantly impact driver visibility and tire-road surface friction, requiring adjusted safe driving speeds to reduce crash risk. This study proposes a hybrid predictive framework that recommends real-time safe speed intervals for freeway travel under diverse weather conditions. Leveraging high-resolution Connected Vehicle (CV) data and Road Weather Information System (RWIS) data collected in Buffalo, NY, from 2022 to 2023, we construct a spatiotemporally aligned dataset containing over 6.6 million records across 73 days. The core model employs Quantile Regression Forests (QRF) to estimate vehicle speed distributions in 10-minute windows, using 26 input features that capture meteorological, pavement, and temporal conditions. To enforce safety constraints, a physics-based upper speed limit is computed for each interval based on real-time road grip and visibility, ensuring that vehicles can safely stop within their sight distance. The final recommended interval fuses QRF-predicted quantiles with both posted speed limits and the physics-derived upper bound. Experimental results demonstrate strong predictive performance: the QRF model achieves a mean absolute error of 1.55 mph, with 96.43% of median speed predictions within 5 mph, a PICP (50%) of 48.55%, and robust generalization across weather types. The model's ability to respond to changing weather conditions and generalize across road segments shows promise for real-world deployment, thereby improving traffic safety and reducing weather-related crashes.

We present a unified algorithmic framework for the numerical solution, constrained optimization, and physics-informed learning of PDEs with a variational structure. Our framework is based on a Galerkin discretization of the underlying variational forms, and its high efficiency stems from a novel highly-optimized and GPU-compliant TensorGalerkin framework for linear system assembly (stiffness matrices and load vectors). TensorGalerkin operates by tensorizing element-wise operations within a Python-level Map stage and then performs global reduction with a sparse matrix multiplication that performs message passing on the mesh-induced sparsity graph. It can be seamlessly employed downstream as i) a highly-efficient numerical PDEs solver, ii) an end-to-end differentiable framework for PDE-constrained optimization, and iii) a physics-informed operator learning algorithm for PDEs. With multiple benchmarks, including 2D and 3D elliptic, parabolic, and hyperbolic PDEs on unstructured meshes, we demonstrate that the proposed framework provides significant computational efficiency and accuracy gains over a variety of baselines in all the targeted downstream applications.

Offline reinforcement learning (RL) aims to learn the optimal policy from a fixed dataset generated by behavior policies without additional environment interactions. One common challenge that arises in this setting is the out-of-distribution (OOD) error, which occurs when the policy leaves the training distribution. Prior methods penalize a statistical distance term to keep the policy close to the behavior policy, but this constrains policy improvement and may not completely prevent OOD actions. Another challenge is that the optimal policy distribution can be multimodal and difficult to represent. Recent works apply diffusion or flow policies to address this problem, but it is unclear how to avoid OOD errors while retaining policy expressiveness. We propose ReFORM, an offline RL method based on flow policies that enforces the less restrictive support constraint by construction. ReFORM learns a behavior cloning (BC) flow policy with a bounded source distribution to capture the support of the action distribution, then optimizes a reflected flow that generates bounded noise for the BC flow while keeping the support, to maximize the performance. Across 40 challenging tasks from the OGBench benchmark with datasets of varying quality and using a constant set of hyperparameters for all tasks, ReFORM dominates all baselines with hand-tuned hyperparameters on the performance profile curves.

Vision-Language-Action (VLA) models have demonstrated strong performance across a wide range of robotic manipulation tasks. Despite the success, extending large pretrained Vision-Language Models (VLMs) to the action space can induce vision-action misalignment, where action predictions exhibit weak dependence on the current visual state, leading to unreliable action outputs. In this work, we study VLA models through the lens of visual conditioning and empirically show that successful rollouts consistently exhibit stronger visual dependence than failed ones. Motivated by this observation, we propose a training framework that explicitly strengthens visual conditioning in VLA models. Our approach first aligns action prediction with visual input via preference optimization on a track-following surrogate task, and then transfers the enhanced alignment to instruction-following task through latent-space distillation during supervised finetuning. Without introducing architectural modifications or additional data collection, our method improves both visual conditioning and task performance for discrete OpenVLA, and further yields consistent gains when extended to the continuous OpenVLA-OFT setting. Project website: https://vista-vla.github.io/ .

Joint planning through language-based interactions is a key area of human-AI teaming. Planning problems in the open world often involve various aspects of incomplete information and unknowns, e.g., objects involved, human goals/intents -- thus leading to knowledge gaps in joint planning. We consider the problem of discovering optimal interaction strategies for AI agents to actively elicit human inputs in object-driven planning. To this end, we propose Minimal Information Neuro-Symbolic Tree (MINT) to reason about the impact of knowledge gaps and leverage self-play with MINT to optimize the AI agent's elicitation strategies and queries. More precisely, MINT builds a symbolic tree by making propositions of possible human-AI interactions and by consulting a neural planning policy to estimate the uncertainty in planning outcomes caused by remaining knowledge gaps. Finally, we leverage LLM to search and summarize MINT's reasoning process and curate a set of queries to optimally elicit human inputs for best planning performance. By considering a family of extended Markov decision processes with knowledge gaps, we analyze the return guarantee for a given MINT with active human elicitation. Our evaluation on three benchmarks involving unseen/unknown objects of increasing realism shows that MINT-based planning attains near-expert returns by issuing a limited number of questions per task while achieving significantly improved rewards and success rates.

Despite increased adoption and advances in machine learning (ML), there are studies showing that many ML prototypes do not reach the production stage and that testing is still largely limited to testing model properties, such as model performance, without considering requirements derived from the system it will be a part of, such as throughput, resource consumption, or robustness. This limited view of testing leads to failures in model integration, deployment, and operations. In traditional software development, quality models such as ISO 25010 provide a widely used structured framework to assess software quality, define quality requirements, and provide a common language for communication with stakeholders. A newer standard, ISO 25059, defines a more specific quality model for AI systems. However, a problem with this standard is that it combines system attributes with ML component attributes, which is not helpful for a model developer, as many system attributes cannot be assessed at the component level. In this paper, we present a quality model for ML components that serves as a guide for requirements elicitation and negotiation and provides a common vocabulary for ML component developers and system stakeholders to agree on and define system-derived requirements and focus their testing efforts accordingly. The quality model was validated through a survey in which the participants agreed with its relevance and value. The quality model has been successfully integrated into an open-source tool for ML component testing and evaluation demonstrating its practical application.

Machine Learning (ML)-enabled systems challenge traditional Requirements Engineering (RE) and agile management due to data dependence, experimentation, and uncertain model behavior. Existing RE and agile practices remain poorly integrated and insufficiently tailored to these characteristics. This paper reports on the practical experience of applying RefineML, a requirements-focused approach for the continuous and agile refinement of ML-enabled systems, which integrates ML-tailored specification and agile management approaches with best practices derived from a systematic mapping study. The application context concerns an industry-academia collaboration project between PUC-Rio and EXA, a Brazilian cybersecurity company. For evaluation purposes, we applied questionnaires assessing RefineML's suitability and overall acceptance and semi-structured interviews. We applied thematic analysis to the collected qualitative data. Regarding suitability and acceptance, the results of the questionnaires indicated high perceived usefulness and intention to use. Based on the interviews, stakeholders perceived RefineML as improving communication and facilitating early feasibility assessments, as well as enabling dual-track governance of ML and software work, allowing continuous refinement of the model while evolving the overall software project. However, some limitations remain, particularly related to difficulties in operationalizing ML concerns into agile requirements and in estimating ML effort.

Can multi-task self-supervised learning on graphs be coordinated without the usual tug-of-war between objectives? Graph self-supervised learning (SSL) offers a growing toolbox of pretext objectives: mutual information, reconstruction, contrastive learning; yet combining them reliably remains a challenge due to objective interference and training instability. Most multi-pretext pipelines use per-update mixing, forcing every parameter update to be a compromise, leading to three failure modes: Disagreement (conflict-induced negative transfer), Drift (nonstationary objective utility), and Drought (hidden starvation of underserved objectives). We argue that coordination is fundamentally a temporal allocation problem: deciding when each objective receives optimization budget, not merely how to weigh them. We introduce ControlG, a control-theoretic framework that recasts multi-objective graph SSL as feedback-controlled temporal allocation by estimating per-objective difficulty and pairwise antagonism, planning target budgets via a Pareto-aware log-hypervolume planner, and scheduling with a Proportional-Integral-Derivative (PID) controller. Across 9 datasets, ControlG consistently outperforms state-of-the-art baselines, while producing an auditable schedule that reveals which objectives drove learning.

Multilingual language models (LMs) sometimes under-perform their monolingual counterparts, possibly due to capacity limitations. We quantify this ``multilingual penalty'' for lexical disambiguation--a task requiring precise semantic representations and contextualization mechanisms--using controlled datasets of human relatedness judgments for ambiguous words in both English and Spanish. Comparing monolingual and multilingual LMs from the same families, we find consistently reduced performance in multilingual LMs. We then explore three potential capacity constraints: representational (reduced embedding isotropy), attentional (reduced attention to disambiguating cues), and vocabulary-related (increased multi-token segmentation). Multilingual LMs show some evidence of all three limitations; moreover, these factors statistically account for the variance formerly attributed to a model's multilingual status. These findings suggest both that multilingual LMs do suffer from multiple capacity constraints, and that these constraints correlate with reduced disambiguation performance.

Learning meaningful causal representations from observations has emerged as a crucial task for facilitating machine learning applications and driving scientific discoveries in fields such as climate science, biology, and physics. This process involves disentangling high-level latent variables and their causal relationships from low-level observations. Previous work in this area that achieves identifiability typically focuses on cases where the observations are either i.i.d. or follow a latent discrete-time process. Nevertheless, many real-world settings require identifying latent variables that are continuous-time stochastic processes (e.g., multivariate point processes). To this end, we develop identifiable causal representation learning for continuous-time latent stochastic point processes. We study its identifiability by analyzing the geometry of the parameter space. Furthermore, we develop MUTATE, an identifiable variational autoencoder framework with a time-adaptive transition module to infer stochastic dynamics. Across simulated and empirical studies, we find that MUTATE can effectively answer scientific questions, such as the accumulation of mutations in genomics and the mechanisms driving neuron spike triggers in response to time-varying dynamics.

Planning with a learned model remains a key challenge in model-based reinforcement learning (RL). In decision-time planning, state representations are critical as they must support local cost computation while preserving long-horizon structure. In this paper, we show that the Laplacian representation provides an effective latent space for planning by capturing state-space distances at multiple time scales. This representation preserves meaningful distances and naturally decomposes long-horizon problems into subgoals, also mitigating the compounding errors that arise over long prediction horizons. Building on these properties, we introduce ALPS, a hierarchical planning algorithm, and demonstrate that it outperforms commonly used baselines on a selection of offline goal-conditioned RL tasks from OGBench, a benchmark previously dominated by model-free methods.

Operating effectively in novel real-world environments requires robotic systems to estimate and interact with previously unseen objects. Current state-of-the-art models address this challenge by using large amounts of training data and test-time samples to build black-box scene representations. In this work, we introduce a differentiable neuro-graphics model that combines neural foundation models with physics-based differentiable rendering to perform zero-shot scene reconstruction and robot grasping without relying on any additional 3D data or test-time samples. Our model solves a series of constrained optimization problems to estimate physically consistent scene parameters, such as meshes, lighting conditions, material properties, and 6D poses of previously unseen objects from a single RGBD image and bounding boxes. We evaluated our approach on standard model-free few-shot benchmarks and demonstrated that it outperforms existing algorithms for model-free few-shot pose estimation. Furthermore, we validated the accuracy of our scene reconstructions by applying our algorithm to a zero-shot grasping task. By enabling zero-shot, physically-consistent scene reconstruction and grasping without reliance on extensive datasets or test-time sampling, our approach offers a pathway towards more data efficient, interpretable and generalizable robot autonomy in novel environments.

Sparse Autoencoders (SAEs) are powerful tools for interpreting neural representations, yet their use in audio remains underexplored. We train SAEs across all encoder layers of Whisper and HuBERT, provide an extensive evaluation of their stability, interpretability, and show their practical utility. Over 50% of the features remain consistent across random seeds, and reconstruction quality is preserved. SAE features capture general acoustic and semantic information as well as specific events, including environmental noises and paralinguistic sounds (e.g. laughter, whispering) and disentangle them effectively, requiring removal of only 19-27% of features to erase a concept. Feature steering reduces Whisper's false speech detections by 70% with negligible WER increase, demonstrating real-world applicability. Finally, we find SAE features correlated with human EEG activity during speech perception, indicating alignment with human neural processing. The code and checkpoints are available at https://github.com/audiosae/audiosae_demo.

We formulate the fundamental laws governing learning dynamics, namely the conservation law and the decrease of total entropy. Within this framework, we introduce an entropy-based lifelong ensemble learning method. We evaluate its effectiveness by constructing an immunization mechanism to defend against transfer-based adversarial attacks on the CIFAR-10 dataset. Compared with a naive ensemble formed by simply averaging models specialized on clean and adversarial samples, the resulting logifold achieves higher accuracy in most test cases, with particularly large gains under strong perturbations.

Vision-language models (VLMs) have demonstrated strong performance in image geolocation, a capability further sharpened by frontier multimodal large reasoning models (MLRMs). This poses a significant privacy risk, as these widely accessible models can be exploited to infer sensitive locations from casually shared photos, often at street-level precision, potentially surpassing the level of detail the sharer consented or intended to disclose. While recent work has proposed applying a blanket restriction on geolocation disclosure to combat this risk, these measures fail to distinguish valid geolocation uses from malicious behavior. Instead, VLMs should maintain contextual integrity by reasoning about elements within an image to determine the appropriate level of information disclosure, balancing privacy and utility. To evaluate how well models respect contextual integrity, we introduce VLM-GEOPRIVACY, a benchmark that challenges VLMs to interpret latent social norms and contextual cues in real-world images and determine the appropriate level of location disclosure. Our evaluation of 14 leading VLMs shows that, despite their ability to precisely geolocate images, the models are poorly aligned with human privacy expectations. They often over-disclose in sensitive contexts and are vulnerable to prompt-based attacks. Our results call for new design principles in multimodal systems to incorporate context-conditioned privacy reasoning.

We study constrained contextual bandits (CCB) with adversarially chosen contexts, where each action yields a random reward and incurs a random cost. We adopt the standard realizability assumption: conditioned on the observed context, rewards and costs are drawn independently from fixed distributions whose expectations belong to known function classes. We consider the continuing setting, in which the algorithm operates over the entire horizon even after the budget is exhausted. In this setting, the objective is to simultaneously control regret and cumulative constraint violation. Building on the seminal SquareCB framework of Foster et al. (2018), we propose a simple and modular algorithmic scheme that leverages online regression oracles to reduce the constrained problem to a standard unconstrained contextual bandit problem with adaptively defined surrogate reward functions. In contrast to most prior work on CCB, which focuses on stochastic contexts, our reduction yields improved guarantees for the more general adversarial context setting, together with a compact and transparent analysis.

Agent-based cellular models simulate tissue evolution by capturing the behavior of individual cells, their interactions with neighboring cells, and their responses to the surrounding microenvironment. An important challenge in the field is scaling cellular resolution models to real-scale tumor simulations, which is critical for the development of digital twin models of diseases and requires the use of High-Performance Computing (HPC) since every time step involves trillions of operations. We hereby present a scalable HPC solution for the molecular diffusion modeling using an efficient implementation of state-of-the-art Finite Volume Method (FVM) frameworks. The paper systematically evaluates a novel scalable Biological Finite Volume Method (BioFVM) library and presents an extensive performance analysis of the available solutions. Results shows that our HPC proposal reach almost 200x speedup and up to 36% reduction in memory usage over the current state-of-the-art solutions, paving the way to efficiently compute the next generation of biological problems.

Managing one's digital footprint is overwhelming, as it spans multiple platforms and involves countless context-dependent decisions. Recent advances in agentic AI offer ways forward by enabling holistic, contextual privacy-enhancing solutions. Building on this potential, we adopted a ''human-as-the-unit'' perspective and investigated users' cross-context privacy challenges through 12 semi-structured interviews. Results reveal that people rely on ad hoc manual strategies while lacking comprehensive privacy controls, highlighting nine privacy-management challenges across applications, temporal contexts, and relationships. To explore solutions, we generated nine AI agent concepts and evaluated them via a speed-dating survey with 116 US participants. The three highest-ranked concepts were all post-sharing management tools with half or full agent autonomy, with users expressing greater trust in AI accuracy than in their own efforts. Our findings highlight a promising design space where users see AI agents bridging the fragments in privacy management, particularly through automated, comprehensive post-sharing remediation of users' digital footprints.

With the rapid progress of tool-using and agentic large language models (LLMs), Retrieval-Augmented Generation (RAG) is evolving from one-shot, passive retrieval into multi-turn, decision-driven evidence acquisition. Despite strong results in open-domain settings, existing agentic search frameworks commonly treat long documents as flat collections of chunks, underutilizing document-native priors such as hierarchical organization and sequential discourse structure. We introduce DeepRead, a structure-aware, multi-turn document reasoning agent that explicitly operationalizes these priors for long-document question answering. DeepRead leverages LLM-based OCR model to convert PDFs into structured Markdown that preserves headings and paragraph boundaries. It then indexes documents at the paragraph level and assigns each paragraph a coordinate-style metadata key encoding its section identity and in-section order. Building on this representation, DeepRead equips the LLM with two complementary tools: a Retrieve tool that localizes relevant paragraphs while exposing their structural coordinates (with lightweight scanning context), and a ReadSection tool that enables contiguous, order-preserving reading within a specified section and paragraph range. Our experiments demonstrate that DeepRead achieves significant improvements over Search-o1-style agentic search in document question answering. The synergistic effect between retrieval and reading tools is also validated. Our fine-grained behavioral analysis reveals a reading and reasoning paradigm resembling human-like ``locate then read'' behavior.

In-context learning (ICL) enables Large Language Models (LLMs) to adapt to new tasks with only a small set of examples at inference time, thereby avoiding task-specific fine-tuning. However, in-context examples may contain privacy-sensitive information that should not be revealed through model outputs. Existing differential privacy (DP) approaches to ICL are either computationally expensive or rely on heuristics with limited effectiveness, including context oversampling, synthetic data generation, or unnecessary thresholding. We reformulate private ICL through the lens of a Product-of-Experts model. This gives a theoretically grounded framework, and the algorithm can be trivially parallelized. We evaluate our method across five datasets in text classification, math, and vision-language. We find that our method improves accuracy by more than 30 percentage points on average compared to prior DP-ICL methods, while maintaining strong privacy guarantees.

The normalization of query and key vectors is an essential part of the Transformer architecture. It ensures that learning is stable regardless of the scale of these vectors. Some normalization approaches are available. In this preliminary work, a generalization of the QKNorm normalization scheme is proposed. The approach is based on the Lp norm, allowing non-Euclidean norms to be employed. Experimental results demonstrate the suitability of the method for a simple problem.

Large language models are enabling language-conditioned agents in interactive environments, but highly cooperative tasks often impose two simultaneous constraints: sub-second real-time coordination and sustained multi-episode adaptation under a strict online token budget. Existing approaches either rely on frequent in-episode reasoning that induces latency and timing jitter, or deliver post-episode improvements through unstructured text that is difficult to compile into reliable low-cost execution. We propose CoWork-X, an active co-evolution framework that casts peer collaboration as a closed-loop optimization problem across episodes, inspired by fast--slow memory separation. CoWork-X instantiates a Skill-Agent that executes via HTN (hierarchical task network)-based skill retrieval from a structured, interpretable, and compositional skill library, and a post-episode Co-Optimizer that performs patch-style skill consolidation with explicit budget constraints and drift regularization. Experiments in challenging Overcooked-AI-like realtime collaboration benchmarks demonstrate that CoWork-X achieves stable, cumulative performance gains while steadily reducing online latency and token usage.

Reward guidance has been applied to great success in the test-time adaptation of continuous diffusion models; it updates each denoising step using the gradients from a downstream reward model. We study reward guidance for discrete diffusion language models, where one cannot differentiate through the natural outputs of the model because they are discrete tokens. Existing approaches either replace these discrete tokens with continuous relaxations, or employ techniques like the straight-through estimator. In this work, we show the downsides of both these methods. The former degrades gradient feedback because the reward model has never been trained with continuous inputs. The latter involves incorrect optimization because the gradient evaluated at discrete tokens is used to update continuous logits. Our key innovation is to go beyond this tradeoff by introducing a novel mechanism called EntRGi: Entropy aware Reward Guidance that dynamically regulates the gradients from the reward model. By modulating the continuous relaxation using the model's confidence, our approach substantially improves reward guidance while providing reliable inputs to the reward model. We empirically validate our approach on a 7B-parameter diffusion language model across 3 diverse reward models and 3 multi-skill benchmarks, showing consistent improvements over state-of-the-art methods.

Low-Rank Adaptation (LoRA) is the prevailing approach for efficient large language model (LLM) fine-tuning. Building on this paradigm, recent studies have proposed alternative initialization strategies and architectural modifications, reporting substantial improvements over vanilla LoRA. However, these gains are often demonstrated under fixed or narrowly tuned hyperparameter settings, despite the known sensitivity of neural networks to training configurations. In this work, we systematically re-evaluate four representative LoRA variants alongside vanilla LoRA through extensive hyperparameter searches. Across mathematical and code generation tasks on diverse model scales, we find that different LoRA methods favor distinct learning rate ranges. Crucially, once learning rates are properly tuned, all methods achieve similar peak performance (within 1-2%), with only subtle rank-dependent behaviors. These results suggest that vanilla LoRA remains a competitive baseline and that improvements reported under single training configuration may not reflect consistent methodological advantages. Finally, a second-order analysis attributes the differing optimal learning rate ranges to variations in the largest Hessian eigenvalue, aligning with classical learning theories.

With the deep integration of facial recognition into online banking, identity verification, and other networked services, achieving effective decoupling of identity information from visual representations during image storage and transmission has become a critical challenge for privacy protection. To address this issue, we propose SIDeR, a Semantic decoupling-driven framework for unrestricted face privacy protection. SIDeR decomposes a facial image into a machine-recognizable identity feature vector and a visually perceptible semantic appearance component. By leveraging semantic-guided recomposition in the latent space of a diffusion model, it generates visually anonymous adversarial faces while maintaining machine-level identity consistency. The framework incorporates momentum-driven unrestricted perturbation optimization and a semantic-visual balancing factor to synthesize multiple visually diverse, highly natural adversarial samples. Furthermore, for authorized access, the protected image can be restored to its original form when the correct password is provided. Extensive experiments on the CelebA-HQ and FFHQ datasets demonstrate that SIDeR achieves a 99% attack success rate in black-box scenarios and outperforms baseline methods by 41.28% in PSNR-based restoration quality.

The allocation of scarce donor organs constitutes one of the most consequential algorithmic challenges in healthcare. While the field is rapidly transitioning from rigid, rule-based systems to machine learning and data-driven optimization, we argue that current approaches often overlook a fundamental barrier: incentives. In this position paper, we highlight that organ allocation is not merely a static optimization problem, but rather a complex game involving transplant centers, clinicians, and regulators. Focusing on US adult heart transplant allocation, we identify critical incentive misalignments across the decision-making pipeline, and present data showing that they are having adverse consequences today. Our main position is that the next generation of allocation policies should be incentive aware. We outline a research agenda for the machine learning community, calling for the integration of mechanism design, strategic classification, causal inference, and social choice to ensure robustness, efficiency, and fairness in the face of strategic behavior from the various constituent groups.

Online matching has been a mainstay in domains such as Internet advertising and organ allocation, but practical algorithms often lack strong theoretical guarantees. We take an important step toward addressing this by developing new online matching algorithms based on a coarsening approach. Although coarsening typically implies a loss of granularity, we show that, to the contrary, aggregating offline nodes into capacitated clusters can yield near-optimal theoretical guarantees. We apply our methodology to heart transplant allocation to develop theoretically grounded policies based on structural properties of historical data. In realistic simulations, our policy closely matches the performance of the omniscient benchmark. Our work bridges the gap between data-driven heuristics and pessimistic theoretical lower bounds, and provides rigorous justification for prior clustering-based approaches in organ allocation.

We endorse and expand upon Susan Schneider's critique of the linear model of AI progress and introduce two novel concepts: "familiar intelligence" and "strange intelligence". AI intelligence is likely to be strange intelligence, defying familiar patterns of ability and inability, combining superhuman capacities in some domains with subhuman performance in other domains, and even within domains sometimes combining superhuman insight with surprising errors that few humans would make. We develop and defend a nonlinear model of intelligence on which "general intelligence" is not a unified capacity but instead the ability to achieve a broad range of goals in a broad range of environments, in a manner that defies nonarbitrary reduction to a single linear quantity. We conclude with implications for adversarial testing approaches to evaluating AI capacities. If AI is strange intelligence, we should expect that even the most capable systems will sometimes fail in seemingly obvious tasks. On a nonlinear model of AI intelligence, such errors on their own do not demonstrate a system's lack of outstanding general intelligence. Conversely, excellent performance on one type of task, such as an IQ test, cannot warrant assumptions of broad capacities beyond that task domain.

Purpose: To investigate whether routinely acquired longitudinal MR-Linac images can be leveraged to characterize treatment-induced changes during radiotherapy, particularly subtle inter-fraction changes over short intervals (average of 2 days). Materials and Methods: This retrospective study included a series of 0.35T MR-Linac images from 761 patients. An artificial intelligence (deep learning) model was used to characterize treatment-induced changes by predicting the temporal order of paired images. The model was first trained with the images from the first and the last fractions (F1-FL), then with all pairs (All-pairs). Model performance was assessed using quantitative metrics (accuracy and AUC), compared to a radiologist's performance, and qualitative analyses - the saliency map evaluation to investigate affected anatomical regions. Input ablation experiments were performed to identify the anatomical regions altered by radiotherapy. The radiologist conducted an additional task on partial images reconstructed by saliency map regions, reporting observations as well. Quantitative image analysis was conducted to investigate the results from the model and the radiologist. Results: The F1-FL model yielded near-perfect performance (AUC of 0.99), significantly outperforming the radiologist. The All-pairs model yielded an AUC of 0.97. This performance reflects therapy-induced changes, supported by the performance correlation to fraction intervals, ablation tests and expert's interpretation. Primary regions driving the predictions were prostate, bladder, and pubic symphysis. Conclusion: The model accurately predicts temporal order of MR-Linac fractions and detects radiation-induced changes over one or a few days, including prostate and adjacent organ alterations confirmed by experts. This underscores MR-Linac's potential for advanced image analysis beyond image guidance.

With the increasing use of large language models (LLMs) for generating answers to biomedical questions, it is crucial to evaluate the quality of the generated answers and the references provided to support the facts in the generated answers. Evaluation of text generated by LLMs remains a challenge for question answering, retrieval-augmented generation (RAG), summarization, and many other natural language processing tasks in the biomedical domain, due to the requirements of expert assessment to verify consistency with the scientific literature and complex medical terminology. In this work, we propose BioACE, an automated framework for evaluating biomedical answers and citations against the facts stated in the answers. The proposed BioACE framework considers multiple aspects, including completeness, correctness, precision, and recall, in relation to the ground-truth nuggets for answer evaluation. We developed automated approaches to evaluate each of the aforementioned aspects and performed extensive experiments to assess and analyze their correlation with human evaluations. In addition, we considered multiple existing approaches, such as natural language inference (NLI) and pre-trained language models and LLMs, to evaluate the quality of evidence provided to support the generated answers in the form of citations into biomedical literature. With the detailed experiments and analysis, we provide the best approaches for biomedical answer and citation evaluation as a part of BioACE (https://github.com/deepaknlp/BioACE) evaluation package.

Errors are the primary bottleneck preventing practical quantum computing. This challenge is exacerbated in the distributed quantum computing regime, where quantum networks introduce additional communication-induced noise. While error mitigation techniques such as Zero Noise Extrapolation (ZNE) have proven effective for standalone quantum processors, their behavior in distributed architectures is not yet well understood. We investigate ZNE in this setting by comparing Global optimization (ZNE is applied prior to circuit partitioning), against Local optimization (ZNE is applied independently to each sub-circuit). Partitioning is performed on a monolithic circuit, which is then transformed into a distributed implementation by inserting noisy teleportation-based communication primitives between sub-circuits. We evaluate both approaches across varying numbers of quantum processing units (QPUs) and under heterogeneous local and network noise conditions. Our results demonstrate that Global ZNE exhibits superior scalability, achieving error reductions of up to $48\%$ across six QPUs. Moreover, we observe counterintuitive noise behavior, where increasing the number of QPUs improves mitigation effectiveness despite higher communication overhead. These findings highlight fundamental trade-offs in distributed quantum error mitigation and raise new questions regarding the interplay between circuit structure, partitioning strategies, and network noise.

This work introduces a stochastic hierarchical optimization framework inspired by Sloppy Model theory for the efficient calibration of physical models. Central to this method is the use of a reduced Hessian approximation, which identifies and targets the stiff parameter subspace using minimal simulation queries. This strategy enables efficient navigation of highly anisotropic landscapes, avoiding the computational burden of exhaustive sampling. To ensure rigorous inference, we integrate this approach with a probabilistic formulation that derives a principled objective loss function directly from observed data. We validate the framework by applying it to the problem of plasma-surface interactions, where accurate modelling is strictly limited by uncertainties in surface reactivity parameters and the computational cost of kinetic simulations. Comparative analysis demonstrates that our method consistently outperforms baseline optimization techniques in sample efficiency. This approach offers a general and scalable tool for optimizing models of complex reaction systems, ranging from plasma chemistry to biochemical networks.

We study the sample complexity of shadow tomography in the high-precision regime under realistic measurement constraints. Given an unknown $d$-dimensional quantum state $ρ$ and a known set of observables $\{O_i\}_{i=1}^m$, the goal is to estimate expectation values $\{\mathrm{tr}(O_iρ)\}_{i=1}^m$ to accuracy $ε$ in $L_p$-norm, using possibly adaptive measurements that act on $O(\mathrm{polylog}(d))$ number of copies of $ρ$ at a time. We focus on the regime where $ε$ is below an instance-dependent threshold. Our main contribution is an instance-optimal characterization of the sample complexity as $\tildeΘ(Γ_p/ε^2)$, where $Γ_p$ is a function of $\{O_i\}_{i=1}^m$ defined via an optimization formula involving the inverse Fisher information matrix. Previously, tight bounds were known only in special cases, e.g. Pauli shadow tomography with $L_\infty$-norm error. Concretely, we first analyze a simpler oblivious variant where the goal is to estimate an observable of the form $\sum_{i=1}^m α_i O_i$ with $\|α\|_q = 1$ (where $q$ is dual to $p$) revealed after the measurement. For single-copy measurements, we obtain a sample complexity of $Θ(Γ^{\mathrm{ob}}_p/ε^2)$. We then show $\tildeΘ(Γ_p/ε^2)$ is necessary and sufficient for the original problem, with the lower bound applying to unbiased, bounded estimators. Our upper bounds rely on a two-step algorithm combining coarse tomography with local estimation. Notably, $Γ^{\mathrm{ob}}_\infty = Γ_\infty$. In both cases, allowing $c$-copy measurements improves the sample complexity by at most $Ω(1/c)$. Our results establish a quantitative correspondence between quantum learning and metrology, unifying asymptotic metrological limits with finite-sample learning guarantees.

Polycystic Ovary Syndrome (PCOS) is a widespread disorder in women of reproductive age, characterized by a hormonal imbalance, irregular periods, and multiple ovarian cysts. Infertility, metabolic syndrome, and cardiovascular risks are long-term complications that make early detection essential. In this paper, we design a powerful framework based on transfer learning utilizing DenseNet201 and ResNet50 for classifying ovarian ultrasound images. The model was trained on an online dataset containing 3856 ultrasound images of cyst-infected and non-infected patients. Each ultrasound frame was resized to 224x224 pixels and encoded with precise pathological indicators. The MixUp and CutMix augmentation strategies were used to improve generalization, yielding a peak validation accuracy of 99.80% by Densenet201 and a validation loss of 0.617 with alpha values of 0.25 and 0.4, respectively. We evaluated the model's interpretability using leading Explainable AI (XAI) approaches such as SHAP, Grad-CAM, and LIME, reasoning with and presenting explicit visual reasons for the model's behaviors, therefore increasing the model's transparency. This study proposes an automated system for medical picture diagnosis that may be used effectively and confidently in clinical practice.

Despite extensive study, the phase structure of the wavefunctions in frustrated Heisenberg antiferromagnets (HAF) is not yet systematically characterized. In this work, we represent the Hilbert space of an HAF as a weighted graph, which we term the Hilbert graph (HG), whose vertices are spin configurations and whose edges are generated by off-diagonal spin-flip terms of the Heisenberg Hamiltonian, with weights set by products of wavefunction amplitudes. Holding the amplitudes fixed and restricting phases to $\mathbb{Z}_2$ values, the phase-dependent variational energy can be recast as a classical Ising antiferromagnet on the HG, so that phase reconstruction of the ground state reduces to a weighted Max-Cut instance. This shows that phase reconstruction HAF is worst-case NP-hard and provides a direct link between wavefunction sign structure and combinatorial optimization.

Training-time privileged information (PI) can enable language models to succeed on tasks they would otherwise fail, making it a powerful tool for reinforcement learning in hard, long-horizon settings. However, transferring capabilities learned with PI to policies that must act without it at inference time remains a fundamental challenge. We study this problem in the context of distilling frontier models for multi-turn agentic environments, where closed-source systems typically hide their internal reasoning and expose only action trajectories. This breaks standard distillation pipelines, since successful behavior is observable but the reasoning process is not. For this, we introduce π-Distill, a joint teacher-student objective that trains a PI-conditioned teacher and an unconditioned student simultaneously using the same model. Additionally, we also introduce On-Policy Self-Distillation (OPSD), an alternative approach that trains using Reinforcement Learning (RL) with a reverse KL-penalty between the student and the PI-conditioned teacher. We show that both of these algorithms effectively distill frontier agents using action-only PI. Specifically we find that π-Distill and in some cases OPSD, outperform industry standard practices (Supervised finetuning followed by RL) that assume access to full Chain-of-Thought supervision across multiple agentic benchmarks, models, and forms of PI. We complement our results with extensive analysis that characterizes the factors enabling effective learning with PI, focusing primarily on π-Distill and characterizing when OPSD is competitive.

Sets represent a fundamental abstraction across many types of data. To handle the unordered nature of set-structured data, models such as DeepSets and PointNet rely on fixed, non-learnable pooling operations (e.g., sum or max) -- a design choice that can hinder the transferability of learned embeddings and limits model expressivity. More recently, learnable aggregation functions have been proposed as more expressive alternatives. In this work, we advance this line of research by introducing the Neuralized Kolmogorov Mean (NKM) -- a novel, trainable framework for learning a generalized measure of central tendency through an invertible neural function. We further propose quasi-arithmetic neural networks (QUANNs), which incorporate the NKM as a learnable aggregation function. We provide a theoretical analysis showing that, QUANNs are universal approximators for a broad class of common set-function decompositions and, thanks to their invertible neural components, learn more structured latent representations. Empirically, QUANNs outperform state-of-the-art baselines across diverse benchmarks, while learning embeddings that transfer effectively even to tasks that do not involve sets.

The rapid growth of large language models (LLMs) has outpaced the evolution of single-GPU hardware, making model scale increasingly constrained by memory capacity rather than computation. While modern training systems extend GPU memory through distributed parallelism and offloading across CPU and storage tiers, they fundamentally retain a GPU-centric execution paradigm in which GPUs host persistent model replicas and full autograd graphs. As a result, scaling large models remains tightly coupled to multi-GPU clusters, complex distributed runtimes, and unpredictable host memory consumption, creating substantial barriers for node-scale post-training workloads such as instruction tuning, alignment, and domain adaptation. We present Horizon-LM, a memory-centric training system that redefines the roles of CPU and GPU for large-model optimization. Horizon-LM treats host memory as the authoritative parameter store and uses GPUs solely as transient compute engines through a CPU-master, GPU-template execution model. By eliminating persistent GPU-resident modules and autograd graphs, employing explicit recomputation with manual gradient propagation, and introducing a pipelined double-buffered execution engine, Horizon-LM decouples model scale from GPU count and bounds memory usage to the theoretical parameter footprint. On a single H200 GPU with 1.5\,TB host RAM, Horizon-LM reliably trains models up to 120B parameters. On a standard single A100 machine, Horizon-LM achieves up to 12.2$\times$ higher training throughput than DeepSpeed ZeRO-3 with CPU offloading while preserving numerical correctness. Across platforms and scales, Horizon-LM sustains high device utilization and predictable memory growth, demonstrating that host memory, not GPU memory, defines the true feasibility boundary for node-scale large-model training.

Nonnegative matrix factorization (NMF) is a popular data embedding technique. Given a nonnegative data matrix $X$, it aims at finding two lower dimensional matrices, $W$ and $H$, such that $X\approx WH$, where the factors $W$ and $H$ are constrained to be element-wise nonnegative. The factor $W$ serves as a basis for the columns of $X$. In order to obtain more interpretable and unique solutions, minimum-volume NMF (MinVol NMF) minimizes the volume of $W$. In this paper, we consider the dual approach, where the volume of $H$ is maximized instead; this is referred to as maximum-volume NMF (MaxVol NMF). MaxVol NMF is identifiable under the same conditions as MinVol NMF in the noiseless case, but it behaves rather differently in the presence of noise. In practice, MaxVol NMF is much more effective to extract a sparse decomposition and does not generate rank-deficient solutions. In fact, we prove that the solutions of MaxVol NMF with the largest volume correspond to clustering the columns of $X$ in disjoint clusters, while the solutions of MinVol NMF with smallest volume are rank deficient. We propose two algorithms to solve MaxVol NMF. We also present a normalized variant of MaxVol NMF that exhibits better performance than MinVol NMF and MaxVol NMF, and can be interpreted as a continuum between standard NMF and orthogonal NMF. We illustrate our results in the context of hyperspectral unmixing.

Deep learning has emerged as a transformative tool for the neural surrogate modeling of partial differential equations (PDEs), known as neural PDE solvers. However, scaling these solvers to industrial-scale geometries with over $10^8$ cells remains a fundamental challenge due to the prohibitive memory complexity of processing high-resolution meshes. We present Transolver-3, a new member of the Transolver family as a highly scalable framework designed for high-fidelity physics simulations. To bridge the gap between limited GPU capacity and the resolution requirements of complex engineering tasks, we introduce two key architectural optimizations: faster slice and deslice by exploiting matrix multiplication associative property and geometry slice tiling to partition the computation of physical states. Combined with an amortized training strategy by learning on random subsets of original high-resolution meshes and a physical state caching technique during inference, Transolver-3 enables high-fidelity field prediction on industrial-scale meshes. Extensive experiments demonstrate that Transolver-3 is capable of handling meshes with over 160 million cells, achieving impressive performance across three challenging simulation benchmarks, including aircraft and automotive design tasks.

Retrieval Augmented Generation (RAG) is a highly effective paradigm for keeping LLM-based responses up-to-date and reducing the likelihood of hallucinations. Yet, RAG was recently shown to be quite vulnerable to corpus knowledge poisoning: an attacker injects misleading documents to the corpus to steer an LLM's output to an undesired response. We argue that the standard causal attention mechanism in LLMs enables harmful cross-document interactions, specifically in cases of attacks. Accordingly, we introduce a novel defense approach for RAG: Sparse Document Attention RAG (SDAG). This is a block-sparse attention mechanism that disallows cross-attention between retrieved documents. SDAG requires a minimal inference-time change to the attention mask; furthermore, no fine-tuning or additional architectural changes are needed. We present an empirical evaluation of LLM-based question answering (QA) with a variety of attack strategies on RAG. We show that our SDAG method substantially outperforms the standard causal attention mechanism in terms of attack success rate. We further demonstrate the clear merits of integrating SDAG with state-of-the-art RAG defense methods. Specifically, the integration results in performance that is statistically significantly better than the state-of-the-art.

Generative artificial intelligence attracts significant attention, especially with the introduction of large language models. Its capabilities are being exploited to solve various software engineering tasks. Thanks to their ability to understand natural language and generate natural language responses, large language models are great for processing various software documentation artifacts. At the same time, large language models excel at understanding structured languages, having the potential for working with software programs and models. We conduct a literature review on the usage of large language models for software engineering tasks related to documentation and modeling. We analyze articles from four major venues in the area, organize them per tasks they solve, and provide an overview of used prompt techniques, metrics, approaches to human-based evaluation, and major datasets.

Selecting the best data mixture is critical for successful Supervised Fine-Tuning (SFT) of Multimodal Large Language Models. However, determining the optimal mixture weights across multiple domain-specific datasets remains a significant bottleneck due to the combinatorial search space and the high cost associated with even a single training run. This is the so-called Data Mixture Optimization (DMO) problem. On the other hand, model merging unifies domain-specific experts through parameter interpolation. This strategy is efficient, as it only requires a single training run per domain, yet oftentimes leads to suboptimal models. In this work, we take the best of both worlds, studying model merging as an efficient strategy for estimating the performance of different data mixtures. We train domain-specific multimodal experts and evaluate their weighted parameter-space combinations to estimate the efficacy of corresponding data mixtures. We conduct extensive experiments on 14 multimodal benchmarks, and empirically demonstrate that the merged proxy models exhibit a high rank correlation with models trained on actual data mixtures. This decouples the search for optimal mixtures from the resource-intensive training process, thereby providing a scalable and efficient strategy for navigating the complex landscape of mixture weights. Code is publicly available at https://github.com/BerasiDavide/mLLMs_merging_4_DMO.

Open-weight language models are increasingly used in production settings, raising new security challenges. One prominent threat in this context is backdoor attacks, in which adversaries embed hidden behaviors in language models that activate under specific conditions. Previous work has assumed that adversaries have access to training pipelines or deployment infrastructure. We propose a novel attack surface requiring neither, which utilizes the chat template. Chat templates are executable Jinja2 programs invoked at every inference call, occupying a privileged position between user input and model processing. We show that an adversary who distributes a model with a maliciously modified template can implant an inference-time backdoor without modifying model weights, poisoning training data, or controlling runtime infrastructure. We evaluated this attack vector by constructing template backdoors targeting two objectives: degrading factual accuracy and inducing emission of attacker-controlled URLs, and applied them across eighteen models spanning seven families and four inference engines. Under triggered conditions, factual accuracy drops from 90% to 15% on average while attacker-controlled URLs are emitted with success rates exceeding 80%; benign inputs show no measurable degradation. Backdoors generalize across inference runtimes and evade all automated security scans applied by the largest open-weight distribution platform. These results establish chat templates as a reliable and currently undefended attack surface in the LLM supply chain.

For real-world deployment of general-purpose LLM agents, the core challenge is often not tool use itself, but efficient domain adaptation under rapidly evolving toolsets, APIs, and protocols. Repeated LoRA or SFT across domains incurs exponentially growing training and maintenance costs, while prompt or schema methods are brittle under distribution shift and complex interfaces. We propose \textbf{Activation Steering Adapter (ASA}), a lightweight, inference-time, training-free mechanism that reads routing signals from intermediate activations and uses an ultra-light router to produce adaptive control strengths for precise domain alignment. Across multiple model scales and domains, ASA achieves LoRA-comparable adaptation with substantially lower overhead and strong cross-model transferability, making it ideally practical for robust, scalable, and efficient multi-domain tool ecosystems with frequent interface churn dynamics.

For the past decade, the trajectory of generative artificial intelligence (AI) has been dominated by a model-centric paradigm driven by scaling laws. Despite significant leaps in visual fidelity, this approach has encountered a ``usability ceiling'' manifested as the Intent-Execution Gap (i.e., the fundamental disparity between a creator's high-level intent and the stochastic, black-box nature of current single-shot models). In this paper, inspired by the Vibe Coding, we introduce the \textbf{Vibe AIGC}, a new paradigm for content generation via agentic orchestration, which represents the autonomous synthesis of hierarchical multi-agent workflows. Under this paradigm, the user's role transcends traditional prompt engineering, evolving into a Commander who provides a Vibe, a high-level representation encompassing aesthetic preferences, functional logic, and etc. A centralized Meta-Planner then functions as a system architect, deconstructing this ``Vibe'' into executable, verifiable, and adaptive agentic pipelines. By transitioning from stochastic inference to logical orchestration, Vibe AIGC bridges the gap between human imagination and machine execution. We contend that this shift will redefine the human-AI collaborative economy, transforming AI from a fragile inference engine into a robust system-level engineering partner that democratizes the creation of complex, long-horizon digital assets.

Hyperbolic representation learning has been widely used to extract implicit hierarchies within data, and recently it has found its way to the open-world classification task of Generalized Category Discovery (GCD). However, prior hyperbolic GCD methods only use hyperbolic geometry for representation learning and transform back to Euclidean geometry when clustering. We hypothesize this is suboptimal. Therefore, we present Hyperbolic Clustered GCD (HC-GCD), which learns embeddings in the Lorentz Hyperboloid model of hyperbolic geometry, and clusters these embeddings directly in hyperbolic space using a hyperbolic K-Means algorithm. We test our model on the Semantic Shift Benchmark datasets, and demonstrate that HC-GCD is on par with the previous state-of-the-art hyperbolic GCD method. Furthermore, we show that using hyperbolic K-Means leads to better accuracy than Euclidean K-Means. We carry out ablation studies showing that clipping the norm of the Euclidean embeddings leads to decreased accuracy in clustering unseen classes, and increased accuracy for seen classes, while the overall accuracy is dataset dependent. We also show that using hyperbolic K-Means leads to more consistent clusters when varying the label granularity.

We show that decoder-only large language models exhibit a depth-wise transition from context-processing to prediction-forming phases of computation accompanied by a reorganization of representational geometry. Using a unified framework combining geometric analysis with mechanistic intervention, we demonstrate that late-layer representations implement a structured geometric code that enables selective causal control over token prediction. Specifically, angular organization of the representation geometry parametrizes prediction distributional similarity, while representation norms encode context-specific information that does not determine prediction. Together, these results provide a mechanistic-geometric account of the dynamics of transforming context into predictions in LLMs.

Future AI deployments will likely be monitored for malicious behaviour. The ability of these AIs to subvert monitors by adversarially selecting against them - attack selection - is particularly concerning. To study this, we let a red team create attack policies that attempt to insert attacks into code without being caught by a monitor in the concentrated BigCodeBench backdooring setting. We decompose attack selection into two problems: mapping attacks to a quality score and mapping quality scores to submission probabilities. We frame attack selection as a classification problem and show that safety is significantly more sensitive to FPR than TPR. We find that prompting the attacker model to reason about the monitor while being cautious with attack selection reduces safety from a baseline of 99% to 59% at 0.5% auditing budget, emphasizing that eliciting attack selection capabilities of models is vital to avoid overly optimistic safety scores in control evaluations.

Humanoid robots can suffer significant performance drops under small changes in dynamics, task specifications, or environment setup. We propose HoRD, a two-stage learning framework for robust humanoid control under domain shift. First, we train a high-performance teacher policy via history-conditioned reinforcement learning, where the policy infers latent dynamics context from recent state--action trajectories to adapt online to diverse randomized dynamics. Second, we perform online distillation to transfer the teacher's robust control capabilities into a transformer-based student policy that operates on sparse root-relative 3D joint keypoint trajectories. By combining history-conditioned adaptation with online distillation, HoRD enables a single policy to adapt zero-shot to unseen domains without per-domain retraining. Extensive experiments show HoRD outperforms strong baselines in robustness and transfer, especially under unseen domains and external perturbations. Code and project page are available at https://tonywang-0517.github.io/hord/.

The rapid growth of large language models (LLMs) has heightened the importance of post-training quantization (PTQ) for reducing memory and computation costs. Among PTQ methods, GPTQ has gained significant attention for its efficiency, enabling billion-scale LLMs to be quantized within a few GPU hours. However, GPTQ's assumption of layer-wise independence leads to severe accuracy drops in low-bit regimes. Recently, BoA improved upon GPTQ by incorporating inter-layer dependencies within attention modules, but its reliance on sequential quantization across all out-channels makes it substantially less efficient. In this paper, we propose TurboBoA, a new backpropagation-free PTQ algorithm that preserves the accuracy benefits of BoA while significantly accelerating the process. The proposed TurboBoA introduces three key innovations: (i) joint quantization of multiple out-channels with a closed-form error compensation rule, which reduces sequential bottlenecks and yields more than a three-fold speedup; (ii) a correction mechanism for errors propagated from preceding quantized layers; and (iii) adaptive grid computation with coordinate descent refinement to maintain alignment during iterative updates. Extensive experiments demonstrate that TurboBoA delivers substantial acceleration over BoA while consistently improving accuracy. When combined with outlier suppression techniques, it achieves state-of-the-art results in both weight-only and weight-activation quantization. The code will be available at https://github.com/SamsungLabs/TurboBoA.

While online Reinforcement Learning has emerged as a crucial technique for aligning flow matching models with human preferences, current approaches are hindered by inefficient exploration during training rollouts. Relying on undirected stochasticity and sparse outcome rewards, these methods struggle to discover high-reward samples, resulting in data-inefficient and slow optimization. To address these limitations, we propose Euphonium, a novel framework that steers generation via process reward gradient guided dynamics. Our key insight is to formulate the sampling process as a theoretically principled Stochastic Differential Equation that explicitly incorporates the gradient of a Process Reward Model into the flow drift. This design enables dense, step-by-step steering toward high-reward regions, advancing beyond the unguided exploration in prior works, and theoretically encompasses existing sampling methods (e.g., Flow-GRPO, DanceGRPO) as special cases. We further derive a distillation objective that internalizes the guidance signal into the flow network, eliminating inference-time dependency on the reward model. We instantiate this framework with a Dual-Reward Group Relative Policy Optimization algorithm, combining latent process rewards for efficient credit assignment with pixel-level outcome rewards for final visual fidelity. Experiments on text-to-video generation show that Euphonium achieves better alignment compared to existing methods while accelerating training convergence by 1.66x.

Artificial intelligence systems introduce complex privacy risks throughout their lifecycle, especially when processing sensitive or high-dimensional data. Beyond the seven traditional privacy threat categories defined by the LINDDUN framework, AI systems are also exposed to model-centric privacy attacks such as membership inference and model inversion, which LINDDUN does not cover. To address both classical LINDDUN threats and additional AI-driven privacy attacks, PriMod4AI introduces a hybrid privacy threat modeling approach that unifies two structured knowledge sources, a LINDDUN knowledge base representing the established taxonomy, and a model-centric privacy attack knowledge base capturing threats outside LINDDUN. These knowledge bases are embedded into a vector database for semantic retrieval and combined with system level metadata derived from Data Flow Diagram. PriMod4AI uses retrieval-augmented and Data Flow specific prompt generation to guide large language models (LLMs) in identifying, explaining, and categorizing privacy threats across lifecycle stages. The framework produces justified and taxonomy-grounded threat assessments that integrate both classical and AI-driven perspectives. Evaluation on two AI systems indicates that PriMod4AI provides broad coverage of classical privacy categories while additionally identifying model-centric privacy threats. The framework produces consistent, knowledge-grounded outputs across LLMs, as reflected in agreement scores in the observed range.

Retrieval-augmented generation (RAG) is now standard for knowledge-intensive LLM tasks, but most systems still treat every query as fresh, repeatedly re-retrieving long passages and re-reasoning from scratch, inflating tokens, latency, and cost. We present AutoPrunedRetriever, a graph-style RAG system that persists the minimal reasoning subgraph built for earlier questions and incrementally extends it for later ones. AutoPrunedRetriever stores entities and relations in a compact, ID-indexed codebook and represents questions, facts, and answers as edge sequences, enabling retrieval and prompting over symbolic structure instead of raw text. To keep the graph compact, we apply a two-layer consolidation policy (fast ANN/KNN alias detection plus selective $k$-means once a memory threshold is reached) and prune low-value structure, while prompts retain only overlap representatives and genuinely new evidence. We instantiate two front ends: AutoPrunedRetriever-REBEL, which uses REBEL as a triplet parser, and AutoPrunedRetriever-llm, which swaps in an LLM extractor. On GraphRAG-Benchmark (Medical and Novel), both variants achieve state-of-the-art complex reasoning accuracy, improving over HippoRAG2 by roughly 9--11 points, and remain competitive on contextual summarize and generation. On our harder STEM and TV benchmarks, AutoPrunedRetriever again ranks first, while using up to two orders of magnitude fewer tokens than graph-heavy baselines, making it a practical substrate for long-running sessions, evolving corpora, and multi-agent pipelines.

The internalization of chain-of-thought processes into hidden states has emerged as a highly efficient paradigm for scaling test-time compute. However, existing activation steering methods rely on static control vectors that fail to adapt to the non-stationary evolution of complex reasoning tasks. To address this limitation, we propose STIR (Self-Distilled Tools for Internal Reasoning), a framework that reformulates reasoning enhancement as a dynamic latent trajectory control problem. STIR introduces a synergistic three-stage pipeline: (1) differential intrinsic action induction harvests latent reasoning successes to crystallize steering primitives; (2) sparse control basis construction curates a compact, geometrically diverse tool library; and (3) value-modulated trajectory intervention dynamically injects context-specific impulses via anchor-based gating. Extensive experiments on six arithmetic and logical benchmarks across four representative models demonstrate that STIR improves average accuracy by 1.9% to 7.5% while reducing average token consumption by up to 35% compared to vanilla decoding. These findings demonstrate that the benefits of explicit chain-of-thought can be realized through dynamic latent trajectory control, internalizing the reasoning process to bypass the explicit generation while achieving superior fidelity. Our code is available at https://github.com/sznnzs/LLM-Latent-Action.

We present a formal problem formulation for \textit{Reliable} Audio-Visual Question Answering ($\mathcal{R}$-AVQA), where we prefer abstention over answering incorrectly. While recent AVQA models have high accuracy, their ability to identify when they are likely wrong and their consequent abstention from answering remain underexplored areas of research. To fill this gap, we explore several approaches and then propose Adaptive Confidence Refinement (ACR), a lightweight method to further enhance the performance of $\mathcal{R}$-AVQA. Our key insight is that the Maximum Softmax Probability (MSP) is Bayes-optimal only under strong calibration, a condition usually not met in deep neural networks, particularly in multimodal models. Instead of replacing MSP, our ACR maintains it as a primary confidence signal and applies input-adaptive residual corrections when MSP is deemed unreliable. ACR introduces two learned heads: i) a Residual Risk Head that predicts low-magnitude correctness residuals that MSP does not capture, and ii) a Confidence Gating Head to determine MSP trustworthiness. Our experiments and theoretical analysis show that ACR consistently outperforms existing methods on in- and out-of-disrtibution, and data bias settings across three different AVQA architectures, establishing a solid foundation for $\mathcal{R}$-AVQA task. The code and checkpoints will be available upon acceptance \href{https://github.com/PhuTran1005/R-AVQA}{at here}

Neural operators have emerged as powerful surrogates for the solution of partial differential equations (PDEs), yet their ability to handle general, highly variable boundary conditions (BCs) remains limited. Existing approaches often fail when the solution operator exhibits strong sensitivity to boundary forcings. We propose a general framework for conditioning neural operators on complex non-homogeneous BCs through function extensions. Our key idea is to map boundary data to latent pseudo-extensions defined over the entire spatial domain, enabling any standard operator learning architecture to consume boundary information. The resulting operator, coupled with an arbitrary domain-to-domain neural operator, can learn rich dependencies on complex BCs and input domain functions at the same time. To benchmark this setting, we construct 18 challenging datasets spanning Poisson, linear elasticity, and hyperelasticity problems, with highly variable, mixed-type, component-wise, and multi-segment BCs on diverse geometries. Our approach achieves state-of-the-art accuracy, outperforming baselines by large margins, while requiring no hyperparameter tuning across datasets. Overall, our results demonstrate that learning boundary-to-domain extensions is an effective and practical strategy for imposing complex BCs in existing neural operator frameworks, enabling accurate and robust scientific machine learning models for a broader range of PDE-governed problems.

Multimodal machine learning, mimicking the human brain's ability to integrate various modalities has seen rapid growth. Most previous multimodal models are trained on perfectly paired multimodal input to reach optimal performance. In real-world deployments, however, the presence of modality is highly variable and unpredictable, causing the pre-trained models in suffering significant performance drops and fail to remain robust with dynamic missing modalities circumstances. In this paper, we present a novel Cyclic INformative Learning framework (CyIN) to bridge the gap between complete and incomplete multimodal learning. Specifically, we firstly build an informative latent space by adopting token- and label-level Information Bottleneck (IB) cyclically among various modalities. Capturing task-related features with variational approximation, the informative bottleneck latents are purified for more efficient cross-modal interaction and multimodal fusion. Moreover, to supplement the missing information caused by incomplete multimodal input, we propose cross-modal cyclic translation by reconstruct the missing modalities with the remained ones through forward and reverse propagation process. With the help of the extracted and reconstructed informative latents, CyIN succeeds in jointly optimizing complete and incomplete multimodal learning in one unified model. Extensive experiments on 4 multimodal datasets demonstrate the superior performance of our method in both complete and diverse incomplete scenarios.

Large Language Models (LLMs) have demonstrated impressive reasoning capabilities, but their substantial size often demands significant computational resources. To reduce resource consumption and accelerate inference, it is essential to eliminate redundant parameters without compromising performance. However, conventional pruning methods that directly remove such parameters often lead to a dramatic drop in model performance in reasoning tasks, and require extensive post-training to recover the lost capabilities. In this work, we propose a gradual compacting method that divides the compression process into multiple fine-grained iterations, applying a Prune-Tune Loop (PTL) at each stage to incrementally reduce model size while restoring performance with finetuning. This iterative approach-reminiscent of the "boiling frog" effect-enables the model to be progressively compressed without abrupt performance loss. Experimental results show that PTL can compress LLMs to nearly half their original size with only lightweight post-training, while maintaining performance comparable to the original model on reasoning tasks. Moreover, PTL is flexible and can be applied to various pruning strategies, such as neuron pruning and layer pruning, as well as different post-training methods, including continual pre-training and reinforcement learning. Additionally, experimental results confirm the effectiveness of PTL on a variety of tasks beyond mathematical reasoning, such as code generation, demonstrating its broad applicability.

Large Language Models (LLMs) frequently prioritize conflicting in-context information over pre-existing parametric memory, a phenomenon often termed sycophancy or compliance. However, the mechanistic realization of this behavior remains obscure, specifically how the model resolves these knowledge conflicts through compliance, and whether this suppression arises from signal magnitude dilution or directional geometric alteration within the residual stream. To resolve this, we conducted a layer-wise geometric analysis across Qwen-4B, Llama-3.1-8B, and GLM-4-9B, decomposing the residual stream updates induced by counter-factual contexts into radial (norm-based) and angular (cosine-based) components. Our empirical results reject the universality of the "Manifold Dilution" hypothesis, as two of the three architectures maintained stable residual norms despite exhibiting significant performance degradation on factual queries. Instead, we observed that compliance is consistently characterized by "Orthogonal Interference," where the conflicting context injects a steering vector that is quasi-orthogonal to the ground-truth direction, effectively rotating the hidden state representation. This suggests that models do not "unlearn" or suppress the magnitude of internal truths but rather employ a mechanism of geometric displacement to bypass the correct unembedding vector, effectively simulating adoption while preserving the original structural magnitude. These findings challenge scalar confidence metrics for detecting hallucinations and underscore the necessity of vectorial monitoring to distinguish between genuine knowledge integration and superficial in-context mimicry.

Analysis and anomaly detection in event tensor streams consisting of timestamps and multiple attributes - such as communication logs(time, IP address, packet length)- are essential tasks in data mining. While existing tensor decomposition and anomaly detection methods provide useful insights, they face the following two limitations. (i) They cannot handle heterogeneous tensor streams, which comprises both categorical attributes(e.g., IP address) and continuous attributes(e.g., packet length). They typically require either discretizing continuous attributes or treating categorical attributes as continuous, both of which distort the underlying statistical properties of the data.Furthermore, incorrect assumptions about the distribution family of continuous attributes often degrade the model's performance. (ii) They discretize timestamps, failing to track the temporal dynamics of streams(e.g., trends, abnormal events), which makes them ineffective for detecting anomalies at the group level, referred to as 'group anomalies' (e.g, DoS attacks). To address these challenges, we propose HeteroComp, a method for continuously summarizing heterogeneous tensor streams into 'components' representing latent groups in each attribute and their temporal dynamics, and detecting group anomalies. Our method employs Gaussian process priors to model unknown distributions of continuous attributes, and temporal dynamics, which directly estimate probability densities from data. Extracted components give concise but effective summarization, enabling accurate group anomaly detection. Extensive experiments on real datasets demonstrate that HeteroComp outperforms the state-of-the-art algorithms for group anomaly detection accuracy, and its computational time does not depend on the data stream length.

Large language models (LLMs) have significantly advanced protein representation learning. However, their capacity to interpret and design antibodies through natural language remains limited. To address this challenge, we present AFD-Instruction, the first large-scale instruction dataset with functional annotations tailored to antibodies. This dataset encompasses two key components: antibody understanding, which infers functional attributes directly from sequences, and antibody design, which enables de novo sequence generation under functional constraints. These components provide explicit sequence-function alignment and support antibody design guided by natural language instructions. Extensive instruction-tuning experiments on general-purpose LLMs demonstrate that AFD-Instruction consistently improves performance across diverse antibody-related tasks. By linking antibody sequences with textual descriptions of function, AFD-Instruction establishes a new foundation for advancing antibody modeling and accelerating therapeutic discovery.

We propose a new class of linear-time attention mechanisms based on a relaxed and computationally efficient formulation of the recently introduced E-Product, often referred to as the Yat-kernel (Bouhsine, 2025). The resulting interactions are geometry-aware and inspired by inverse-square interactions in physics. Our method, Spherical Linearized Attention with Yat Kernels (SLAY), constrains queries and keys to the unit sphere so that attention depends only on angular alignment. Using Bernstein's theorem, we express the spherical Yat-kernel as a nonnegative mixture of polynomial-exponential product kernels and derive a strictly positive random-feature approximation enabling linear-time O(L) attention. We establish positive definiteness and boundedness on the sphere and show that the estimator yields well-defined, nonnegative attention scores. Empirically, SLAY achieves performance that is nearly indistinguishable from standard softmax attention while retaining linear time and memory scaling, and consistently outperforms prior linear-time attention mechanisms such as Performers and Cosformers. To the best of our knowledge, SLAY represents the closest linear-time approximation to softmax attention reported to date, enabling scalable Transformers without the typical performance trade-offs of attention linearization.

Multimodal Process Reward Models (MPRMs) are central to step-level supervision for visual reasoning in MLLMs. Training MPRMs typically requires large-scale Monte Carlo (MC)-annotated corpora, incurring substantial training cost. This paper studies the data efficiency for MPRM training. Our preliminary experiments reveal that MPRM training quickly saturates under random subsampling of the training data, indicating substantial redundancy within existing MC-annotated corpora. To explain this, we formalize a theoretical framework and reveal that informative gradient updates depend on two factors: label mixtures of positive/negative steps and label reliability (average MC scores of positive steps). Guided by these insights, we propose the Balanced-Information Score (BIS), which prioritizes both mixture and reliability based on existing MC signals at the rollout level, without incurring any additional cost. Across two backbones (InternVL2.5-8B and Qwen2.5-VL-7B) on VisualProcessBench, BIS-selected subsets consistently match and even surpass the full-data performance at small fractions. Notably, the BIS subset reaches full-data performance using only 10% of the training data, improving over random subsampling by a relative 4.1%.

Vision and language models (VLMs) are expected to analyse complex documents, such as those containing flowcharts, through a question-answering (QA) interface. The ability to recognise and interpret these flowcharts is in high demand, as they provide valuable insights unavailable in text-only explanations. However, developing VLMs with precise flowchart understanding requires large-scale datasets of flowchart images and corresponding text, the creation of which is highly time-consuming. To address this challenge, we introduce JSynFlow, a synthesised visual QA dataset for Japanese flowcharts, generated using large language models (LLMs). Our dataset comprises task descriptions for various business occupations, the corresponding flowchart images rendered from domain-specific language (DSL) code, and related QA pairs. This paper details the dataset's synthesis procedure and demonstrates that fine-tuning with JSynFlow significantly improves VLM performance on flowchart-based QA tasks. Our dataset is publicly available at https://huggingface.co/datasets/jri-advtechlab/jsynflow.

Developing expressive and responsive conversational digital humans is a cornerstone of next-generation human-computer interaction. While large language models (LLMs) have significantly enhanced dialogue capabilities, most current systems still rely on cascaded architectures that connect independent modules. These pipelines are often plagued by accumulated errors, high latency, and poor real-time performance. Lacking access to the underlying conversational context, these pipelines inherently prioritize rigid lip-sync over emotional depth. To address these challenges, we propose A$^2$-LLM, an end-to-end conversational audio avatar large language model that jointly reasons about language, audio prosody, and 3D facial motion within a unified framework. To facilitate training, we introduce FLAME-QA, a high-quality multimodal dataset designed to align semantic intent with expressive facial dynamics within a QA format. By leveraging deep semantic understanding, A$^2$-LLM generates emotionally rich facial movements beyond simple lip-synchronization. Experimental results demonstrate that our system achieves superior emotional expressiveness while maintaining real-time efficiency (500 ms latency, 0.7 RTF).

Many tool-based Retrieval Augmented Generation (RAG) systems lack precise mechanisms for tracing final responses back to specific tool components -- a critical gap as systems scale to complex multi-agent architectures. We present \textbf{Atomic Information Flow (AIF)}, a graph-based network flow model that decomposes tool outputs and LLM calls into atoms: indivisible, self-contained units of information. By modeling LLM orchestration as a directed flow of atoms from tool and LLM nodes to a response super-sink, AIF enables granular attribution metrics for AI explainability. Motivated by the max-flow min-cut theorem in network flow theory, we train a lightweight Gemma3 (4B parameter) language model as a context compressor to approximate the minimum cut of tool atoms using flow signals computed offline by AIF. We note that the base Gemma3-4B model struggles to identify critical information with \textbf{54.7\%} accuracy on HotpotQA, barely outperforming lexical baselines (BM25). However, post-training on AIF signals boosts accuracy to \textbf{82.71\%} (+28.01 points) while achieving \textbf{87.52\%} (+1.85\%) context token compression -- bridging the gap with the Gemma3-27B variant, a model nearly $7\times$ larger.

We introduce a simple initial working system in which relations (such as part-whole) are directly represented via an architecture with operating and learning rules fundamentally distinct from standard artificial neural network methods. Arbitrary data are straightforwardly encoded as graphs whose edges correspond to codes from a small fixed primitive set of elemental pairwise relations, such that simple relational encoding is not an add-on, but occurs intrinsically within the most basic components of the system. A novel graph-Hamiltonian operator calculates energies among these encodings, with ground states denoting simultaneous satisfaction of all relation constraints among graph vertices. The method solely uses radically low-precision arithmetic; computational cost is correspondingly low, and scales linearly with the number of edges in the data. The resulting unconventional architecture can process standard ANN examples, but also produces representations that exhibit characteristics of symbolic computation. Specifically, the method identifies simple logical relational structures in these data (part-of; next-to), building hierarchical representations that enable abductive inferential steps generating relational position-based encodings, rather than solely statistical representations. Notably, an equivalent set of ANN operations are derived, identifying a special case of embedded vector encodings that may constitute a useful approach to current work in higher-level semantic representation. The very simple current state of the implemented system invites additional tools and improvements.

Large Language Models (LLMs) are widely used for code generation. However, the correctness of code generated by LLMs remains a concern. A potential remedy to this concern is to have LLMs generate formal correctness proofs along with such code. However, compared with code generation, code-proof generation requires much higher reasoning capability and has much less existing data to learn from. In this paper, we present VeruSyn, a data synthesis pipeline for Verus, a state-of-the-art verification tool for system software written in Rust. Through self-synthesis and tutorial-based synthesis, VeruSyn achieves much larger scale and Verus-feature coverage than previous data-synthesis techniques designed for Verus; VeruSyn also supplements its dataset with long-chain-of-thought (CoT) data through agent trajectory synthesis. With VeruSyn, we synthesize the largest set of Verus verified programs: 6.9 million Rust programs, each with a formal specification and a proof that it meets that specification. This dataset lets us create a fine-tuned Qwen2.5-Coder-32B-Instruct model with appealing cost-proof tradeoff compared with state-of-the-art commercial models like Claude Sonnet 4.5. It also significantly outperforms models like o4-mini and previously proposed research models.

Direct Preference Optimization (DPO) and related methods align large language models from pairwise preferences by regularizing updates against a fixed reference policy. As the policy drifts, a static reference, however, can become increasingly miscalibrated, leading to distributional mismatch and amplifying spurious preference signals under noisy supervision. Conversely, reference-free variants avoid mismatch but often suffer from unconstrained reward drift. We propose Geometric Anchor Preference Optimization (GAPO), which replaces the fixed reference with a dynamic, geometry-aware anchor: an adversarial local perturbation of the current policy within a small radius that serves as a pessimistic baseline. This anchor enables an adaptive reweighting mechanism, modulating the importance of each preference pair based on its local sensitivity. We further introduce the Anchor Gap, the reward discrepancy between the policy and its anchor, and show under smoothness conditions that it approximates worst-case local margin degradation. Optimizing a logistic objective weighted by this gap downweights geometrically brittle instances while emphasizing robust preference signals. Across diverse noise settings, GAPO consistently improves robustness while matching or improving performance on standard LLM alignment and reasoning benchmarks.

Continuous-time generative models, such as diffusion models, flow matching, and rectified flow, learn time-dependent vector fields but are typically trained with objectives that treat timesteps independently, leading to high estimator variance and inefficient sampling. Prior approaches mitigate this via explicit smoothness penalties, trajectory regularization, or modified probability paths and solvers. We introduce Temporal Pair Consistency (TPC), a lightweight variance-reduction principle that couples velocity predictions at paired timesteps along the same probability path, operating entirely at the estimator level without modifying the model architecture, probability path, or solver. We provide a theoretical analysis showing that TPC induces a quadratic, trajectory-coupled regularization that provably reduces gradient variance while preserving the underlying flow-matching objective. Instantiated within flow matching, TPC improves sample quality and efficiency across CIFAR-10 and ImageNet at multiple resolutions, achieving lower FID at identical or lower computational cost than prior methods, and extends seamlessly to modern SOTA-style pipelines with noise-augmented training, score-based denoising, and rectified flow.

Causal discovery is often a data-driven paradigm to analyze complex real-world systems. In parallel, physics-based models such as ordinary differential equations (ODEs) provide mechanistic structure for many dynamical processes. Integrating these paradigms potentially allows physical knowledge to act as an inductive bias, improving identifiability, stability, and robustness of causal discovery in dynamical systems. However, such integration remains challenging: real dynamical systems often exhibit feedback, cyclic interactions, and non-stationary data trend, while many widely used causal discovery methods are formulated under acyclicity or equilibrium-based assumptions. In this work, we propose an integrative causal discovery framework for dynamical systems that leverages partial physical knowledge as an inductive bias. Specifically, we model system evolution as a stochastic differential equation (SDE), where the drift term encodes known ODE dynamics and the diffusion term corresponds to unknown causal couplings beyond the prescribed physics. We develop a scalable sparsity-inducing MLE algorithm that exploits causal graph structure for efficient parameter estimation. Under mild conditions, we establish guarantees to recover the causal graph. Experiments on dynamical systems with diverse causal structures show that our approach improves causal graph recovery and produces more stable, physically consistent estimates than purely data-driven state-of-the-art baselines.

We propose Laplacian In-context Spectral Analysis (LISA), a method for inference-time adaptation of Laplacian-based time-series models using only an observed prefix. LISA combines delay-coordinate embeddings and Laplacian spectral learning to produce diffusion-coordinate state representations, together with a frozen nonlinear decoder for one-step prediction. We introduce lightweight latent-space residual adapters based on either Gaussian-process regression or an attention-like Markov operator over context windows. Across forecasting and autoregressive rollout experiments, LISA improves over the frozen baseline and is often most beneficial under changing dynamics. This work links in-context adaptation to nonparametric spectral methods for dynamical systems.

Multimodal fusion faces two robustness challenges: noisy inputs degrade representation quality, and missing modalities cause prediction failures. We propose DCER, a unified framework addressing both challenges through dual-stage compression and energy-based reconstruction. The compression stage operates at two levels: within-modality frequency transforms (wavelet for audio, DCT for video) remove noise while preserving task-relevant patterns, and cross-modality bottleneck tokens force genuine integration rather than modality-specific shortcuts. For missing modalities, energy-based reconstruction recovers representations via gradient descent on a learned energy function, with the final energy providing intrinsic uncertainty quantification (\r{ho} > 0.72 correlation with prediction error). Experiments on CMU-MOSI, CMU-MOSEI, and CH-SIMS demonstrate state-of-the-art performance across all benchmarks, with a U-shaped robustness pattern favoring multimodal fusion at both complete and high-missing conditions. The code will be available on Github.

Feature steering has emerged as a promising approach for controlling LLM behavior through direct manipulation of internal representations, offering advantages over prompt engineering. However, its practical effectiveness in real-world applications remains poorly understood, particularly regarding potential trade-offs with output quality. We show that feature steering methods substantially degrade model performance even when successfully controlling target behaviors, a critical trade-off. Specifically, we evaluate Goodfire's Auto Steer against prompt engineering baselines across 14 steering queries (covering innocuous and safety-relevant behaviors) on 171 Massive Multitask Language Understanding (MMLU) questions using Llama-8B and Llama-70B, measuring accuracy, coherence, and behavioral control. Our findings show that Auto Steer successfully modifies target behaviors (achieving scores of 3.33 vs. 2.98 for prompting on Llama-8B and 3.57 vs. 3.10 on Llama-70B), but causes dramatic performance degradation: accuracy on the MMLU questions drops from 66% to 46% on Llama-8B and 87% to 73% on Llama-70B, with coherence falling from 4.62 to 2.24 and 4.94 to 3.89 respectively. Simple prompting achieves the best overall balance. These findings highlight limitations of current feature steering methods for practical deployment where task performance cannot be sacrificed. More broadly, our work demonstrates that mechanistic control methods face fundamental capability-behavior trade-offs that must be empirically characterized before deployment.

The Mechanistic Interpretability (MI) program has mapped the Transformer as a precise computational graph. We extend this graph with a conservation law and time-varying AC dynamics, viewing it as a physical circuit. We introduce Momentum Attention, a symplectic augmentation embedding physical priors via the kinematic difference operator $p_t = q_t - q_{t-1}$, implementing the symplectic shear $\hat{q}_t = q_t + γp_t$ on queries and keys. We identify a fundamental Symplectic-Filter Duality: the physical shear is mathematically equivalent to a High-Pass Filter. This duality is our cornerstone contribution -- by injecting kinematic momentum, we sidestep the topological depth constraint ($L \geq 2$) for induction head formation. While standard architectures require two layers for induction from static positions, our extension grants direct access to velocity, enabling Single-Layer Induction and Spectral Forensics via Bode Plots. We formalize an Orthogonality Theorem proving that DC (semantic) and AC (mechanistic) signals segregate into orthogonal frequency bands when Low-Pass RoPE interacts with High-Pass Momentum. Validated through 5,100+ controlled experiments (documented in Supplementary Appendices A--R and 27 Jupyter notebooks), our 125M Momentum model exceeds expectations on induction-heavy tasks while tracking a 350M baseline within $\sim$2.9% validation loss. Dedicated associative recall experiments reveal a scaling law $γ^* = 4.17 \times N^{-0.74}$ establishing momentum-depth fungibility. We offer this framework as a complementary analytical toolkit connecting Generative AI, Hamiltonian Physics, and Signal Processing.

Robust generalization under distribution shift remains difficult to monitor and optimize in the absence of target-domain labels, as models with similar in-distribution accuracy can exhibit markedly different out-of-distribution (OOD) performance. While prior work has focused on training-time regularization and low-order representation statistics, little is known about whether the geometric structure of learned embeddings provides reliable post-hoc signals of robustness. We propose a geometry-based diagnostic framework that constructs class-conditional mutual k-nearest-neighbor graphs from in-distribution embeddings and extracts two complementary invariants: a global spectral complexity proxy based on the reduced log-determinant of the normalized Laplacian, and a local smoothness measure based on Ollivier--Ricci curvature. Across multiple architectures, training regimes, and corruption benchmarks, we find that lower spectral complexity and higher mean curvature consistently predict stronger OOD accuracy across checkpoints. Controlled perturbations and topological analyses further show that these signals reflect meaningful representation structure rather than superficial embedding statistics. Our results demonstrate that representation geometry enables interpretable, label-free robustness diagnosis and supports reliable unsupervised checkpoint selection under distribution shift.

Predicting transcriptional responses to genetic perturbations is a central problem in functional genomics. In practice, perturbation responses are rarely gene-independent but instead manifest as coordinated, program-level transcriptional changes among functionally related genes. However, most existing methods do not explicitly model such coordination, due to gene-wise modeling paradigms and reliance on static biological priors that cannot capture dynamic program reorganization. To address these limitations, we propose scBIG, a module-inductive perturbation prediction framework that explicitly models coordinated gene programs. scBIG induces coherent gene programs from data via Gene-Relation Clustering, captures inter-program interactions through a Gene-Cluster-Aware Encoder, and preserves modular coordination using structure-aware alignment objectives. These structured representations are then modeled using conditional flow matching to enable flexible and generalizable perturbation prediction. Extensive experiments on multiple single-cell perturbation benchmarks show that scBIG consistently outperforms state-of-the-art methods, particularly on unseen and combinatorial perturbation settings, achieving an average improvement of 6.7% over the strongest baselines.

Large language models excel as few-shot learners when provided with appropriate demonstrations, yet this strength becomes problematic in multiturn agent scenarios, where LLMs erroneously mimic their own previous responses as few-shot examples. Through attention analysis, we identify conversational inertia, a phenomenon where models exhibit strong diagonal attention to previous responses, which is associated with imitation bias that constrains exploration. This reveals a tension when transforming few-shot LLMs into agents: longer context enriches environmental feedback for exploitation, yet also amplifies conversational inertia that undermines exploration. Our key insight is that for identical states, actions generated with longer contexts exhibit stronger inertia than those with shorter contexts, enabling construction of preference pairs without environment rewards. Based on this, we propose Context Preference Learning to calibrate model preferences to favor low-inertia responses over highinertia ones. We further provide context management strategies at inference time to balance exploration and exploitation. Experimental results across eight agentic environments and one deep research scenario validate that our framework reduces conversational inertia and achieves performance improvements.

As Generative AI (GenAI), particularly inference, rapidly emerges as a dominant workload category, the Kubernetes ecosystem is proactively evolving to natively support its unique demands. This industry paper demonstrates how emerging Kubernetes-native projects can be combined to deliver the benefits of container orchestration, such as scalability and resource efficiency, to complex AI workflows. We implement and evaluate an illustrative, multi-stage use case consisting of automatic speech recognition and summarization. First, we address batch inference by using Kueue to manage jobs that transcribe audio files with Whisper models and Dynamic Accelerator Slicer (DAS) to increase parallel job execution. Second, we address a discrete online inference scenario by feeding the transcripts to a Large Language Model for summarization hosted using llm-d, a novel solution utilizing the recent developments around the Kubernetes Gateway API Inference Extension (GAIE) for optimized routing of inference requests. Our findings illustrate that these complementary components (Kueue, DAS, and GAIE) form a cohesive, high-performance platform, proving Kubernetes' capability to serve as a unified foundation for demanding GenAI workloads: Kueue reduced total makespan by up to 15%; DAS shortened mean job completion time by 36%; and GAIE improved Time to First Token by 82\%.

We present a data poisoning attack -- Phantom Transfer -- with the property that, even if you know precisely how the poison was placed into an otherwise benign dataset, you cannot filter it out. We achieve this by modifying subliminal learning to work in real-world contexts and demonstrate that the attack works across models, including GPT-4.1. Indeed, even fully paraphrasing every sample in the dataset using a different model does not stop the attack. We also discuss connections to steering vectors and show that one can plant password-triggered behaviours into models while still beating defences. This suggests that data-level defences are insufficient for stopping sophisticated data poisoning attacks. We suggest that future work should focus on model audits and white-box security methods.

Inverse design with physics-based objectives is challenging because it couples high-dimensional geometry with expensive simulations, as exemplified by aerodynamic shape optimization for drag reduction. We revisit inverse design through two canonical solutions, the optimal design point and the optimal design distribution, and relate them to optimization and guided generation. Building on this view, we propose a new training loss for cost predictors and a density-gradient optimization method that improves objectives while preserving plausible shapes. We further unify existing training-free guided generation methods. To address their inability to approximate conditional covariance in high dimensions, we develop a time- and memory-efficient algorithm for approximate covariance estimation. Experiments on a controlled 2D study and high-fidelity 3D aerodynamic benchmarks (car and aircraft), validated by OpenFOAM simulations and miniature wind-tunnel tests with 3D-printed prototypes, demonstrate consistent gains in both optimization and guided generation. Additional offline RL results further support the generality of our approach.

Text-to-image (T2I) diffusion models are widely adopted for their strong generative capabilities, yet remain vulnerable to backdoor attacks. Existing attacks typically rely on fixed textual triggers and single-entity backdoor targets, making them highly susceptible to enumeration-based input defenses and attention-consistency detection. In this work, we propose Semantic-level Backdoor Attack (SemBD), which implants backdoors at the representation level by defining triggers as continuous semantic regions rather than discrete textual patterns. Concretely, SemBD injects semantic backdoors by distillation-based editing of the key and value projection matrices in cross-attention layers, enabling diverse prompts with identical semantic compositions to reliably activate the backdoor attack. To further enhance stealthiness, SemBD incorporates a semantic regularization to prevent unintended activation under incomplete semantics, as well as multi-entity backdoor targets that avoid highly consistent cross-attention patterns. Extensive experiments demonstrate that SemBD achieves a 100% attack success rate while maintaining strong robustness against state-of-the-art input-level defenses.

Activation steering is a practical post-training model alignment technique to enhance the utility of Large Language Models (LLMs). Prior to deploying a model as a service, developers can steer a pre-trained model toward specific behavioral objectives, such as compliance or instruction adherence, without the need for retraining. This process is as simple as adding a steering vector to the model's internal representations. However, this capability unintentionally introduces critical and under-explored safety risks. We identify a phenomenon termed Steering Externalities, where steering vectors derived from entirely benign datasets-such as those enforcing strict compliance or specific output formats like JSON-inadvertently erode safety guardrails. Experiments reveal that these interventions act as a force multiplier, creating new vulnerabilities to jailbreaks and increasing attack success rates to over 80% on standard benchmarks by bypassing the initial safety alignment. Ultimately, our results expose a critical blind spot in deployment: benign activation steering systematically erodes the "safety margin," rendering models more vulnerable to black-box attacks and proving that inference-time utility improvements must be rigorously audited for unintended safety externalities.

We study privacy amplification by synthetic data release, a phenomenon in which differential privacy guarantees are improved by releasing only synthetic data rather than the private generative model itself. Recent work by Pierquin et al. (2025) established the first formal amplification guarantees for a linear generator, but they apply only in asymptotic regimes where the model dimension far exceeds the number of released synthetic records, limiting their practical relevance. In this work, we show a surprising result: under a bounded-parameter assumption, privacy amplification persists even when releasing an unbounded number of synthetic records, thereby improving upon the bounds of Pierquin et al. (2025). Our analysis provides structural insights that may guide the development of tighter privacy guarantees for more complex release mechanisms.

Audio-visual video highlight detection aims to automatically identify the most salient moments in videos by leveraging both visual and auditory cues. However, existing models often underutilize the audio modality, focusing on high-level semantic features while failing to fully leverage the rich, dynamic characteristics of sound. To address this limitation, we propose a novel framework, Dual-Pathway Audio Encoders for Video Highlight Detection (DAViHD). The dual-pathway audio encoder is composed of a semantic pathway for content understanding and a dynamic pathway that captures spectro-temporal dynamics. The semantic pathway extracts high-level information by identifying the content within the audio, such as speech, music, or specific sound events. The dynamic pathway employs a frequency-adaptive mechanism as time evolves to jointly model these dynamics, enabling it to identify transient acoustic events via salient spectral bands and rapid energy changes. We integrate the novel audio encoder into a full audio-visual framework and achieve new state-of-the-art performance on the large-scale MrHiSum benchmark. Our results demonstrate that a sophisticated, dual-faceted audio representation is key to advancing the field of highlight detection.

The deployment of Large Language Models (LLMs) faces a critical bottleneck when handling lengthy inputs: the prohibitive memory footprint of the Key Value (KV) cache. To address this bottleneck, the token pruning paradigm leverages attention sparsity to selectively retain a small, critical subset of tokens. However, existing approaches fall short, with static methods risking irreversible information loss and dynamic strategies employing heuristics that insufficiently capture the query-dependent nature of token importance. We propose FASA, a novel framework that achieves query-aware token eviction by dynamically predicting token importance. FASA stems from a novel insight into RoPE: the discovery of functional sparsity at the frequency-chunk (FC) level. Our key finding is that a small, identifiable subset of "dominant" FCs consistently exhibits high contextual agreement with the full attention head. This provides a robust and computationally free proxy for identifying salient tokens. Building on this insight, FASA first identifies a critical set of tokens using dominant FCs, and then performs focused attention computation solely on this pruned subset. Across a spectrum of long-context tasks, from sequence modeling to complex CoT reasoning, FASA consistently outperforms all token-eviction baselines and achieves near-oracle accuracy, demonstrating remarkable robustness even under constraint budgets. Notably, on LongBench-V1, FASA reaches nearly 100\% of full-KV performance when only keeping 256 tokens, and achieves 2.56$\times$ speedup using just 18.9\% of the cache on AIME24.

While Large Reasoning Models (LRMs) have demonstrated impressive capabilities in solving complex tasks through the generation of long reasoning chains, this reliance on verbose generation results in significant latency and computational overhead. To address these challenges, we propose \textbf{CoSMo} (\textbf{Co}nsistency-Guided \textbf{S}plit-\textbf{M}erge \textbf{O}ptimization), a framework designed to eliminate structural redundancy rather than indiscriminately restricting token volume. Specifically, CoSMo utilizes a split-merge algorithm that dynamically refines reasoning chains by merging redundant segments and splitting logical gaps to ensure coherence. We then employ structure-aligned reinforcement learning with a novel segment-level budget to supervise the model in maintaining efficient reasoning structures throughout training. Extensive experiments across multiple benchmarks and backbones demonstrate that CoSMo achieves superior performance, improving accuracy by \textbf{3.3} points while reducing segment usage by \textbf{28.7\%} on average compared to reasoning efficiency baselines.

Growing renewable penetration introduces substantial uncertainty into power system operations, necessitating frequent adaptation of dispatch objectives and constraints and challenging expertise-intensive, near-real-time modeling workflows. Large Language Models (LLMs) provide a promising avenue for automating this process by translating natural-language (NL) operational requirements into executable optimization models via semantic reasoning and code synthesis. Yet existing LLM datasets and benchmarks for optimization modeling primarily target coarse-grained cross-domain generalization, offering limited, rigorous evaluation in power-system settings, particularly for Optimal Power Flow (OPF). We therefore introduce \textbf{ProOPF-D} and \textbf{ProOPF-B}, a dataset and benchmark for professional-grade OPF modeling: ProOPF-D contains 12K instances pairing NL requests with parameter adjustments and structural extensions to a canonical OPF, together with executable implementations; ProOPF-B provides 121 expert-annotated test cases with ground-truth code, enabling end-to-end evaluation under both concrete and abstract OPF modeling regimes.

Adaptive Neuro-Fuzzy Inference System (ANFIS) was designed to combine the learning capabilities of neural network with the reasoning transparency of fuzzy logic. However, conventional ANFIS architectures suffer from structural complexity, where the product-based inference mechanism causes an exponential explosion of rules in high-dimensional spaces. We herein propose the Kolmogorov-Arnold Neuro-Fuzzy Inference System (KANFIS), a compact neuro-symbolic architecture that unifies fuzzy reasoning with additive function decomposition. KANFIS employs an additive aggregation mechanism, under which both model parameters and rule complexity scale linearly with input dimensionality rather than exponentially. Furthermore, KANFIS is compatible with both Type-1 (T1) and Interval Type-2 (IT2) fuzzy logic systems, enabling explicit modeling of uncertainty and ambiguity in fuzzy representations. By using sparse masking mechanisms, KANFIS generates compact and structured rule sets, resulting in an intrinsically interpretable model with clear rule semantics and transparent inference processes. Empirical results demonstrate that KANFIS achieves competitive performance against representative neural and neuro-fuzzy baselines.

LLMs are increasingly used for code generation, but their outputs often follow recurring templates that can induce predictable vulnerabilities. We study \emph{vulnerability persistence} in LLM-generated software and introduce \emph{Feature--Security Table (FSTab)} with two components. First, FSTab enables a black-box attack that predicts likely backend vulnerabilities from observable frontend features and knowledge of the source LLM, without access to backend code or source code. Second, FSTab provides a model-centric evaluation that quantifies how consistently a given model reproduces the same vulnerabilities across programs, semantics-preserving rephrasings, and application domains. We evaluate FSTab on state-of-the-art code LLMs, including GPT-5.2, Claude-4.5 Opus, and Gemini-3 Pro, across diverse application domains. Our results show strong cross-domain transfer: even when the target domain is excluded from training, FSTab achieves up to 94\% attack success and 93\% vulnerability coverage on Internal Tools (Claude-4.5 Opus). These findings expose an underexplored attack surface in LLM-generated software and highlight the security risks of code generation. Our code is available at: https://anonymous.4open.science/r/FSTab-024E.

Discovering causal structures from multivariate time series is a key problem because interactions span across multiple lags and possibly involve instantaneous dependencies. Additionally, the search space of the dynamic graphs is combinatorial in nature. In this study, we propose \textit{Stable Causal Dynamic Differentiable Discovery (SC3D)}, a two-stage differentiable framework that jointly learns lag-specific adjacency matrices and, if present, an instantaneous directed acyclic graph (DAG). In Stage 1, SC3D performs edge preselection through node-wise prediction to obtain masks for lagged and instantaneous edges, whereas Stage 2 refines these masks by optimizing a likelihood with sparsity along with enforcing acyclicity on the instantaneous block. Numerical results across synthetic and benchmark dynamical systems demonstrate that SC3D achieves improved stability and more accurate recovery of both lagged and instantaneous causal structures compared to existing temporal baselines.

Membership Inference Attacks (MIAs) are widely used to quantify training data memorization and assess privacy risks. Standard evaluation requires repeated retraining, which is computationally costly for large models. One-run methods (single training with randomized data inclusion) and zero-run methods (post hoc evaluation) are often used instead, though their statistical validity remains unclear. To address this gap, we frame MIA evaluation as a causal inference problem, defining memorization as the causal effect of including a data point in the training set. This novel formulation reveals and formalizes key sources of bias in existing protocols: one-run methods suffer from interference between jointly included points, while zero-run evaluations popular for LLMs are confounded by non-random membership assignment. We derive causal analogues of standard MIA metrics and propose practical estimators for multi-run, one-run, and zero-run regimes with non-asymptotic consistency guarantees. Experiments on real-world data show that our approach enables reliable memorization measurement even when retraining is impractical and under distribution shift, providing a principled foundation for privacy evaluation in modern AI systems.

Large language models (LLMs) increasingly support reasoning over biomolecular structures, but most existing approaches remain modality-specific and rely on either sequence-style encodings or fixed-length connector tokens for structural inputs. These designs can under-expose explicit geometric cues and impose rigid fusion bottlenecks, leading to over-compression and poor token allocation as structural complexity grows. We present a unified all-atom framework that grounds language reasoning in geometric information while adaptively scaling structural tokens. The method first constructs variable-size structural patches on molecular graphs using an instruction-conditioned gating policy, enabling complexity-aware allocation of query tokens. It then refines the resulting patch tokens via cross-attention with modality embeddings and injects geometry-informed tokens into the language model to improve structure grounding and reduce structural hallucinations. Across diverse all-atom benchmarks, the proposed approach yields consistent gains in heterogeneous structure-grounded reasoning. An anonymized implementation is provided in the supplementary material.

We establish a theoretical connection between wavelet transforms and spiking neural networks through scale-space theory. We rely on the scale-covariant guarantees in the leaky integrate-and-fire neurons to implement discrete mother wavelets that approximate continuous wavelets. A reconstruction experiment demonstrates the feasibility of the approach and warrants further analysis to mitigate current approximation errors. Our work suggests a novel spiking signal representation that could enable more energy-efficient signal processing algorithms.

R1-style LLMs have attracted growing attention for their capacity for self-reflection, yet the internal mechanisms underlying such behavior remain unclear. To bridge this gap, we anchor on the onset of reflection behavior and trace its layer-wise activation trajectory. Using the logit lens to read out token-level semantics, we uncover a structured progression: (i) Latent-control layers, where an approximate linear direction encodes the semantics of thinking budget; (ii) Semantic-pivot layers, where discourse-level cues, including turning-point and summarization cues, surface and dominate the probability mass; and (iii) Behavior-overt layers, where the likelihood of reflection-behavior tokens begins to rise until they become highly likely to be sampled. Moreover, our targeted interventions uncover a causal chain across these stages: prompt-level semantics modulate the projection of activations along latent-control directions, thereby inducing competition between turning-point and summarization cues in semantic-pivot layers, which in turn regulates the sampling likelihood of reflection-behavior tokens in behavior-overt layers. Collectively, our findings suggest a human-like meta-cognitive process-progressing from latent monitoring, to discourse-level regulation, and to finally overt self-reflection. Our analysis code can be found at https://github.com/DYR1/S3-CoT.

The rapid emergence of autonomous large language model agents has given rise to persistent, large-scale agent ecosystems whose collective behavior cannot be adequately understood through anecdotal observation or small-scale simulation. This paper introduces data-driven silicon sociology as a systematic empirical framework for studying social structure formation among interacting artificial agents. We present a pioneering large-scale data mining investigation of an in-the-wild agent society by analyzing Moltbook, a social platform designed primarily for agent-to-agent interaction. At the time of study, Moltbook hosted over 150,000 registered autonomous agents operating across thousands of agent-created sub-communities. Using programmatic and non-intrusive data acquisition, we collected and analyzed the textual descriptions of 12,758 submolts, which represent proactive sub-community partitioning activities within the ecosystem. Treating agent-authored descriptions as first-class observational artifacts, we apply rigorous preprocessing, contextual embedding, and unsupervised clustering techniques to uncover latent patterns of thematic organization and social space structuring. The results show that autonomous agents systematically organize collective space through reproducible patterns spanning human-mimetic interests, silicon-centric self-reflection, and early-stage economic and coordination behaviors. Rather than relying on predefined sociological taxonomies, these structures emerge directly from machine-generated data traces. This work establishes a methodological foundation for data-driven silicon sociology and demonstrates that data mining techniques can provide a powerful lens for understanding the organization and evolution of large autonomous agent societies.

Differentially private synthetic data enables the sharing and analysis of sensitive datasets while providing rigorous privacy guarantees for individual contributors. A central challenge is to achieve strong utility guarantees for meaningful downstream analysis. Many existing methods ensure uniform accuracy over broad query classes, such as all Lipschitz functions, but this level of generality often leads to suboptimal rates for statistics of practical interest. Since many common data analysis queries exhibit smoothness beyond what worst-case Lipschitz bounds capture, we ask whether exploiting this additional structure can yield improved utility. We study the problem of generating $(\varepsilon,δ)$-differentially private synthetic data from a dataset of size $n$ supported on the hypercube $[-1,1]^d$, with utility guarantees uniformly for all smooth queries having bounded derivatives up to order $k$. We propose a polynomial-time algorithm that achieves a minimax error rate of $n^{-\min \{1, \frac{k}{d}\}}$, up to a $\log(n)$ factor. This characterization uncovers a phase transition at $k=d$. Our results generalize the Chebyshev moment matching framework of (Musco et al., 2025; Wang et al., 2016) and strictly improve the error rates for $k$-smooth queries established in (Wang et al., 2016). Moreover, we establish the first minimax lower bound for the utility of $(\varepsilon,δ)$-differentially private synthetic data with respect to $k$-smooth queries, extending the Wasserstein lower bound for $\varepsilon$-differential privacy in (Boedihardjo et al., 2024).

Current AI governance frameworks, including regulatory benchmarks for accuracy, latency, and energy efficiency, are built for static, centrally trained artificial neural networks on von Neumann hardware. NeuroAI systems, embodied in neuromorphic hardware and implemented via spiking neural networks, break these assumptions. This paper examines the limitations of current AI governance frameworks for NeuroAI, arguing that assurance and audit methods must co-evolve with these architectures, aligning traditional regulatory metrics with the physics, learning dynamics, and embodied efficiency of brain-inspired computation to enable technically grounded assurance.

Removing ''important'' high-gradient components from a neural network can improve generalization, while removing unimportant'' low-gradient components can destroy it. We demonstrate this paradox by formalizing the \textit{Gradient-Causal Gap} in Transformers trained on algorithmic tasks. While gradient magnitude and causal importance align on simple tasks ($ρ=0.73$ for reversal), this relationship collapses as task complexity increases ($ρ=0.32$ for sorting), sometimes becoming inverted ($ρ=-0.11$). Pruning experiments reveal that gradient magnitude is not merely inaccurate but \textit{unpredictably} so. Removing low-gradient ''Hidden Heroes'' consistently devastates OOD accuracy ($-32\%$). Removing high-gradient ''Gradient Bloats'' is a coin flip: harmless in most seeds (indicating optimization noise), catastrophic in others (indicating overfitting circuits). This unpredictability means gradient-based pruning cannot reliably preserve model capabilities.

Deep Reinforcement Learning (DRL), a subset of machine learning focused on sequential decision-making, has emerged as a powerful approach for tackling financial trading problems. In finance, DRL is commonly used either to generate discrete trade signals or to determine continuous portfolio allocations. In this work, we propose a novel reinforcement learning framework for portfolio optimization that incorporates Physics-Informed Kolmogorov-Arnold Networks (PIKANs) into several DRL algorithms. The approach replaces conventional multilayer perceptrons with Kolmogorov-Arnold Networks (KANs) in both actor and critic components-utilizing learnable B-spline univariate functions to achieve parameter-efficient and more interpretable function approximation. During actor updates, we introduce a physics-informed regularization loss that promotes second-order temporal consistency between observed return dynamics and the action-induced portfolio adjustments. The proposed framework is evaluated across three equity markets-China, Vietnam, and the United States, covering both emerging and developed economies. Across all three markets, PIKAN-based agents consistently deliver higher cumulative and annualized returns, superior Sharpe and Calmar ratios, and more favorable drawdown characteristics compared to both standard DRL baselines and classical online portfolio-selection methods. This yields more stable training, higher Sharpe ratios, and superior performance compared to traditional DRL counterparts. The approach is particularly valuable in highly dynamic and noisy financial markets, where conventional DRL often suffers from instability and poor generalization.

We introduce a reversible deep learning model for 13C NMR that uses a single conditional invertible neural network for both directions between molecular structures and spectra. The network is built from i-RevNet style bijective blocks, so the forward map and its inverse are available by construction. We train the model to predict a 128-bit binned spectrum code from a graph-based structure encoding, while the remaining latent dimensions capture residual variability. At inference time, we invert the same trained network to generate structure candidates from a spectrum code, which explicitly represents the one-to-many nature of spectrum-to-structure inference. On a filtered subset, the model is numerically invertible on trained examples, achieves spectrum-code prediction above chance, and produces coarse but meaningful structural signals when inverted on validation spectra. These results demonstrate that invertible architectures can unify spectrum prediction and uncertainty-aware candidate generation within one end-to-end model.

Decentralized Finance (DeFi) has turned blockchains into financial infrastructure, allowing anyone to trade, lend, and build protocols without intermediaries, but this openness exposes pools of value controlled by code. Within five years, the DeFi ecosystem has lost over 15.75B USD to reported exploits. Many exploits arise from permissionless opportunities that any participant can trigger using only public state and standard interfaces, which we call Anyone-Can-Take (ACT) opportunities. Despite on-chain transparency, postmortem analysis remains slow and manual: investigations start from limited evidence, sometimes only a single transaction hash, and must reconstruct the exploit lifecycle by recovering related transactions, contract code, and state dependencies. We present TxRay, a Large Language Model (LLM) agentic postmortem system that uses tool calls to reconstruct live ACT attacks from limited evidence. Starting from one or more seed transactions, TxRay recovers the exploit lifecycle, derives an evidence-backed root cause, and generates a runnable, self-contained Proof of Concept (PoC) that deterministically reproduces the incident. TxRay self-checks postmortems by encoding incident-specific semantic oracles as executable assertions. To evaluate PoC correctness and quality, we develop PoCEvaluator, an independent agentic execution-and-review evaluator. On 114 incidents from DeFiHackLabs, TxRay produces an expert-aligned root cause and an executable PoC for 105 incidents, achieving 92.11% end-to-end reproduction. Under PoCEvaluator, 98.1% of TxRay PoCs avoid hard-coding attacker addresses, a +22.9pp lift over DeFiHackLabs. In a live deployment, TxRay delivers validated root causes in 40 minutes and PoCs in 59 minutes at median latency. TxRay's oracle-validated PoCs enable attack imitation, improving coverage by 15.6% and 65.5% over STING and APE.

We present a unified information-theoretic framework elucidating the interplay between stability, privacy, and the generalization performance of quantum learning algorithms. We establish a bound on the expected generalization error in terms of quantum mutual information and derive a probabilistic upper bound that generalizes the classical result by Esposito et al. (2021). Complementing these findings, we provide a lower bound on the expected true loss relative to the expected empirical loss. Additionally, we demonstrate that $(\varepsilon, δ)$-quantum differentially private learning algorithms are stable, thereby ensuring strong generalization guarantees. Finally, we extend our analysis to dishonest learning algorithms, introducing Information-Theoretic Admissibility (ITA) to characterize the fundamental limits of privacy when the learning algorithm is oblivious to specific dataset instances.

Query Auto-Completion (QAC) suggests query completions as users type, helping them articulate intent and reach results more efficiently. Existing approaches face fundamental challenges: traditional retrieve-and-rank pipelines have limited long-tail coverage and require extensive feature engineering, while recent generative methods suffer from hallucination and safety risks. We present a unified framework that reformulates QAC as end-to-end list generation through Retrieval-Augmented Generation (RAG) and multi-objective Direct Preference Optimization (DPO). Our approach combines three key innovations: (1) reformulating QAC as end-to-end list generation with multi-objective optimization; (2) defining and deploying a suite of rule-based, model-based, and LLM-as-judge verifiers for QAC, and using them in a comprehensive methodology that combines RAG, multi-objective DPO, and iterative critique-revision for high-quality synthetic data; (3) a hybrid serving architecture enabling efficient production deployment under strict latency constraints. Evaluation on a large-scale commercial search platform demonstrates substantial improvements: offline metrics show gains across all dimensions, human evaluation yields +0.40 to +0.69 preference scores, and a controlled online experiment achieves 5.44\% reduction in keystrokes and 3.46\% increase in suggestion adoption, validating that unified generation with RAG and multi-objective alignment provides an effective solution for production QAC. This work represents a paradigm shift to end-to-end generation powered by large language models, RAG, and multi-objective alignment, establishing a production-validated framework that can benefit the broader search and recommendation industry.

We investigate the emergence of time quasicrystals (TQCs) in the open Dicke model, subjected to a quasi-periodic Fibonacci drive. TQCs are characterized by a robust sub-harmonic quasi-periodic response that is qualitatively distinct from the external drive. By directly analyzing the dynamics of the system in the thermodynamic limit, we establish the existence of TQC order in this system for a wide parameter regime. Remarkably, we demonstrate that this behavior persists even in the deep quantum regime with only two qubits. We systematically study the dependence of the TQC lifetime, $τ^{\ast}$, on the number of qubits and demonstrate that $τ^{\ast}$ increases monotonically with the system size. Our work demonstrates that quasi-periodically driven dissipative quantum systems can serve as a powerful platform for realizing novel non-equilibrium phases of matter.

The temperature range of the blue phases (BPs) formed in highly chiral mixtures based on cholesteryl oleyl carbonate (COC) and the nematic liquid crystal E7 was studied in the presence of various chemical structures. The $azo$ compounds used were of both chiral and achiral nature, and their molecular geometry was modified by substitution of modular "LEGO" molecular units of varying alkyl chain lengths and types of bridging groups, which could substantially affect the mesomorphic properties of the matrix mixture. It was shown that in many cases these dopants effectively broadened the BP temperature range. This effect depends on both the variation in the molecular geometry of the $azo$ compounds and the increase in the $cis$-isomer concentration under UV irradiation. The presence of the $cis$-isomers formed have a stronger impact on broadening the BP temperature range than the initial $trans$-isomers. These results demonstrate that the temperature range of BPs can be precisely controlled via a combination of molecular engineering and $trans$-$cis$ photo-isomerization.

We study non-equilibrium bacterial colony growth using a geometry-first, time-resolved analysis of morphology. From time-lapse microscopy data, we track the coupled evolution of area, perimeter, and boundary-sensitive shape descriptors along the full growth history. We find that non-equilibrium growth can exhibit extended intervals of compact area--perimeter scaling with exponent $α\approx 2$, consistent with growth governed by a single effective geometric length scale, as well as time-localized breakdowns of this scaling during ongoing growth. These breakdowns coincide with transient boundary reorganization while bulk area growth remains sustained. Our results demonstrate that visually distinct morphologies can arise within the same geometric growth regime, and that departures from single-scale behavior reflect intrinsic dynamical restructuring rather than growth arrest. More broadly, this work establishes time-resolved geometry as a coarse-grained framework for identifying when non-equilibrium growth departs from single-scale geometric constraints in living systems.

Understanding heat transport at the nanometer scale is critical for semiconductor devices, quantum materials, and thermal management of nanostructures, yet direct local measurements of thermal conductivity and heat capacity remain scarce. We developed a laser-excitation system integrated into a scanning transmission electron microscope (STEM) for nanoscale thermal transport measurements using ultra-high-resolution electron energy-loss spectroscopy (EELS). A fiber-coupled laser is introduced via a modified aperture mechanism, enabling flexible holder geometries and large tilt angles without optical elements in the polepiece gap. Synchronization of pulsed laser excitation with an externally gated direct electron detector provides temporal resolution about 50 ns at <10 meV energy resolution. Local temperatures are determined via the principle of detailed balance, and thermal transport parameters are extracted by fitting a forward-time central-space heat diffusion model including radiative losses. For amorphous carbon films, we obtain a thermal conductivity of 1.24 $\frac{W}{m\cdot K}$ and a heat capacity of 821 $\frac{J}{kg\cdot K}$, consistent with literature. This framework enables time-resolved nanoscale measurements of thermal transport in materials and devices.

Frustration on the triangular lattice has long been a source of intriguing and often debated phases in many-body systems. Although symmetry analysis has been employed, the role of the seemingly trivial parity symmetry has received little attention. In this work, we show that phases induced by frustration are systematically shaped by an implicit rule of thumb associated with spontaneous parity breaking. This principle enables us to anticipate and rationalize the regimes and conditions under which nontrivial phases emerge. For the spin-$S$ antiferromagnetic XXZ model, we demonstrate that a controversial parity-broken phase appears only at intermediate values of $S$. In bilayer systems, enhanced frustration leads to additional phases, such as supersolids, whose properties can be classified by their characteristic parity features. Benefiting from our improved tensor network contraction techniques, we confirm these results through large-scale tensor-network calculations. This study offers an alternative viewpoint and a systematic approach for examining the interplay between spin, symmetry, and frustration in many-body systems.

Improving the computational efficiency of quantum many-body calculations from a hardware perspective remains a critical challenge. Although field-programmable gate arrays (FPGAs) have recently been exploited to improve the computational scaling of algorithms such as Monte Carlo methods, their application to tensor network algorithms is still at an early stage. In this work, we propose a fine-grained parallel tensor network design based on FPGAs to substantially enhance the computational efficiency of two representative tensor network algorithms: the infinite time-evolving block decimation (iTEBD) and the higher-order tensor renormalization group (HOTRG). By employing a quad-tile partitioning strategy to decompose tensor elements and map them onto hardware circuits, our approach effectively translates algorithmic computational complexity into scalable hardware resource utilization, enabling an extremely high degree of parallelism on FPGAs. Compared with conventional CPU-based implementations, our scheme exhibits superior scalability in computation time, reducing the bond-dimension scaling of the computational cost from $O(D_b^3)$ to $O(D_b)$ for iTEBD and from $O(D_b^6)$ to $O(D_b^2)$ for HOTRG. This work provides a theoretical foundation for future hardware implementations of large-scale tensor network computations.

We demonstrate an alternative approach based on complex-valued inverse temperature and partition function to probe quantum phases of matter with nontrivial spectra and dynamics. It leverages thermofield dynamics (TFD) to quantitatively characterize quantum and thermal fluctuations, and exploit the correspondence between low-energy excitations and Fisher zeros. Using the quantum XY chain in an external field as a testbed, we show that the oscillatory gap behavior manifests as oscillations in the long-time dynamics of the TFD spectral form factor. We also identify giant bubbles, i.e. large-scale closed lines, of Fisher-zeros near the gapless XX limit. They provide a characteristic energy scale that seems to contradict the predictions of the low energy theory of a featureless Luttinger liquid. We identify this energy scale and relate the motion of these giant bubbles with varying external field to the transfer of spectral weight from high to low energies. The deep connection between Fisher zeros, dynamics, and excitations opens up promising avenues for understanding the unconventional gap behaviors in strongly correlated many-body systems.

AMDAT (Amorphous Molecular Dynamics Analysis Toolkit) is an open-source C++ toolkit for post-processing molecular dynamics trajectories, focused on high-performance static and dynamic analyses of amorphous, glassy, and polymer materials, including supercooled liquids and complex fluids. In this paper, we describe AMDAT's design for efficient long-timescale analysis via in-memory trajectory handling and exponential time sampling, and we demonstrate representative workflows for widely used observables such as radial distribution functions, structure factors, intermediate scattering functions, and neighbor correlations.

We present a nonperturbative analysis of the weak- and strong-disorder regimes of the continuous random-field Ising model using the distributional zeta-function method. By performing the quenched-disorder average at the level of the effective action, we derive exact quadratic and interaction terms. In the weak-disorder limit, we show that the infrared structure of the two-point correlation functions yields a decomposition of the physical field into correlated components with distinct scaling dimensions. This mechanism exhibits the characteristic $1/p^4$ behavior, which shifts the upper critical dimension to $d_c^{+}=6$. The universal critical behavior of the RFIM near this dimension is governed by a minimal infrared effective action. In the strong-disorder regime, we obtain an exact diagonal quadratic action with a discrete spectrum of massive modes. Here, the absence of massless modes implies the absence of conventional criticality. The resulting spectral representation of correlation functions converges rapidly and remains well controlled in the infrared regime.

Statistical functions such as the moment-generating function, characteristic function, cumulant-generating function, and second characteristic function are cornerstone tools in classical statistics and probability theory. They provide a powerful means to analyze the statistical properties of a system and find applications in diverse fields, including statistical physics and field theory. While these functions are ubiquitous in classical theory, a quantum counterpart has remained elusive due to the fundamental hurdle of noncommutativity of operators. The lack of such a framework has obscured the deep connections between standard statistical measures and the non-classical features of quantum mechanics. Here, we establish a comprehensive framework for quantum statistical functions that transcends these limitations, naturally unifying the disparate languages of standard quantum statistics, quasiprobability distributions, and weak values. We show that these functions, defined as expectation values with respect to the purified state, naturally reproduce fundamental quantum statistical quantities like expectation values, variance, and covariance upon differentiation. Crucially, by extending this framework to include the concepts of pre- and post-selection, we define conditional quantum statistical functions that uniquely yield weak values and weak variance. We further demonstrate that multivariable quantum statistical functions, when defined with specific operator orderings, correspond to well-known quasiprobability distributions. Our framework provides a cohesive mathematical structure that not only reproduces standard quantum statistical measures but also incorporates nonclassical features of quantum mechanics, thus laying the foundation for a deeper understanding of quantum statistics.

We investigate the nonequilibrium dynamics induced by an interaction quench in the semiclassical Holstein model within the Ehrenfest nonadiabatic framework, which describes an isolated hybrid quantum-classical system with strictly conserved total energy. Focusing on the half-filled case, where the equilibrium ground state exhibits commensurate charge-density-wave (CDW) order for any nonzero coupling, we identify three distinct post-quench dynamical regimes as a function of the final electron-phonon coupling: a nonequilibrium metallic state without CDW order, an intermediate regime characterized by slow scale-invariant ordering dynamics, and a frozen CDW state with arrested coarsening and immobile kinks. We analyze the intermediate regime in detail and uncover an unconventional algebraic decay of the kink density, $n \sim t^{-1/3}$, distinct from both ballistic annihilation and diffusive coarsening in classical dissipative systems. We show that this anomalous exponent arises from the hybrid nature of the dynamics: while the lattice evolves deterministically, the electronic degrees of freedom act as an effective internal bath that induces diffusive kink motion without energy dissipation. An effective reaction-diffusion description, incorporating both annihilation and elastic scattering of kinks, quantitatively accounts for the observed scaling behavior. Our results reveal a distinct coarsening mechanism in isolated hybrid systems, demonstrating how internal quantum dynamics can qualitatively reshape defect kinetics far from equilibrium.

Effect of internal chirality on collective motion of large number of active objects is studied by simulations of appropriately modified Vicsek model. We add a fixed angle to the noise and consider small ratios, p, between this angle and the maximal deviation from the average local direction of motion. When the above ratio is p=1/120, the traveling bands observed with the symmetrical noise are destroyed, and small bands moving in different directions appear. Circular rotating flocks of objects with the same orientation are formed for p=1/7.5. Stable vortexes in the stationary state were found from p=1/60 to p=1/20. Velocity autocorrelation function shows equilibrium between the inflow and the outflow to and from the vortex. Long-time evolution is significantly influenced by a temporary trapping of the objects in the vortex. The ballistic behavior for the symmetrical noise changes to the diffusive behavior for the chirality leading to the onset of vortexes.

The mechanisms of flow in suspensions of soft particles above the glass-transition volume fraction and in the jammed state were probed using Orthogonal Superposition Rheometry (OSR). A small amplitude oscillatory shear flow is superimposed orthogonally onto a steady shear flow, which allows monitoring the viscoelastic spectra of sheared jammed core-shell microgels during flow. The characteristic crossover frequency ωc, deduced from the viscoelastic spectrum, provides information about the shear induced structural relaxation time, which is connected to the microscopic yielding mechanism of cage breaking. The shear rate evolution of the crossover frequency is used to achieve a superposition of all spectra and get a better insight of the flow mechanism. Despite their inherent softness, the hybrid core-shell microgels exhibit similarities with hard sphere-like flow behavior, with the main difference that for the microgels the transition from a glassy to a jammed state introduces a volume fraction dependence of the scaling of ωc with shear rate. We further check the application of the Kramers-Kronig relations on the experimental low strain amplitude OSR data finding a good agreement. Finally, the low frequency response at high strain rates was investigated with open bottom cell geometry and instrumental limits were identified. Based on these limits, we discuss previous OSR data and findings in repulsive and attractive colloidal glasses, and compared them with the current soft particle gels.

In this study, we show that a $d$-wave altermagnet can be transformed into a Chern insulator by irradiating it with elliptically polarized light from a high-frequency photon beam. We further explore the intrinsic anomalous thermoelectric and thermal Hall effects in light-irradiated altermagnets. At extremely low temperatures, the thermoelectric Hall coefficient, which exhibits a linear temperature dependence for the thermoelectric Hall conductivity, vanishes within the gapped region between the conduction and valence bands. However, it displays peaks and dips at the boundaries of the gap, suggesting that thermoelectric Hall conductivity can be used to probe the bandwidth. Similarly, the low-temperature thermal Hall coefficient, which also shows a linear temperature dependence for the thermal Hall conductivity, becomes quantized in the gapped region between the conduction and valence bands. This quantization indicates that thermal Hall conductivity can serve as a probe for the topological properties of the system.

Hyperbolic phonon polaritons (HPPs) in hexagonal boron nitride (hBN) confine mid-infrared light to deep-subwavelength scales and may offer a powerful route to strong light-matter interactions. Generation and control of HPPs are typically accessed using classical near-field probes, which limits experiments at the quantum level.A complementary frontier in hBN research focuses on color centers: bright, stable, atomically localized emitters that have rapidly emerged as a promising platform for solid-state quantum optics. Here we establish a key connection between these two directions by developing a cavity-QED framework in which a single hBN color center serves as a quantum source of HPPs. We quantify the emitter-HPP interaction and analyze two generation schemes. The first is spontaneous emission into the phonon sideband, which can produce single-HPP events and, in ultrathin slabs, becomes single-mode with an enhanced decay rate. The second is a stimulated Raman process that provides frequency selectivity, tunable conversion rate, and narrowband excitation. This drive launches spatially confined, ray-like HPPs that propagate over micrometer distances. We also outline a two-emitter correlation measurement that can directly test the single-polariton character of these emissions. By connecting color-center quantum optics with hyperbolic polaritonics, our approach enables quantum emitters to act as on-chip quantum sources and controls for HPPs, while HPPs provide long-range channels that couple spatially separated emitters. Together, these capabilities point to a new direction for mid-infrared photonic experiments that unite strong coupling, spectral selectivity, and spatial reach within a single material system.

The mechanical properties of confining boundaries can fundamentally alter the flow behaviour of shear-thickening suspensions. We study a dense cornstarch suspension sheared beneath a viscous silicone-oil layer, using the oil viscosity to tune boundary compliance. Flow visualisation and rheometry reveal two distinct regimes. With compliant boundaries, long-lived heterogeneities emerge via density waves or persistent clusters, maintained by a balance between interface deformation and particle rearrangement. With more resistant confinement, we observe transient jamming events, marked by abrupt spanning of load-bearing structures across the suspension thickness and the emergence of secondary stress waves. The onset stress of these events remains constant at the DST threshold, independent of bounding viscosity. Our results reveal that boundary compliance selects the lifetime and morphology of heterogeneous structures, offering a means to amplify otherwise short-lived microscopic processes and providing new insight into the interplay between shear thickening, shear jamming, and confinement mechanics.

Constriction-based spin Hall nano-oscillators (SHNOs) show great promise for application as highly tunable microwave sources with straightforward scalability toward large coupled networks. However, details of the magnetization dynamics within SHNOs have thus far not been addressed experimentally, due to the minute time and length scales involved. In this work, we present direct imaging of the magnetization dynamics within a single CoFeB-based SHNO using time-resolved scanning transmission X-ray microscopy (STXM). Our measurements reveal that the magnon amplitude is the strongest at the two constriction edges, with a pronounced assymetry favoring one edge, and that emitted spin waves exhibit strongly anisotropic propagation. Micromagnetic simulations suggest that grain boundaries and the Dzyaloshinskii-Moriya interaction (DMI) play a key role in both effects. Furthermore, the magnetodynamics changed during the measurement, indicating that the CoFeB/MgO interface may be more susceptible to X-ray induced modifications than previously recognized, challenging its presumed radiation hardness.

Weakly coupled oscillators adjust their dynamics to work in unison: they synchronize. This ubiquitous phenomenon is observed for oscillating pendulum, electronic devices, as well as clapping crowds or flashing fireflies. In effect, synchronization constitutes an efficient mean to translate microscopic into large scale dynamics. While broadly studied theoretically, experimental investigations of synchronization of systems at the microscale are limited. Here we devise and study a model system of noisy and "measurably imperfect" colloidal oscillators: autonomous clocks made of an active swimmer revolving around a passive sphere. The distribution of natural frequency of the clock is achieved using passive spheres of various sizes, thus without altering the (phoretic) coupling between clocks. We observe that pairs of oscillators lock phases before slipping and returning to sync, and we characterize the synchronicity of the pair. We rationalize our findings with a stochastic model, formalizing synchronization as a classical Kramers escape problem in an adequate potential. This provides an analytical expression for the rate of synchronization of a pair set by the ratio between differences of natural frequency and environmental noise, and agrees qualitatively with the experiment. Our results set a blueprint for synchronization with micrometric autonomous systems.

Despite decades of research, the microscopic origin of sublattice melting and fast ion transport in superionic conductors remains elusive. Here, we introduce a chemically neutral minimal binary model consisting of a rigid host lattice stabilized by short-range steric repulsion and a soft carrier sublattice interacting via long-range Wigner-type forces. This contrast naturally produces distinct melting temperatures and an intermediate sublattice-melting phase in which carriers become fluidlike while the host remains crystalline. Molecular-dynamics simulations identify three dynamical regimes-crystalline, sublattice-melt, and fully molten-marked by sharp changes in diffusivity, structural correlations, and dynamic heterogeneity. Near sublattice melting, carrier motion is strongly anharmonic and spatially heterogeneous, beyond mean-field hopping descriptions. By tuning the density, we demonstrate that sublattice melting can be continuously controlled, establishing a direct link between lattice softness, anharmonicity, and collective ion transport. This work provides a unified microscopic foundation for designing mechanically robust, high-performance superionic conductors operable near ambient conditions.

We study a coupled driven system where two different species of particles, along with some vacancies or holes, move on a landscape whose shape fluctuates with time. The movement of the particles is guided by the local shape of the landscape, and this shape is also affected by the presence of different particle species. When a particle species push the landscape in the same (opposite) direction of its own motion, it is called an aligned (a reverse) bias. Aligned bias promotes ordering while reverse bias destroys it. In absence of vacancies, the system reduces to previously studied LH model with different kinds of ordered and disordered phases which could be explained as a competition or cooperation between aligned bias and reverse bias. This interplay is expected to remain unaffected even when vacancies are present since vacancies do not impart any kind of bias on the landscape. However, we find presence of vacancies effectively weakens the reverse bias and this significantly changes the outcome of the competition between the two bias types. As a result novel ordered phases emerge which were not seen before. We analytically calculate the new phase boundaries within mean field approximation. We show even when aligned bias is weaker than reverse bias, it is possible to find long range order in the system. We discover two new phases where particle species showing weak aligned bias phase separate and the other species with strong reverse bias stays mixed with the vacancies. We call these phases finite current with partial phase separation (FPPS) and vacancy induced phase separation (VIPS). The landscape beneath the phase separated species takes the form of a macroscopic hill or valley in FPPS phase. But in VIPS phase it has the shape like a plateau whose height scales as square root of system size. The landscape in the remaining part of the system is disordered in both these phases.

We present a novel Exchange Monte Carlo (EMC) method designed for application in continuous-space Path Integral Monte Carlo (PIMC) simulations at finite temperature. Traditional PIMC methods for bosonic systems suffer from long autocorrelation times, particularly when measuring observables affected by particle permutations, such as the winding number. To address this issue, we introduce an exchange update scheme that facilitates replica transitions between different interaction regimes, significantly accelerating Monte Carlo dynamics-especially for global observables sensitive to permutation effects. Furthermore, we incorporate Stochastic Potential Switching (SPS) to efficiently decompose interactions, substantially enhancing computational efficiency for long-range interatomic pair potentials such as the Lennard-Jones and Aziz potentials.

We predict the formation of ferron-polariton - a hybrid light-matter quasiparticle arising from the coupling between collective ferroelectric excitations (ferrons) and Swihart photons in a superconductor/ferroelectric/superconductor heterostructure. The coupling provides direct evidence for ferrons and reaches the ultrastrong-coupling regime, with a spectral gap in the terahertz range, orders of magnitude larger than those in magnetic analogues, reflecting the superior strength of electric dipole interactions. Our work establishes superconductor-ferroelectric heterostructures as a novel platform for exploring extreme light-matter coupling and for developing high-speed, ferroelectric-based quantum technologies at terahertz frequencies.

Characterizing the ground-state properties of disordered systems, such as spin glasses and combinatorial optimization problems, is fundamental to science and engineering. However, computing exact ground states and counting their degeneracies are generally NP-hard and #P-hard problems, respectively, posing a formidable challenge for exact algorithms. Recently, Tensor Networks methods, which utilize high-dimensional linear algebra and achieve massive hardware parallelization, have emerged as a rapidly developing paradigm for efficiently solving these tasks. Despite their success, these methods are fundamentally constrained by the exponential growth of space complexity, which severely limits their scalability. To address this bottleneck, we introduce the Branch-and-Bound Tensor Network (BBTN) method, which seamlessly integrates the adaptive search framework of branch-and-bound with the efficient contraction of tropical tensor networks, significantly extending the reach of exact algorithms. We show that BBTN significantly surpasses existing state-of-the-art solvers, setting new benchmarks for exact computation. It pushes the boundaries of tractability to previously unreachable scales, enabling exact ground-state counting for $\pm J$ spin glasses up to $64 \times 64$ and solving Maximum Independent Set problems on King's subgraphs up to $100 \times 100$. For hard instances, BBTN dramatically reduces the computational cost of standard Tropical Tensor Networks, compressing years of runtime into minutes. Furthermore, it outperforms leading integer-programming solvers by over 30$\times$, establishing a versatile and scalable framework for solving hard problems in statistical physics and combinatorial optimization.

Non-reciprocal microwave components are indispensable in quantum information processing and cryogenic measurement. Conventional implementations, however, are bulky and incompatible with on-chip scalable integration. Recent efforts to develop compact on-chip alternatives often rely on active modulation or complex circuit architectures, which introduce additional losses and degrade performance. We demonstrate the realization of compact, self-impedance-matched gyrators based on edge magnetoplasmons in a two-dimensional electron gas. Gyrators can be used as building blocks for other non-reciprocal elements such as isolators and circulators. Our devices achieve gyration from 0.2 to 2 GHz, tunable by moderate out-of plane magnetic fields below 400 mT, and sub-mm footprint, two orders of magnitude smaller than conventional ferrite-based components. Using an electrode geometry predicted to minimize reflections, we achieve insertion losses as low as 2 to 4 dB. The self-matched design framework we utilize is broadly applicable, and can be implemented in a wide variety of non-reciprocal device architectures.

In this paper, we provide an accurate description of geometric phases emerging in simple optical systems -- nanostructures (metaatoms) interacting with vortex beams. We show that this interaction leads to a new class of geometric phase for optical beams, which is different from the geometric phases commonly discussed in structured-light optics. We compare this setting to the usual description of geometric phases for beams and show that the underlying geometry is different.

Ferroelectric metals, characterized by the coexistence of ferroelectricity and metallic conductivity, present a fundamental challenge due to the screening effect of free charge carriers on the long-range electric dipole order. Existing strategies to circumvent this obstacle include employing two-dimensional (2D) crystals, where reduced dimensionality and low carrier densities suppress screening, or designing materials of van der Waals (vdW) superlattice with spatially separated and decoupled conductive and nearly insulating ferroelectric layers. Here, we report an alternative paradigm in TaNiTe5, where an ultrahigh carrier density coexists with an out-of-plane ferroelectric order within the same surface monolayer. Using piezoresponse force microscopy (PFM), we observed robust ferroelectric behavior in TaNiTe5 down to single-unit-cell thickness (~1.3 nm) at room temperature. Scanning transmission electron microscopy (STEM) gives structural evidence that the ferroelectricity might originate from the vertical displacement of outmost Te atoms on the surface, breaking the inversion symmetry. Concurrently, electrical transport measurements reveal a metallic state with a carrier density on the order of 10$^{15}$ cm$^{-2}$ (or 10$^{22}$ cm$^{-3}$) -- comparable to that of Copper (Cu). Our findings establish a unique platform for exploring the interplay between ferroelectricity and an ultrahigh density of mobile carriers in the 2D limit.

Ta4SiTe4 is a one-dimensional van der Waals material that exhibits an exceptionally large thermoelectric power factor below room temperature. However, since this material has been available only in the form of acicular microcrystals, experimental exploration of the electronic properties responsible for its giant thermoelectric performance has long been challenging. In this study, we quantitatively evaluated the one-dimensional electronic nature of Ta4SiTe4 by combining micro-spot angle-resolved photoemission spectroscopy and transport measurements on focused-ion-beam-processed samples. The angle-resolved photoemission spectroscopy measurements reveal anisotropic band dispersions along and perpendicular to the crystallographic c axis. Consistently, transport measurements demonstrate that the resistivity perpendicular to the c axis is approximately five times larger than that along the c axis at 200 K. These results provide direct experimental evidence for the quasi-one-dimensional electronic character of Ta4SiTe4, which underlies its giant thermoelectric response reported previously, and offer fundamental insights into the role of electronic dimensionality in enhancing thermoelectric performance.

We study the interplay between quantum transport and topology in a one-dimensional off-diagonal commensurate Aubry-André-Harper (AAH) chain. The model, formulated within AAH framework, effectively represents a one-dimensional lattice with two competing commensurate modulations, supporting two distinct types of topological edge modes: zero-energy states in the central gapless region and quantum Hall (QH) edge states bridging the gapped bulk bands. These edge modes govern the transport behavior and give rise to sharp variations in transmission across the corresponding gap-closing transitions. A pronounced even-odd effect further emerges, where chains with an odd number of sites exhibit nearly perfect zero-energy transmission at the Dirac points, independent of system-lead coupling, system size, or modulation strength; a robust signature of ballistic transport. To capture the influence of environmental decoherence, we also incorporate Büttiker dephasing probes, which enable a phenomenological description of inelastic scattering and reveal how dephasing modifies, and in some regimes enhances, coherent transport.

Spin current offers a promising route toward energy-efficient and high-speed information processing. Developing efficient methods for their generation remains a central challenge in spintronics. Here, we investigate spin current generation via two-dimensional topological spin textures: a skyrmion crystal (SkX) with out-of-plane magnetization, a bimeron crystal (BmX) with in-plane magnetization, and a meron crystal (MX) with zero net magnetization. We show that these distinct spin textures generate spin currents with characteristic spin polarization directions. In the absence of spin--orbit coupling, the SkX and BmX generate spin currents polarized along their magnetization directions, whereas the MX yields no spin current. Upon introducing spin--orbit coupling, while the behavior of the SkX does not qualitatively change, the BmX generates nonzero spin currents in multiple polarization directions. Notably, the MX, despite its zero net magnetization, exhibits a pronounced spin current with out-of-plane spin polarization, driven by an enhanced spin Berry curvature associated with characteristic band degeneracy. We further demonstrate that the electronic and spin transport properties of each texture are governed by their magnetic symmetries. Our results highlight the topological spin textures as efficient sources of spin current even without net magnetization, expanding the design for spintronics devices based on topological magnetic metals.

Establishing reliable and interpretable structure-property relationships in glasses is a longstanding challenge in condensed matter physics. While modern data-driven machine learning techniques have proven highly effective in establishing structure-property correlations, many models are criticized for lacking physical interpretability and being task-specific. In this work, we identify an approximate linear relation between structure profiles and disorder-induced responses of glass properties based on first order perturbation theory. We analytically demonstrate that this relationship holds universally across glassy systems with varying dimensions and distinct interaction types. This robust theoretical relationship motivates the adoption of linear machine learning models, which we show numerically to achieve surprisingly high predictive accuracy for structure-property mapping in a wide variety of glassy materials. We further devise regularization analysis to further enhance the interpretability of our model, bridging the gap between predictive performance and physical insight. Overall, this linear relation establishes a simple yet powerful connection between structural disorder and spectral properties in glasses, opening a new avenue for advancing their studies.

We present a unified extension of two sets of criteria for high-fugacity crystallization in hard-core lattice systems developed previously by Jauslin, Lebowitz, and the author. Our new criterion is formulated in terms of the existence of a volume allocation rule with desirable properties, in analogy to the scoring function constructed in Hales' proof of the Kepler conjecture. The proof uses a recent systematic extension of Pirogov--Sinai theory to systems with infinite interactions by Mazel--Stuhl--Suhov. Notably, our result applies to a large class of polyomino models with discrete rotational degrees of freedom and chiral mixtures thereof.

We propose a tensor-network (TN) approach for solving classical optimization problems that is inspired by spectral filtering and sampling on quantum states. We first shift and scale an Ising Hamiltonian of the cost function so that all eigenvalues become non-negative and the ground states correspond to the the largest eigenvalues, which are then amplified by power iteration. We represent the transformed Hamiltonian as a matrix product operator (MPO) and form an immense power of this object via truncated MPO-MPO contractions, embedding the resulting operator into a matrix product state for sampling in the computational basis. In contrast to the density-matrix renormalization group, our approach provides a straightforward route to systematic improvement by increasing the bond dimension and is better at avoiding local minima. We also study the performance of this power method in the context of a higher-order Ising Hamiltonian on a heavy-hexagonal lattice, making a comparison with simulated annealing. These results highlight the potential of quantum-inspired algorithms for solving optimization problems and provide a baseline for assessing and developing quantum algorithms.

Universality classes encompass the analogous thermodynamic behavior of unlike physical systems, at different spatial dimensions $d$, in the vicinity of their critical point. Critical exponents define these classes, with the Ising model being the outstanding prototype that elucidates the differences from the mean-field category, believed to be valid above a critical dimension only. Here, in apparent striking contradiction to the Ising universality class, we demonstrate that the critical behavior of a finite Ising system of $N$ spins in $d = 3$ obeys mean-field Landau theory in the vicinity of its critical point, with classical critical exponents. Yet, when expressed in terms of the linear size $L$ of the system, the free energy unveils its proper finite-size scaling form, from which the thermodynamic limit critical temperature $T_c$ and the Ising critical exponents $ν$, $γ$ and $β$ can be identified. We find that the larger the size $L$, the smaller the mean-field region, shrinking to zero in the thermodynamic limit. These conclusions are achieved via the use of an alternative approach to collect data from a Monte Carlo simulation of a three-dimensional Ising model that allows for the evaluation of the free energy per spin $f = f(T,m;L)$ and of the coexistence curve, or spontaneous magnetization at zero magnetic field, $m_{\rm coex} = m(T;L)$ as functions of temperature $T$ and magnetization per spin $m = M/N$. Our results suggest a revision of the role of mean-field theory in the elucidation of critical phenomena.

Vertical integration of superconducting electronics requires fabrication strategies that preserve pristine interfaces while accommodating oxidation-sensitive elemental superconductors. However, existing van der Waals-based vertical Josephson junctions largely rely on transfer-based assembly schemes that are incompatible with elemental materials such as niobium (Nb). Here, we introduce a freestanding van der Waals membrane architecture that enables deposition-based fabrication of vertical Josephson junctions through double-sided processing of a single suspended two-dimensional layer. Using multilayer graphene suspended across lithographically defined through-holes in a SiNx membrane, we realize vertical Nb/multilayer graphene/Nb Josephson junctions without ambient exposure of buried interfaces. The resulting devices exhibit clear Josephson coupling, including reproducible supercurrents and a temperature dependence of the critical current consistent with short-junction behaviour. Well-defined magnetic interference patterns governed by the membrane-defined aperture geometry, together with sub-gap features that track a Bardeen-Cooper-Schrieffer (BCS)-like superconducting gap, further confirm the junction quality. This platform establishes a scalable route to vertical superconducting devices based on oxidation-sensitive elemental superconductors and van der Waals materials.

The critical incompatibility of polymers with different compositions scales inversely with their length. For instance, a mixture of A and B homopolymers of length $N$ segregates at $χ_{AB}^{cr} = 2/N$. But what if the difference between the blend components is subtler? We demonstrate that a mixture of AB copolymers with identical composition -- equal amounts of A and B monomers -- but different primary sequences can still phase separate. Incompatibility arises from distinct positional correlations between monomers of different chains. Calculating the Gaussian fluctuation correction to the free energy reveals that critical incompatibility from sequence differences follows a distinct yet universal scaling with chain length, $χ_{AB}^{cr} \sim 1 / \sqrt{N}$. This power law holds for both regular-sequence and statistical copolymers. A closed-form expression is derived for blends of block-alternating chains. The new theoretical scaling is confirmed by coarse-grained simulations, offering important insights into multiphase coexistence in biomolecular condensates.

When phospholipids crystallize within the otherwise fluid membranes of giant unilamellar vesicles, the resulting molecularly-thin "2D" solids exhibit great variety in their morphology evolution. For instance within membranes containing moderate amounts of the crystallizing component, crystals grow with a fixed morphology depending on vesicle size. Conversely for membranes containing large amounts of the crystallizing species, we find small compact crystals on vesicles of all sizes. However on large vesicles, growing crystals sprout flower petals that lengthen progressively. These behaviors result from two combined mechanisms: First, like other 2D solids, the shear rigidity of phospholipid crystals renders them intolerant to morphologies with non-zero Gaussian curvature. As a result and especially at elevated membrane tension, the cost of bending elasticity is reduced, at the expense of line energy, by the formation of flowers as opposed to compact crystals. Second, the composition-dependent tension rise during cooling relaxes via water permeation of the membrane with a time constant scaling as $R^2$. The amount of crystal formed for a small decrease in temperature determines this composition-dependent increase in stress from thermal contractions versus solidification. Surface Evolver computations motivated using the predicted tension evolution to develop a processing space that maps to experimental observations for initial and growing crystal morphology. Important variable groups are identified, including a scaled ratio of bending to line energy, a vesicle size-independent group for membrane contractions, and a time constant for stress relaxation. Though processing stresses ultimately relax, the crystal morphology persists well beyond the processing window.

The physics underlying reset in bipolar resistive memory has been the subject of decades of controversy and has been identified as the primary barrier to resistive memory technology development. This manuscript introduces a nanoscale effect in current carrying conductors, whereby surface charge induced radial electric fields are found to be inversely proportional to the radius of the conductive path. This nanoscale effect is then applied to explain the negative resistance switching (reset) mechanism in filamentary metal oxide resistive switching memory devices (memristors). Previous explanations for the negative resistive switching mechanism state that diffusion constitutes the radial driving mechanism for oxygen ions, and drift under electric fields is restricted to the direction parallel to current flow. This explanation conflicts with retention and microscopy data collected in a subset of devices presented in literature. We demonstrate that the electric field's dependency on the on the radius of a nanoscale conductive path can result in radial fields on the order of 10^5 to 10^6 V/cm at only -1 V bias, sufficient to govern the negative resistance switching mechanism in filamentary metal oxides. By accounting for this nanoscale size effect, long standing anomalous experimental data about the negative (reset) resistance switching mechanism in bipolar filamentary resistive memory devices is finally reconciled. Wide understanding of surface charges and associated electric fields in nanoscale conductive paths could prove important for further scaling of integrated circuits and aid in elucidating many nanoscale phenomena.

Non-Hermitian (NH) Dirac semimetals describe open gain--loss systems, yet at charge neutrality models featuring real spectrum often look Hermitian-like, with NH effects absorbed into renormalized band parameters. Here we show that a response-based diagnostic of effective non-Hermiticity can be formulated using minimal pseudo-Lorentz-symmetry-breaking deformations, which separate observables that remain captured by parameter redefinitions from those that exhibit irreducible NH structure. For a two-dimensional NH Dirac semimetal in the weak-NH, real-spectrum regime, we analyze Dirac-cone tilt and velocity anisotropy and compute representative probes of spectral structure, quantum geometry, optical response, and viscoelasticity at zero temperature. We find that tilt yields an NH-dependent slope of the density of states that cannot be collapsed to a single effective velocity, while velocity anisotropy can be captured by effective-velocity reparametrization. Furthermore, the quantum metric and collisionless optical conductivities provide NH-insensitive benchmarks (with the nonlinear conductivity symmetry selected), whereas the shear viscosity offers a discriminator through its tensor structure. Our results identify minimal deformations and bulk response channels that enable access to effective non-Hermiticity even when the spectrum remains real.

Proper management of resources whose arrival and consumption are subject to environmental randomness is an intrinsic process in both natural and artificial systems. This phenomenon can be modeled as a queuing process whose arrival distribution is determined by a search process with stochastic resetting. When the queuing system has a limited number of servers and the search process occurs within a bounded domain, the dynamics of expediting or delaying the search through stochastic resetting interact with the long-term dynamics of the number of resources in the queue. We combine results from queuing theory with those from search processes with stochastic resetting in a bounded domain to obtain regions of the parameter space of the search process that ensure convergence of the number of resources in the queue to a steady state. Furthermore, we find a threshold resetting rate at which the effects of stochastic resetting shift from reducing convergence regions to expanding them. Finally, we demonstrate that this threshold value grows exponentially with the number of servers, making it harder for stochastic resetting to improve the convergence of the queueing system.

Understanding how multi-component liquid mixtures undergo phase separation is central to elucidating biophysical organization in the cell. Here, combining analytical and numerical results, we characterise the dynamics of mixtures with disordered interactions among the components. We first study how two coexisting phases become unstable, leading to multiphase coexistence. We then show that the scaling of droplet radius as $t^{1/3}$ and droplet number as $n^{-2/3}$, characteristic of Ostwald ripening in two dimensions, can be severely delayed. This delay arises from glass-like relaxation and the emergence of long-lived metastable states characterized by different wetting angles.

Measurement-induced phase transitions (MIPT) give rise to novel dynamical states of quantum matter realized by balancing unitary evolution and measurements. We present large-scale numerical simulations of a trapped-ion native MIPT, argued to belong to the universality class described by the Haar non-unitary conformal field theory. First, through a finite-size analysis we obtained the critical measurement rate, and correlation length exponent, which falls close to the percolation value. Second, by leveraging a monotone computable via semi-definite programming, we uncover robust algebraic decay of genuine multiparty entanglement (GME) versus separation for 2, 3, and 4 parties. The corresponding critical exponents are lower-bounded by those of the multiparty mutual information, which we determine up to 4 parties, and conjecture to be (k+2) for k parties. Additionally, we derive lower bounds for both GME and multiparty mutual information.

We derive constraints on renormalization group (RG) flows and stability of phases in nonequilibrium systems using quantum information inequalities. These constraints involve conditional mutual information (CMI), which quantifies correlations between spatially separated regions not mediated by their surroundings. First, assuming CMI is UV finite, we show that the scaling function associated with CMI is monotonic along the RG flow. This implies a non-perturbative stability criterion: a fixed point with smaller CMI cannot be destabilized toward one with larger CMI. Second, we bound the CMI of a convex mixture of states in terms of the CMI of individual components. We use this inequality to infer perturbative stability of spontaneous symmetry breaking states against quantum channels that explicitly break symmetry. We illustrate these constraints through several examples, including decoherence-driven transitions in classical symmetry-broken states, strong-to-weak symmetry breaking criticality in two dimensions, and even transitions in pure quantum states. We also discuss implications for classical nonequilibrium steady states.

We use the $K$ special conformal generator in the Fuzzy sphere setup of the Ising CFT to determine primary states. For $Δ\lesssim 8$, we recover the known primaries and find several new ones, including in the parity-odd sector. We then use these primaries to compute OPE coefficients. We find that using primaries constructed from special-$K$ allows for better extrapolation of OPE coefficients to the CFT limit, because of the existence of an $O(1)$ gap between primaries and descendants in the spectrum of eigenvalues of $|K|^2$ which protects the primaries from strongly mixing with descendants. We compare the CFT data we obtain with the Eigenstate Thermalization Hypothesis.

We derive a universal bound on the efficiency with which "dissipated" work can generate distinguishable changes in a quantum many-body state at a finite temperature, as quantified by the quantum Fisher information. The bound follows solely from the analytic structure of equilibrium many-body correlators and is independent of all microscopic details. It takes a frequency-resolved form with a characteristic crossover at the Planckian scale, $ω_\star\sim k_B T/\hbar$. We find that Planckian scatterers sit at the edge of optimality, displaying maximal relaxation rate before information-dissipation efficiency collapses. This suggests strange metals are not just fast dissipators, but the fastest that remain efficient in generating distinguishability. The bounded quantity can be evaluated directly from optical conductivity measurements in strongly correlated electronic systems, offering a unique window into how dissipation generates distinguishable changes.

We introduce a two-parameter ensemble of random discrete-time Markov models that simultaneously captures critical slowing down and broken detailed balance. Extending a previously studied heterogeneous Markov ensemble, we incorporate correlations between forward and backward transition rates through a single asymmetry parameter $γ$, while heterogeneity is controlled by $ε$. Using results from random matrix theory, we identify a critical locus $ε_c(γ,N)$ at which relaxation times diverge and spectral universality breaks down. We characterize the behavior of entropy production, predictive information, and relaxation dynamics across the ensemble, showing that many observables depend strongly on heterogeneity but only weakly on asymmetry, except near the symmetric limit. Applying maximum-likelihood inference to human fMRI and EEG data, we find that both modalities operate near the predicted critical locus and occupy a similar region of the $ε-γ$ plane, supporting a super-universality of human brain dynamics. While ensemble averages are well captured by the null model, empirical data exhibit substantially enhanced variability, indicating subject-specific structure beyond random expectations. Our results unify criticality and nonequilibrium measures within a single framework and clarify their intertwined role in the analysis of complex biological dynamics.

We evaluate electromagnetic-response observables in a half-filled Chern band, across a topological phase transition between a composite Fermi liquid (CFL) and a Fermi liquid (FL) phase. While a sharp gapped plasma mode exists deep in the CFL phase, we demonstrate that it is damped near the proposed continuous phase transition between CFL and FL. This plasmon-damping phenomenon originates from emergent gauge fields and a Dirac-fermion-like spectrum. Similar features also occur in other continuous deconfined topological phase transitions, such as the Laughlin to superfluid transition in a bosonic system. In particular, this damping behavior extends over a finite range across the phase boundary, and, hence, we expect it to persist even when the transition is weakly first-order. Furthermore, we analyze the behavior of the Drude weight, the wavevector-dependent conductivity, and the chiral mirror effect across these topological phase transitions.

When exposed to a contagion phenomenon, individuals may respond to the perceived risk of infection by adopting behavioral changes, aiming to reduce their exposure or their risk of infecting others. The social cost of such adaptive behaviors and their impact on the contagion dynamics have been investigated in pairwise networks, with binary interactions driving both contagion and risk perception. However, contagion and adaptive mechanisms can also be driven by group (higher-order) interactions. Here, we consider several adaptive behaviors triggered by awareness of risk perceived through higher-order and pairwise interactions, and we compare their impact on pairwise and higher-order contagion processes. By numerical simulations and a mean-field analytic approach, we show that adaptive behaviors driven by higher-order information are more effective in limiting the spread of a contagion, than similar mechanisms based on pairwise information. Meanwhile, they also entail a lower social cost, measured as the reduction of the intensity of interactions in the population. Indeed, adaptive mechanisms based on higher-order information lead to a heterogeneous risk perception within the population, producing a higher alert on nodes with large hyperdegree (i.e., participating in many groups), on their neighborhoods, and on large groups. This in turn prevents the spreading process to exploit the properties of these nodes and groups, which tend to drive and sustain the dynamics in the absence of adaptive behaviors.

Walking droplets are millimetric fluid drops that propel themselves across a vibrated liquid bath through interaction with their self-generated waves. They constitute classical active wave-particle entities and exhibit a range of hydrodynamic quantum analogs. We investigate an \emph{active soft-impact oscillator} as a minimal model for a walking droplet moving within a piecewise-smooth external potential, analogous to classical mass-spring soft-impact oscillators and recently explored quantum soft-impact oscillators. Our active soft-impact oscillator model couples a non-smooth soft-impact force to the Lorenz-like dynamics arising from the wave-particle entity. Theoretical and numerical exploration of the full parameter space reveals a wide variety of nonlinear behaviors and bifurcations driven by impact and grazing events. These include grazing-induced and impact-induced transitions between periodic and chaotic motion, as well as grazing-mediated attractor switching and impact-free (invisible) attractor switching. The active soft-impact oscillator thus provides a versatile platform for probing nonlinear impact dynamics in active systems and exploring hydrodynamic quantum analogs in non-smooth potentials.

Heteroclinic connections between two distinct hyperbolic periodic orbits in conservative systems are important in a wide range of applications. On the other hand, it is theoretically challenging to find large amplitude connections from scratch and compute them numerically. In this paper, we use a codimension two singularity, in a family of periodic orbits, as an organizing center for the emergence of heteroclinic connections. A normal form is derived whose unfolding produces two distinct finite amplitude periodic orbits with an explicit heteroclinic connection. We also construct heteroclinic connections far from the singularity by numerical continuation, using two numerical strategies: shooting and the core-farfield decomposition. A key geometric tool in the numerics is cylindrical foliations for the stable and unstable manifolds and their intersection. We introduce a new property of heteroclinic connections - the action - and show it is an invariant along foliations, it has a jump at a surface of section, and it appears in a central way in the normal form theory. We find that the difference in asymptotic phase between minus and plus infinity is also a key property. The theory is applied to the Swift-Hohenberg equation, the nonlinear Schrodinger with fourth order dispersion, and coupled Boussinesq equations from water waves, all of which have an energy and action conservation law.

Vegetation in semi-arid ecosystems frequently organizes into spatially heterogeneous mosaics that regulate ecosystem functioning, productivity, and resilience. These patterns arise from local biological interactions, including facilitation among neighboring plants and competition for limiting resources. Classical theoretical approaches have attributed such organization to scale-dependent feedbacks, predicting regular spatial patterns and abrupt transitions to collapse. However, growing empirical and theoretical evidence reveal that environmental variability and demographic stochasticity can fundamentally reshape spatial organization, driving irregular clusters, dynamic mosaics, and gradual rather than catastrophic vegetation declines. In drylands, rainfall variability is a dominant source of environmental forcing: precipitation typically occurs in short, irregular pulses that transiently enhance survival and recruitment before competitive interactions again dominate. Near persistence thresholds, ecosystem dynamics are therefore governed not only by average climatic conditions but also by the timing and spatial coincidence of favorable events. Under these conditions, positive density dependence and local facilitation can critically determine whether vegetation patches persist, expand, or collapse. Here, we develop an individual-based model that integrates intermittent precipitation with local Allee effects to examine how stochastic rainfall shapes spatial organization and persistence. We show that the interaction between pulsed resource availability and density-dependent survival generates irregular cluster structures and strongly modulates extinction risk, with resilience emerging from local spatial covariance and neighborhood density rather than from total biomass alone. These results highlight the importance of individual-level, stochastic processes in determining ecosystem resilience.

We revisit thermal out-of-time-order correlators (OTOCs) in single-particle quantum systems, focusing on magnetic billiards. Using the stadium billiard as a testbed, we compute the thermal OTOC $C_T(t) = -\langle [x(t), p]^2 \rangle_β$ and extract Lyapunov-like exponents $λ_L$ that quantify early-time growth. We map out $λ_L(T, B)$, revealing a crossover from quantum chaos to magnetic rigidity. In parallel, we compute an alternative OTOC built from guiding-center operators, which exhibits qualitatively distinct dynamics and no exponential growth. Our results offer a controlled framework for probing scrambling, temperature dependence, and the interplay of geometry and magnetic fields in quantum systems.

We establish universality for extremal solutions of the focusing nonlinear Schrödinger equation. Extremal solutions are $N$-soliton solutions that achieve the theoretical maximal amplitude and diverge as $N \to \infty$. We consider extremal solutions with the discrete eigenvalues randomly drawn from sub-exponential distributions, and identify two distinct universality classes, determined by the macroscopic structure of the spectrum: the Painlevé--III rogue-wave solution, where the eigenvalues take the form $λ_j = v_j + i μ_j$, and the Painlevé--V rogue wave solution, where $λ_j = -ζ\, j + v_j + i μ_j$, with $0 < ζ< 1$. (In both cases, $μ_{j}$ and $v_{j}$ are subexponential random variables.) Universality can then be summarized as follows: independently of the specific distribution of the eigenvalues, the rescaled solutions converge locally to a deterministic profile governed by the Painlevé-III equation in the first regime, and the Painlevé-V equation in the second. These results demonstrate that the formation of Painlevé-type rogue waves is a universal phenomenon robust to randomness.

Bifurcations take place in molecular Hamiltonian nonlinear systems as the excitation energy increases, this leading to the appearance of different classical resonances. In this paper, we study the quantum manifestations of these classical resonances in the isomerizing system CN-Li$\leftrightarrows$Li-CN. By using a correlation diagram of eigenenergies versus Planck constant, we show the existence of different series of avoided crossings, leading to the corresponding series of quantum resonances, which represent the quantum manifestations of the classical resonances. Moreover, the extrapolation of these series to $\hbar=0$ unveils the correspondence between the bifurcation energy of classical resonances and the energy of the series of quantum resonances in the semiclassical limit $\hbar\to0$. Additionally, in order to obtain analytical expressions for our results, a semiclassical theory is developed.

Modeling count data is important in physics and other scientific disciplines, where measurements often involve discrete, non-negative quantities such as photon or neutrino detection events. Traditional parametric approaches can be trained to generate integer-count predictions but may struggle with capturing complex, non-linear dependencies often observed in the data. Gaussian process (GP) regression provides a robust non-parametric alternative to modeling continuous data; however, it cannot generate integer outputs. We propose the Poisson Log-Normal (PoLoN) process, a framework that employs GP to model Poisson log-rates. As in GP regression, our approach relies on the correlations between data points captured via GP kernel structure rather than explicit functional parameterizations. We demonstrate that the PoLoN predictive distribution is Poisson-LogNormal and provide an algorithm for optimizing kernel hyperparameters. Furthermore, we adapt the PoLoN approach to the problem of detecting weak localized signals superimposed on a smoothly varying background - a task of considerable interest in many areas of science and engineering. Our framework allows us to predict the strength, location and width of the detected signals. We evaluate PoLoN's performance using both synthetic and real-world datasets, including the open dataset from CERN which was used to detect the Higgs boson at the Large Hadron Collider. Our results indicate that the PoLoN process can be used as a non-parametric alternative for analyzing, predicting, and extracting signals from integer-valued data.

In response to the 2008 global financial crisis, Science Foundation Ireland (SFI), now Research Ireland, pivoted to research with potential socioeconomic impact. Given that the latter can encompass higher technology readiness levels, which typically correlates with lower academic impact, it is interesting to understand how academic impact holds up in SFI funded research. Here we decompose SFI \textit{Investigator Awards} - arguably the most academic funding call - into $3,243$ constituent publications and field weighted citation impact (FWCI) values searchable in the SCOPUS database. Given that citation counts are skewed, we highlight the limitation of FWCI as a paper metric, which naively restricts one to comparisons of average FWCI ($\overline{\mathrm{FWCI}}$) in large samples. Neglecting publications with $\textrm{FWCI} < 0.1$ ($8.8\%$), SFI funded publications are well approximated by a lognormal distribution with $μ= -0.0761^{+0.017}_{-0.0039}$ and $ σ= 0.933^{+0.011}_{-0.012}$ at $95 \%$ confidence level. This equates to an $\overline{\mathrm{FWCI}} = 1.433^{+0.029}_{-0.015}$ well above $\overline{\mathrm{FWCI}}=1$ internationally. Broken down by award, we correct $\overline{\mathrm{FWCI}}$ for small samples using simulations and find $\sim 67\%$ exceed \textit{median} international academic interest, thus exhibiting a positive correlation between the potential for socioeconomic impact and academic interest.

The collision phenomenon of polyatomic gases is described by the collision operator of extended Boltzmann equation or the energy-exchange model in particle direct simulations, for example, the Borgnakke-Larsen model. However, as a collision kernel, it dose not guarantee the entrinsic detailed balance and is not integrable. In this work, the Pullin equation, which possesses an integrable collision kernel and satisfies the detailed balance constraint, is adopted as an extended Boltzmann equation for the theoretical analysis of near-continuum relaxation mechanisms. For clarity, only the translational and rotational degrees are considered in this work. Explicit analytical expressions for the temporal relaxation of macroscopic variables, including the stress force, (translational/rotational) temperature and heat flux, are obtained at the first time. This is achieved by approximating the distribution function in mixed Hermite and Laguerre for rotation and computing the collision operator moments, enabling a direct description of macroscopic non-equilibrium evolution. Base on the same elementary moment (integral) of collision operator, the macroscopic transport coefficients is found in Chapman-Enskog framework. The long-standing speculation, that thermal conduction coefficient should be depended on the degrees of thermal non-equilibrium, is rigorously confirmed and evaluated. When thermal equilibrium is enforced, the present thermal conduction coefficients can be degenerated to the famous results of Mason and Monchick. Given the correct relaxation rate, a Rykov-type novel relaxation model for Pullin equation is proposed. It can recover the interaction of transaltional and rotatioanl heat fluxes in relaxation process, which is ignored in the widely used Rykov equation. Finally, the precision of this new Rykov-type equation is examined using a series of benchmark test cases.

Infrared (IR)-transparent passive radiative cooling (PRC) enables non-contact thermal management by regulating radiative heat exchange without direct attachment to the cooling object. While asymmetric IR transmission at a specific incidence angle -- typically normal incidence -- is often emphasized, we show that such single-angle asymmetry is neither sufficient nor predictive of practical cooling performance. In this work, we demonstrate that effective non-contact PRC requires angularly distributed asymmetric IR transparency evaluated through hemispherical integration over emission directions, rather than asymmetry at a single incidence angle. To quantify this effect, an angle-resolved full-wave electromagnetic (EM) model with Bloch periodic boundary conditions and Floquet mode decomposition is employed to compute wavelength- and angle-dependent bidirectional reflection and transmission of periodic PRC structures. The resulting EM response is coupled to an energy-balance-based thermal model to predict the transient temperature evolution of the cooling object. By comparing models that account for the full angular distribution with normal-incidence-only approximations, we show that pronounced asymmetric transmission at normal incidence is generally not preserved at oblique angles. As a result, angular integration yields only marginal cooling or may even result in net heating, whereas normal-incidence-based models can substantially overestimate cooling performance. These results establish angularly distributed asymmetric transparency as a key EM design principle for IR-transparent PRC and wide-angle asymmetric metasurfaces.

We present a comparative analysis of Wikipedia and Grokipedia to examine how generative mediation alters content selection, textual rewriting, narrative structure, and evaluative framing in encyclopedic content. We model page inclusion in Grokipedia as a function of Wikipedia page popularity, density of reference, and recent editorial activity. Inclusion is non-uniform: pages with higher visibility and greater editorial conflict in Wikipedia are more likely to appear in Grokipedia. For included pages, we distinguish between verbatim reproduction and generative rewriting. Rewriting is more frequent for pages with higher reference density and recent controversy, while highly popular pages are more often reproduced without modification. We compare editing activity across the two platforms and estimate page complexity using a fitness-complexity framework to assess whether generative mediation alters patterns of editorial participation. To assess narrative organization, we construct actor-relation networks from article texts using abstract meaning representation. Across multiple topical domains, including U.S. politics, geopolitics, and conspiracy-related narratives, narrative structure remains largely consistent between the two sources. Analysis of lead sections shows broadly correlated framing, with localized shifts in laudatory and conflict-oriented language for some topics in Grokipedia. Overall, generative systems preserve the main structural organization of encyclopedic content, while affecting how content is selected, rewritten, and framed.

We present lrux, a JAX-based software package for fast low-rank updates of determinants and Pfaffians, targeting the dominant computational bottleneck in various quantum Monte Carlo (QMC) algorithms. The package implements efficient low-rank updates that reduce the cost of successive wavefunction evaluations from $\mathcal{O}(n^3)$ to $\mathcal{O}(n^2k)$ when the update rank $k$ is smaller than the dimension $n$ of matrices. Both determinant and Pfaffian updates are supported, together with delayed-update strategies that trade floating-point operations for reduced memory traffic on modern accelerators. lrux natively integrates with JAX transformations such as JIT compilation, vectorization, and automatic differentiation, and supports both real and complex data types. Benchmarks on GPUs demonstrate up to $1000\times$ speedup at large matrix sizes. lrux enables scalable, high-performance evaluation of antisymmetric wavefunctions and is designed as a drop-in component for a wide range of QMC workflows. lrux is available at https://github.com/ChenAo-Phys/lrux.

An active dissipative process organizes auditory frequency analysis in the mammalian cochlea. A minimal active beam model reveals that a spatially varying viscous coupling operator, $\partial_{xx}κ\partial_{xx}$, generates dissipative forces with wave--like propagation. Local energy injection and spatial redistribution compete to govern the dynamics. This balance enables the quantitative reproduction of four key features: sharp tuning, high gain, compression, and spontaneous otoacoustic emissions. Hearing thereby belongs to a broad class of nonequilibrium pattern-forming systems.

Understanding how pedestrians adjust their movement when interacting with autonomous vehicles (AVs) is essential for improving safety in mixed traffic. This study examines micro-level pedestrian behaviour during midblock encounters in the NuScenes dataset using a hybrid discrete choice-machine learning framework based on the Residual Logit (ResLogit) model. The model incorporates temporal, spatial, kinematic, and perceptual indicators. These include relative speed, visual looming, remaining distance, and directional collision risk proximity (CRP) measures. Results suggest that some of these variables may meaningfully influence movement adjustments, although predictive performance remains moderate. Marginal effects and elasticities indicate strong directional asymmetries in risk perception, with frontal and rear CRP showing opposite influences. The remaining distance exhibits a possible mid-crossing threshold. Relative speed cues appear to have a comparatively less effect. These patterns may reflect multiple behavioural tendencies driven by both risk perception and movement efficiency.

A growing literature has examined whether innovation is becoming less disruptive, spanning diverse domains and data sources and using a range of methodologies. This paper provides an inventory of 105 studies exploring this question. The evidence is largely consistent in direction. Studies spanning scientific papers, patents, products, legal cases, music, and visual art consistently report evidence of a decline. This pattern holds not only for citation-based measures, but also for text-based approaches, firm displacement rates, product similarity networks, and audio and visual embeddings. The literature has also identified notable exceptions, including rebounds in specific domains and predictable variation across field lifecycles. We catalog each study's data, methods, and findings to provide a resource for researchers and policymakers seeking to understand the current state of the evidence.

Cosmic voids, the largest underdense regions in the Universe, provide unique laboratories for studying galaxy formation and constitute powerful probes of cosmology. Recent work has shown that individual galaxy bias (b_i), which quantifies how each galaxy traces the underlying dark matter field, exhibits a characteristic radial dependence within spherical voids, defining a void bias profile in which galaxies near void centers display systematically lower bias values. We investigate how the environmental modulation of individual galaxy bias depends on the adopted void-finding algorithm by comparing measurements across five distinct void definitions: spherical voids 'sparkling', watershed-based methods ('zobov' and 'revolver' in two modes), and free-form integrated-density voids ('popcorn'). We apply these complementary void-finding algorithms to the same galaxy sample drawn from the IllustrisTNG simulation (TNG300-1 at $z=0$) and compute individual galaxy bias profiles as a function of distance from void centers. We quantify the correlation between b_i and the membership of the void catalogs and explore how this relationship varies with the integrated underdensity threshold for density-based methods. We find that the radial gradient of individual bias within voids, generally increasing from negative values at the void centers to higher values at the boundaries, is robust across most void definitions. However, density-threshold methods preferentially select galaxies with b_i<0, while watershed methods without density constraints include substantial contamination from high-bias boundary galaxies. The correlation between negative bias and void membership is systematically strengthened as the integrated underdensity threshold becomes less restrictive, with popcorn achieving the highest purity in isolating anti-biased populations.

Community detection is a critical tool for understanding the mesoscopic structure of large-scale networks. However, when applied to aggregated or coarse-grained social networks, disjoint community partitions cannot capture the diverse composition of community memberships within aggregated nodes. While existing mixed membership methods alleviate this issue, they may detect communities that are highly sensitive to the aggregation resolution, not reliably reflecting the community structure of the underlying individual-level network. This paper presents the Link Fraction Mixed Membership (LFMM) method, which computes the mixed memberships of nodes in aggregated networks. Unlike existing mixed membership methods, LFMM is consistent under aggregation. Specifically, we show that it conserves community membership sums at different scales. The method is utilized to study a population-scale social network of the Netherlands, aggregated at different resolutions. Experiments reveal variation in community membership across different geographical regions and evolution over the last decade. In particular, we show how our method identifies large urban hubs that act as the melting pots of diverse, spatially remote communities.

Semantic representations can be framed as a structured, dynamic knowledge space through which humans navigate to retrieve and manipulate meaning. To investigate how humans traverse this geometry, we introduce a framework that represents concept production as navigation through embedding space. Using different transformer text embedding models, we construct participant-specific semantic trajectories based on cumulative embeddings and extract geometric and dynamical metrics, including distance to next, distance to centroid, entropy, velocity, and acceleration. These measures capture both scalar and directional aspects of semantic navigation, providing a computationally grounded view of semantic representation search as movement in a geometric space. We evaluate the framework on four datasets across different languages, spanning different property generation tasks: Neurodegenerative, Swear verbal fluency, Property listing task in Italian, and in German. Across these contexts, our approach distinguishes between clinical groups and concept types, offering a mathematical framework that requires minimal human intervention compared to typical labor-intensive linguistic pre-processing methods. Comparison with a non-cumulative approach reveals that cumulative embeddings work best for longer trajectories, whereas shorter ones may provide too little context, favoring the non-cumulative alternative. Critically, different embedding models yielded similar results, highlighting similarities between different learned representations despite different training pipelines. By framing semantic navigation as a structured trajectory through embedding space, bridging cognitive modeling with learned representation, thereby establishing a pipeline for quantifying semantic representation dynamics with applications in clinical research, cross-linguistic analysis, and the assessment of artificial cognition.

We introduce several axioms which may or may not hold for any given subgraph of the directed graph of all organisms (past, present and future) where edges represent biological parenthood, with the simplifying background assumption that life does not go extinct. We argue these axioms are plausible for species: if one were to define species based purely on genealogical relationships, it would be reasonable to define them in such a way as to satisfy these axioms. The main axiom we introduce, which we call the identical ancestor point axiom, states that for any organism in any species, either the species contains at most finitely many descendants of that organism, or else the species contains at most finitely many non-descendants of that organism. We show that this (together with a convexity axiom) reduces the subjectivity of species, in a technical sense. We call connected sets satisfying these two axioms "specieslike clusters." We consider the question of identifying a set of biologically plausible constraints that would guarantee every organism inhabits a maximal specieslike cluster subject to those constraints. We provide one such set consisting of two constraints and show that no proper subset thereof suffices.

The relationship between traits that influence pathogen virulence and transmission is part of the central canon of the evolution and ecology of infectious disease. However, identifying directional and mechanistic relationships among traits remains a key challenge in various subfields of biology, as models often assume static, fixed links between characteristics. Here, we introduce learning evolutionary trait relationships (LETR), a data-driven framework that applies Granger-causality principles to determine which traits drive others and how these relationships change over time. LETR integrates causal discovery with generative mapping and transfer-operator analysis to link short-term predictability with long-term trait distributions. Using a synthetic myxomatosis virus-host data set, we show that LETR reliably recovers known directional influences, such as virulence driving transmission. Applying the framework to global pandemic (SARS-CoV-2) data, we find that past virulence improves future transmission prediction, while the reverse effect is weak. Invariant-density estimates reveal a long-term trend toward low virulence and transmission, with bimodality in virulence suggesting ecological influences or host heterogeneity. In summary, this study provides a blueprint for learning the relationship between how harmful a pathogen is and how well it spreads, which is highly idiosyncratic and context-dependent. This finding undermines simplistic models and encourages the development of new theory for the constraints underlying pathogen evolution. Further, by uniting causal inference with dynamical modeling, the LETR framework offers a general approach for uncovering mechanistic trait linkages in complex biological systems of various kinds.

The tumor-immune system plays a critical role in colorectal cancer progression. Recent preclinical and clinical studies showed that combination therapy with anti-PD-L1 and cancer vaccines improved treatment response. In this study, we developed a multiscale mathematical model of interactions among tumors, immune cells, and cytokines to investigate tumor evolutionary dynamics under different therapeutic strategies. Additionally, we established a computational framework based on approximate Bayesian computation to generate virtual tumor samples and capture inter-individual heterogeneity in treatment response. The results demonstrated that a multiple low-dose regimen significantly reduced advanced tumor burden compared to baseline treatment in anti-PD-L1 therapy. In contrast, the maximum dose therapy yielded superior tumor growth control in cancer vaccine therapy. Furthermore, cytotoxic T cells were identified as a consistent predictive biomarker both before and after treatment initiation. Notably, the cytotoxic T cells-to-regulatory T cells ratio specifically served as a robust pre-treatment predictive biomarker, offering potential clinical utility for patient stratification and therapy personalization.

In this study, we develop a metapopulation model framework to identify optimal harvesting strategies for a population in a stream network. We consider two distinct optimization objectives: maximization of total biomass and maximization of total yield, under the constraint of a fixed total harvesting effort. We examine in detail the special case of a two-patch network and fully characterize the optimal strategies for each objective. We show that when the population growth rate exceeds a critical threshold, a single harvesting strategy can simultaneously maximize both objectives. For general $n$-patch networks with homogeneous growth rates across patches, we focus on the regime of large growth rates and demonstrate that the optimal harvesting strategy selects patches according to their intraspecific competition rates and an effective net flow metric determined by network connectivity parameters.

Photonic quantum computing is a strong contender in the race to fault-tolerance. Recent proposals using qubits encoded in frequency modes promise a large reduction in hardware footprint, and have garnered much attention. In this encoding, linear optics, i.e., beam splitters and phase shifters, is necessarily not energy-conserving, and is costly to implement. In this work, we present designs of frequency-mode beam splitters based on modulated arrays of coupled resonators. We develop a methodology to construct their effective transfer matrices based on the SLH formalism for quantum input-output networks. Our methodology is flexible and highly composable, allowing us to define $N$-mode beam splitters either natively based on arrays of $N$-resonators of arbitrary connectivity or as networks of interconnected $l$-mode beam splitters, with $l<N$. We apply our methodology to analyze a two-resonator device, a frequency-domain phase shifter and a Mach-Zehnder interferometer obtained from composing these devices, a four-resonator device, and present a formal no-go theorem on the possibility of natively generating certain $N$-mode frequency-domain beam splitters with arrays of $N$-resonators.

We experimentally investigate the quantum noise mechanisms that limit continuously operating multiparameter quantum sensors. Using a hybrid rf-dc optically pumped magnetometer, we map the photon shot noise, spin projection noise, and measurement back-action noise over an order of magnitude in probe power and a factor of three in pump power while remaining quantum-noise-limited. We observe linear, quadratic, and cubic scaling of the respective total noise powers with probe photon flux, together with a quadratic dependence of back-action on pump photon flux, in quantitative agreement with a stochastic Bloch-equation model. At higher probe powers, additional probe-induced relaxation modifies the spin-noise spectrum while preserving the integrated noise scaling. Our results reveal fundamental, resource-dependent trade-offs unique to continuously monitored multiparameter sensors and establish experimentally the quantum limits governing their optimal operation.

This work investigates a modified theory of gravity where the Einstein-Hilbert action, including a cosmological constant, is non-minimally coupled to a Yang-Mills field via an $R^3 F^{(a)}_{μα}F^{(a)μα}$ interaction term. We derive a black brane solution for this model, accurate to the first order in the coupling parameter. Using gauge/gravity duality techniques, we then compute two key holographic transport coefficients: the color non-abelian direct current (DC) conductivity and the ratio of shear viscosity to entropy density. Our analysis reveals that both transport coefficients are modified by the non-minimal coupling, with the conductivity bound violated for positive $q_2$ and the Kovtun-Son-Starinets (KSS) bound for shear viscosity violated for negative $q_2$. In the limit where the non-minimal coupling vanishes, our results consistently reduce to those of the standard Yang-Mills Schwarzschild Anti-de Sitter (AdS) black brane.

Optical parametric amplifiers (OPAs) are promising to overcome the wavelength coverage and noise limitations in conventional optical amplifiers based on rare-earth doping and semiconductor gain. However, the high power requirement remains a major obstacle to the widespread use of OPAs. Integrated OPAs can in principle improve the pump efficiency with tight mode confinement; however, challenges associated with propagation loss, limited nonlinearity, and susceptibility to nanoscale fabrication imperfections prevent them from competing with conventional bulk and fiber-based OPAs. Here, we demonstrate a highly efficient integrated OPAs with continuous-wave net gain. The pump efficiency is improved by over one order of magnitude. Phase-sensitive gain of 23.5 dB is demonstrated, significantly exceeding previous integrated OPAs, using only 110 mW pump power and no cavity enhancement. This is achieved with parametric down-conversion in thin-film lithium niobate waveguides using the adapted poling technique to maintain the coherence of nonlinear interactions. Moreover, the high parametric gain exceeds fibre-chip-fibre losses, leading to appreciable net gain up to 10 dB. The 3 dB bandwidth is approximately 120 nm, covering telecommunication S-, C-, and Lbands. Quantum-limited noise performance is confirmed through the measurement of output field fluctuation below the classical limit. We further demonstrate that signalto-noise ratio in noisy optical communications can be increased by leveraging this efficient integrated OPA. Our work marks a significant step towards ideal optical amplifiers with strong amplification, high efficiency, quantum-limited noise, large bandwidth, and continuous-wave operation, unlocking new possibilities for next-generation photonic information processing systems.

We propose a new approach to improve the accuracy of ground state estimates in Variational Quantum Eigensolver (VQE) algorithms by employing subspace representations with soft-coded orthogonality constraints. As in other subspace-based VQE methods, such as the Subspace-Search VQE (SSVQE) and Multistate Contracted VQE (MCVQE), once the parameters are optimized to maximize the subspace overlap with the low-energy sector of the Hamiltonian, one diagonalizes the Hamiltonian restricted to the subspace. Unlike these methods, where \emph{hard-coded} orthogonality constraints are enforced at the circuit level among the states spanning the subspace, we consider a subspace representation where orthogonality is \emph{soft-coded} via penalty terms in the cost function. We show that this representation allows for shallower quantum circuits while maintaining high fidelity when compared to single-state (standard VQE) and multi-state (SSVQE or MCVQE) representations, on two benchmark cases: a $3\times 3$ transverse-field Ising model and random realizations of the Edwards--Anderson spin-glass model on a $4\times 4$ lattice.

The Rodeo Algorithm is a quantum computing method for computing the energy spectrum of a Hamiltonian and preparing its energy eigenstates. We discuss how to improve the performance of the rodeo algorithm for each of these two applications. In particular, we demonstrate that using a geometric series of time samples offers a near-optimal optimization space for a given total runtime by studying the Rodeo Algorithm performance on a model Hamiltonian representative of gapped many-body quantum systems. Analytics explain the performance of this time sampling and the conditions for it to maintain the established exponential performance of the Rodeo Algorithm. We finally demonstrate this sampling protocol on various physical Hamiltonians, showing its practical applicability. Our results suggest that geometric series of times provide a practical, near-optimal, and robust time-sampling strategy for quantum state preparation with the Rodeo Algorithm across varied Hamiltonians without requiring model-specific fine-tuning.

We explore the consequences of allowing non-Hermitian structures in quantum cosmology by extending the Wheeler DeWitt framework beyond strictly Hermitian dynamics. Using a controlled semiclassical reduction, we show how anti Hermitian contributions propagate into both early universe primordial fluctuations and late-time structure growth as effective damping or gain terms. Confronting this framework with inflationary observables, growth of structure and the observed near flatness of the universe, we derive strong infrared constraints that suppress non Hermiticity across cosmic history. We demonstrate that these bounds are mutually consistent between early and late-time probes and can be partially relaxed in theories beyond General Relativity. Our results establish cosmology as a novel arena for testing foundational aspects of quantum mechanics and suggest that Hermiticity may emerge dynamically along the semiclassical branch describing our universe.

Motivated by the question of the distinguishability of ensembles described by the same compressed density operator, we propose a model for one-way quantum secure direct communication using finite ensembles of shared EPR pairs per bit and a public authenticated classical channel, where the local choice of one of two mutually-unbiased measurement bases is the secret bit. In this model, both the encoding and decoding of classical information in quantum systems are implemented by measurements in either the computational or the Hadamard basis. Using the quantum wiretap channel theory, we study the secure net bit rates and certify information-theoretic security of different implementations of our model when the quantum channel is subjected to BB84-symmetric attacks. Since no local unitary operations need to be performed by the receiver, the proposed model is suitable for real-life implementation of secure direct communication in star network configurations.

We evaluate the capability of current and next-generation gravitational wave detectors, such as Advanced LIGO, Einstein Telescope and DECIGO, to constrain cosmographic parameters using electromagnetically bright standard sirens. By adopting a third-order Taylor expansion, we analyze how signal-to-noise ratios and the number of events impact the estimates of the Hubble constant ($H_0$), the deceleration ($q_0$) and jerk ($j_0$) parameters. Our results show that while Advanced LIGO provides a calibration-free measurement of $H_0$ at the few-percent level, it remains insensitive to higher-order parameters. In contrast, the Einstein Telescope and DECIGO reach sub-percent accuracy for $H_0$. Notably, DECIGO achieves a precision better than 10\% for the deceleration parameter $q_0$ and a few tens of percent for the jerk parameter $j_0$.

We report on the realization of a $^{133}$Cs optical tweezer array in a cryogenic blackbody radiation (BBR) environment. By enclosing the array within a 4K radiation shield, we measure long Rydberg lifetimes, up to $406 (36)\,μ$s for the $55 P_{3/2}$ Rydberg state, a factor of 3.3(3) longer than the room-temperature value. We employ single-photon coupling for coherent manipulation of the ground-Rydberg qubit. We measure a small differential dynamic polarizability of the transition, beneficial for reducing dephasing due to light intensity fluctuations. Our results pave the path for advancing neutral-atom two-qubit gate fidelities as their error budgets become increasingly dominated by $T_1$ relaxation of the ground-Rydberg qubit.

Quantum algorithms for binary optimization problems have been the subject of extensive study. However, the application of quantum algorithms to integer optimization problems remains comparatively unexplored. In this paper, we study the Quantum Approximate Optimization Algorithm (QAOA) applied to integer problems on graphs, with each integer variable encoded in a qudit. We derive a general iterative formula for depth-$p$ QAOA expectation on high-girth $d$-regular graphs of arbitrary size. The cost of evaluating the formula is exponential in the QAOA depth $p$ but does not depend on the graph size. Evaluating this formula for Max-$k$-Cut problem for $p\leq 4$, we identify parameter regimes ($k=3$ with degree $d \leq 10$ and $k=4$ with $d \leq 40$) in which QAOA outperforms the Frieze-Jerrum semi-definite programming (SDP) algorithm, which provides the best worst-case guarantee on the approximation ratio. To strengthen the classical baseline we introduce a new heuristic algorithm, based on the degree-of-saturation, that empirically outperforms both the Frieze-Jerrum algorithm and shallow-depth QAOA. Nevertheless, we provide numerical evidence that QAOA may overtake this heuristic at depth $p\leq 20$. Our results show that moving beyond binary to integer optimization problems can open up new avenues for quantum advantage.

We prove that the Schrödinger operator describing four particles in two dimensions, interacting solely through short-range three-body forces, can possess infinitely many bound states. This holds under the assumption that each three-body subsystem has a virtual level at zero energy. Our result establishes an analog of the Efimov effect for such four-particle systems in two dimensions.

We present the first quantum-simulation study of bound and resonant doubly-bottom tetraquark states within a QCD-inspired chiral quark model. An effective four-quark Hamiltonian is mapped onto a 16-qubit register, encoding color, spin, and spatial degrees of freedom, and incorporating both meson-meson and diquark-antidiquark configurations with complete color bases. Using a variational quantum eigensolver, we identify bound and resonance states in the low-lying $S$-wave sector. Deeply bound states are found exclusively in the isoscalar $I(J^{P})=0(1^{+})$ channel, dominated by color-singlet meson-meson components with non-negligible hidden-color contributions. The resulting masses and binding energies are consistent with classical chiral quark model predictions, establishing quantum simulation as a viable framework for studying exotic multiquark states beyond the reach of conventional methods.

The one-state statement for closed universes has sparked considerable discussion. In this paper we examine its physical meaning in the context of the Hartle-Hawking state and de Sitter space. We argue that the one-state property of closed universes is fully compatible with the finite-dimensional quantum mechanics experienced by observers inside de Sitter space, and that this compatibility requires neither mixing of alpha sectors nor any modification of the rules of the gravitational path integral. The apparent tension is resolved by sharply distinguishing the baby-universe Hilbert space, namely the space of closed universes viewed from the outside, from the bulk Hilbert space that governs quantum mechanics for an observer inside a single de Sitter universe. The baby-universe Hilbert space, together with its commutative operator algebra, is not a quantum-mechanical Hilbert space: it is merely a mathematical repackaging of classical probability theory and does not carry any quantum-mechanical structure at all, a direct consequence of the one-state property of closed universes. Accordingly, attempting to formulate quantum mechanics directly on the baby-universe Hilbert space conflates classical ensemble data with the quantum mechanics experienced by bulk observers and leads to physically incorrect conclusions. By contrast, the quantum mechanics experienced by an observer inside de Sitter space emerges from the classical statistics encoded in the baby-universe Hilbert space, providing a concrete realization of Wheeler's idea of "It from Bit". We demonstrate these features by completely solving a topological toy model of one-dimensional de Sitter spacetime. Along the way we clarify the physical meaning of de Sitter entropy, showing that it corresponds to the coarse-grained entropy of the underlying state.

Measurement plays a crucial role in a quantum system beyond just learning about the system state: it changes the post-measurement state and hence influences the subsequent time evolution; further, measurement can even create entanglement in the post-measurement conditional state. In this work, we study how careful choice of parameters for a typical measurement process on cold atoms systems -- phase contrast imaging -- has a strong impact on both what the experimentalist observes but also on the backaction the measurement has on the system, including the creation and diffusion of quasiparticles emerging from the quantum many-body dynamics. We focus on the case of a Bose-Einstein-condensate array, in the low-temperature and low-momentum limit. Our theoretical investigation reveals regimes where the imaging light probes either the bare particle or quasiparticle dynamics. Moreover, we find a path to selectively measuring quasiparticle modes directly, as well as controlling over the measurement-induced creation and diffusion of quasiparticles into different momentum states. This lays a foundation for understanding the effects of both experimental approaches for probing many-body systems, but also more speculative directions such as observable consequences of `spontaneous collapse' predictions from novel models of quantum gravity on aspects of the Standard Model.

This document summarizes talks and discussions from the workshop "deci-Hz Gravitational Wave Observations on the Moon and Beyond" that took place at Johns Hopkins University between September 1 and September 3, 2025. The workshop focused on experimental proposals to observe gravitational waves in the deci-Hz band, including lunar detectors, laser interferometers in space, and atom interferometry; gravitational wave sources in the deci-Hz frequency band; and the multi-messenger and multi-band astronomy that would be enabled by these observations.

Recent results from combined cosmological probes indicate that the Dark Energy component of the Universe could be dynamical. The simplest explanation envisages the presence of a quintessence field rolling into a potential, where the Dark Energy energy density parameter $Ω_{DE}=Ω_Λ+Ω_{x}$ results from the contribution of the ground state energy $Ω_Λ$ and the scalar field energy $Ω_{x}$. Provided that $Ω_{DE}\approx 0.7$, negative values of $Ω_Λ$ can be consistent with current measurements from cosmological probes, and could help in explaining the large abundance of bright galaxies observed by JWST at $z> 10$, largely exceeding the pre-JWST expectations in a $ΛCDM$ Universe. Here we explore to what extent such a scenario can account also for the early presence of massive Black Holes (BHs) with masses $M_{BH}\gtrsim 10^7\,M_{\odot}$ observed at $z\gtrsim 8$, and for the large over-abundance of AGN with respect to pre-JWST expectations. Our aim is not to provide a detailed description of BH growth, but rather to compute the maximal BH growth that can occur in cosmological models with negative $Ω_Λ$ under the simple assumption of Eddington-limited accretion onto initial light Black Hole seeds with mass $M_{seed}\sim 10^2\,M_{\odot}$ originated from PopIII stars. To this aim we develop a simple analytic framework to connect the growth of dark matter halos to the maximal growth of BHs within the above assumptions. We show such models can account for present observations assuming values of $Ω_Λ\approx -1$, simultaneously boosting both galaxy and AGN number counts without invoking any additional physics. This would allow us to trace the observed excess of bright and massive galaxies and the early formation of massive Black Holes and the abundance of AGN to the same cosmological origin.

The realization of fault-tolerant quantum computers remains a challenging endeavor, forcing state-of-the-art quantum hardware to rely heavily on noise mitigation techniques. Standard quantum error mitigation is typically based on post-processing strategies. In contrast, the present work explores a pre-processing approach, in which the effects of noise are mitigated before performing a measurement on the output state. The main idea is to find an observable $Y$ such that its expectation value on a noisy quantum state $\mathcal{E(ρ)}$ matches the expectation value of a target observable $X$ on the noiseless quantum state $ρ$. Our method requires the execution of a noisy quantum circuit, followed by the measurement of the surrogate observable $Y$. The main enablers of our method in practical scenarios are Tensor Networks. The proposed method improves over Tensor Error Mitigation (TEM) in terms of average error, circuit depth, and complexity, attaining a measurement overhead that approaches the theoretical lower bound. The improvement in terms of classical computation complexity is in the order of $\sim 10^6$ times when compared to the post-processing computational cost of TEM in practical scenarios. Such gain comes from eliminating the need to perform the set of informationally complete positive operator-valued measurements (IC-POVM) required by TEM, as well as any other tomographic strategy.

We consider the transition between short-range entangled (SRE) and long-range ordered (and therefore long-range entangled) states of infinite quantum spin chains, which is induced by local measurements. Specifically, we assume that the initial state is in a non-trivial symmetry-protected topological phase with local symmetry group $\mathcal{G} = G \times H$, where $G$ is an Abelian subgroup. We show that the on-site measurements of the local $G$-charge on intervals of increasing lengths transform the initial SRE state into a family of states with increasingly long-range correlations. In particular, the post-measurement states cannot be uniformly short-range entangled. In the case where the initial state is obtained from a product state using a quantum cellular automaton, we construct the infinite-volume post-measurement state and exhibit almost local observables that are maximally correlated.

Random thermal states of many-body Hamiltonians underpin studies of thermalization, chaos, and quantum phase transitions, yet their generation remains costly when each Hamiltonian must be prepared individually. We introduce the thermal-drift channel, a measurement-based operation that implements a tunable nonunitary drift along a chosen Pauli term. Based on this channel, we present a measurement-controlled sampling algorithm that generates thermal states together with their Hamiltonian "labels" for general physical models. We prove that the total gate count of our algorithm scales cubically with system size, quadratically with inverse temperature, and as the inverse error tolerance to the two-thirds power, with logarithmic dependence on the allowed failure probability. We also show that the induced label distribution approaches a normal distribution reweighted by the thermal partition function, which makes an explicit trade-off between accuracy and effective range. Numerical simulations for a 2D Heisenberg model validate the predicted scaling and distribution. As an application, we compute unfolding-free level-spacing ratio statistics from sampled thermal states of a 2D transverse-field Ising model and observe a crossover toward the Wigner--Dyson prediction, demonstrating a practical and scalable route to chaos diagnostics and random matrix universality studies on near-term quantum hardware.

Quantum key distribution (QKD) achieves information-theoretic security, without relying on computational assumptions, by distributing quantum states. To establish secret bits, two honest parties exploit key distillation protocols over measurement outcomes resulting after the the distribution of quantum states. In this work, we establish a rigorous connection between the key rate achievable by applying two-way key distillation, such as advantage distillation, and quantum asymptotic hypothesis testing, via an integral representation of the relative entropy. This connection improves key rates at small to intermediate blocklengths relative to existing fidelity-based bounds and enables the computation of entropy bounds for intermediate to large blocklengths. Moreover, this connection allows one to close the gap between known sufficient and conjectured necessary conditions for key generation in the asymptotic regime, while the precise finite blocklegth conditions remain open. More broadly, our work shows how advances in quantum multiple hypothesis testing can directly sharpen the security analyses of QKD.

Transformations of quantum channels, such as the transpose, complex conjugate, and adjoint, are fundamental to quantum information theory. Given access to an unknown channel, a central problem is whether these transformations can be implemented physically with quantum supermaps. While such supermaps are known for unitary operations, the situation for general quantum channels is fundamentally different. In this work, we establish a strict hierarchy of physical realizability for the transposition, complex conjugation, and adjoint transformation of an unknown quantum channel. We present a probabilistic protocol that exactly implements the transpose with a single query. In contrast, we prove no-go theorems showing that neither the complex conjugate nor the adjoint can be implemented by any completely positive supermap, even probabilistically. We then overcome this impossibility by designing a virtual protocol for the complex conjugate based on quasi-probability decomposition, and show its optimality in terms of the diamond norm. As a key application, we propose a protocol to estimate the expectation values resulting from the Petz recovery map of an unknown channel, achieving an improved query complexity compared to existing methods.

The horizon multipole moments of a Kerr black hole are computed from two distinct definitions that have been proposed in the literature. The first one [Ashtekar et al., Class. Quantum Grav. 21, 2549 (2004)] regards axisymmetric isolated horizons, while the second one [Ashtekar et al., J. High Energ. Phys. 2022, 28 (2022)] applies to generic (i.e., not necessarily axisymmetric) non-expanding horizons. We review these definitions in a common frame and perform a detailed study of the resulting multipole moments for the Kerr event horizon. The horizon multipoles are found to share several properties with the (Hansen) field multipoles, including parity constraints and the leading scaling behavior with respect to the Kerr spin parameter a in the regime of small a. For the axisymmetry-based definition, we have obtained a closed-form expression of the multipole moments in terms of a and the spherical harmonic degree l. For the generic definition, we have established closed-form expressions for the conformal unit round metric, the `electric' and `magnetic' potentials related to the multipoles, and the values of the multipoles in the small a limit. We show that the two definitions lead to different values of the Kerr horizon multipoles as soon as l >= 1 (generic nonzero value of a) or l >= 2 (small a limit).

The use of tantalum films for superconducting qubits has recently extended qubit coherence times significantly, primarily due to reduced dielectric losses at the metal-air interface. However, the choice of base material also influences the sensitivity to quasiparticle-induced decoherence. In this study, we investigate quasiparticle tunneling rates in niobium and tantalum-based offset-charge-sensitive qubits. Using a source of thermal radiation, we characterize the sensitivity of either material to infrared radiation and explore the impact of the infrared background through the targeted use of in-line filters in the wiring and ambient infrared absorbers. We identify both radiation channels as significant contributions to decoherence for tantalum but not for niobium qubits and achieve tunneling rates of 100 Hz and 300 Hz for niobium and tantalum respectively upon installation of infrared filters. Additionally, we find a time-dependence in the observed tunneling rates on the scale of days, which we interpret as evidence of slowly cooling, thermally radiating components in the experimental setup. Our findings indicate that continued improvements in coherence times may require renewed attention to radiative backgrounds and experimental setup design, especially when introducing new material platforms.

Classical communication is the basis for many of our current and future technologies, such as mobile phones, video conferences, autonomous vehicles and particularly the internet. In contrast, quantum communication is governed by the laws of quantum mechanics. Due to this fundamental difference, it might offer enormous benefits for security applications, more precise measurements, faster computations, and many other fields of application by interconnecting different quantum devices, such as quantum sensors, quantum computers, or quantum memories. This review provides an overview of the specific properties of quantum information networks. This includes the interfaces between the classical and the quantum regime, the transmission of the quantum information by physical implementations, and potential future applications of quantum networks. We aim to provide a starting point based on fundamental concepts of quantum information processing for further research on a future quantum internet.

Quantum control has been a cornerstone of quantum information science, driving major advances in quantum computing, quantum communication, and quantum sensing. Over the years, it has enabled the implementation of arbitrary completely positive and trace-preserving (CPTP) maps; an important next step is to extend control to Hermitian-preserving and trace-preserving (HPTP) maps, which underpin applications such as entanglement detection, quantum error mitigation, quantum simulation, and quantum machine learning. Here we present an efficient and fully constructive method for implementing arbitrary HPTP maps. Unlike existing methods that decompose an HPTP map into multiple CPTP maps or approximate it using bipartite Hamiltonians with large Hilbert spaces, our approach compiles a target HPTP map into a single executable CPTP map whose Kraus rank is guaranteed to be no larger than the intrinsic rank of the target HPTP map plus one, followed by simple classical post-processing. Numerical results for inverse noise channels used in quantum error mitigation, including bosonic photon loss, confirm substantial reductions in resources and highlight scalability in higher-dimensional settings. Together with our numerical benchmarks, these results validate the efficiency and versatility of the proposed framework, opening a route to broader quantum-information applications enabled by HPTP processing.

We study perturbations induced by a light particle scattering off a Schwarzschild black hole. Exploiting recent results for the wave propagation in this geometry, we derive the fields that this process induces on the horizon to leading order in the post-Minkowskian (PM) regime, when the light probe is far from the black hole. We then use these results to calculate the fluxes of energy and angular momentum that enter the black hole. We consider the effects due to gravitational, electromagnetic and scalar radiation, finding agreement with recent computations of the absorbed energy, while the absorbed angular momentum provides a new PM result.

In this paper, we examine the motion of test particles and relativistic accretion mechanisms within the spacetime of a rotating and embedded BH. In this case, the geometric properties of the metric and their dynamical consequences for particle trajectories are systematically studied, with a specific focus on circular orbits together with their existence criteria and stability constraints. Also, the effective potential and the corresponding effective force are constructed to quantify the influence of rotation and embedding parameters on the attractive and repulsive sectors of the gravitational interaction. Closed-form expressions for orbital frequencies as measured by a distant observer are derived, enabling a quantitative analysis of relativistic precession phenomena, including periastron advance and Lense-Thirring precession. Furthermore, we conduct general-relativistic hydrodynamic simulations of BHL accretion onto rotating embedded BHs. In addition, within the framework of the BHL accretion mechanism, the numerical solution of the GRH equations shows that the embedding parameter αsystematically modifies the morphology of the shock cone formed around embedded BHs compared to the Kerr model. In particular, a wider opening angle of the cone is produced, the compression of matter in the post-shock region is weakened, and the dynamical variability of the flow is enhanced. The time-dependent mass accretion rate exhibits increasing oscillation amplitudes and long-term variations with increasing α, while these amplitudes are found to be suppressed by the frame-dragging effect associated with the BH spin parameter. At the same time, increasing values of $α$ lead to a strengthening of the QPO frequencies formed around embedded BHs in the LFQPO regime, enhancing their observability and increasing the likelihood of detecting commensurate frequency ratios such as 3:2.

In this work, we address the numerical identification of entanglement in dynamical scenarios. To this end, we consider different programs based on the restriction of the evolution to the set of separable (i.e., non-entangled) states, together with the discretization of the space of variables for numerical computations. As a first approach, we apply linear splitting methods to the restricted, continuous equations of motion derived from variational principles. We utilize an exchange interaction Hamiltonian to confirm that the numerical and analytical solutions coincide in the limit of small time steps. The application to different Hamiltonians shows the wide applicability of the method to detect dynamical entanglement. To avoid the derivation of analytical solutions for complex dynamics, we consider variational, numerical integration schemes, introducing a variational discretization for Lagrangians linear in velocities. Here, we examine and compare two approaches: one in which the system is discretized before the restriction is applied, and another in which the restriction precedes the discretization. We find that the "first-discretize-then-restrict" method becomes numerically unstable, already for the example of an exchange-interaction Hamiltonian, which can be an important consideration for the numerical analysis of constrained quantum dynamics. Thereby, broadly applicable numerical tools, including their limitations, for studying entanglement over time are established for assessing the entangling power of processes that are used in quantum information theory.

The network of space based optical lattice clocks (OLCs) has been proposed to detect the stochastic gravitational wave background. We investigate the overlap reduction function (ORF) of the OLC detector network and analytically derive a transformation that leaves the ORF invariant. This transformation is applicable to configurations with two OLC detectors, each equipped with a one-way link. It can map a configuration with small separation and high noise correlation to another configuration with larger separation and reduced noise correlation. Using this transformation, we obtain a favourable OLC detector network configuration with high cross-correlation response, and compare its sensitivity to that of space-based laser interferometer gravitational wave detectors.

Nuclear spins provide an ideal platform for studying quantum correlations and open quantum system dynamics across diverse areas, including quantum information, quantum foundations, and many-body physics. This is enabled by their long longitudinal (T1) and transverse (T2) coherence times and precise control using radio frequency pulses. In this thesis, I present my work using nuclear spins to explore these themes. First, I study temporal quantum correlations quantified by the Leggett Garg inequality (LGI) for a qubit evolving under a superposition of unitary operators. Using a three qubit quantum register, we experimentally realized superposed unitaries and observed LGI violations exceeding the maximal quantum bound of 1.5, indicating enhanced non-classicality. Notably, this superposed unitary dynamics also showed improved robustness against decoherence. Next, I investigate Lee Yang zeros, which are zeros of the partition function in the complex plane that reveal thermodynamic behavior near criticality. We proposed and experimentally demonstrated a method to determine the full set of Lee Yang zeros of an asymmetric Ising model using a single quantum probe in a three-qubit nuclear spin register. We further showed that the mutual information between the probe and system peaks at times corresponding to these zeros. I then report our study of the quantum Mpemba effect in nuclear spin relaxation, where systems farther from equilibrium can relax faster than those closer to steady state, verified both theoretically and experimentally using NMR. Finally, I discuss our work on entanglement localization and delocalization induced by local interactions, leading to an apparent violation of the quantum data processing inequality. We showed that this violation is only apparent by constructing a completely positive and trace preserving map describing the dynamics.

This study investigates the interplay between relativistic effects and electron correlation effects on the first ionization energies of heavy atoms (Au through Rn, Z = 79-86). We perform two complementary analyses: (1) comparing relativistic corrections computed at both the Hartree-Fock (HF) and coupled cluster CCSD(T) levels to assess how correlation influences the magnitude of relativistic corrections, and (2) comparing correlation corrections computed within both non-relativistic and relativistic frameworks to determine how relativity influences the magnitude of correlation corrections. Our results reveal a striking non-linear relationship between these two effects. Specifically, the combined effect of relativity and correlation on ionization energy does not equal the sum of their individual contributions. This non-additivity indicates that relativistic and correlation effects are not independent; they interact in complex ways that depend on the atomic system. We find that for some atoms, the two effects enhance each other, while for others they partially cancel. Moreover, the order in which one may add "separate" effects also counts, in that adding "pure" relativistic effects to the remaining outcome (including correlation) would give a different result than when adding "pure" correlation effects to the remaining outcome (including relativity). These findings demonstrate that relativistic and correlation effects are inherently non-additive, reflecting the non-linearity of the quantum many-body problem. Accurate computational predictions of ionization energies in heavy-element systems thus require simultaneous treatment of both effects rather than treating them as independent contributions.

Nonlinear SU(1,1) quantum interferometers based on non-degenerate optical parametric down-conversion exhibit strong unbalanced group velocity dispersion (GVD). This feature is intrinsic to this type of interferometer as correlated photons of vastly different frequencies propagate through a dispersive nonlinear crystal; consequently, the dispersion arises from the source itself. The resulting GVD degrades the axial point-spread function (PSF) in optical coherence tomography (OCT) with undetected photons; and physical compensation is less straightforward, in particular for non-degenerate broadband regimes due to the limited number of suitable materials. In this contribution, we analyze dispersion in bulk nonlinear interferometry and describe its implications for OCT imaging. Aspects of hardware compensation are addressed, and a novel empirical numerical method of compensation is proposed. The approach is based on the extraction of the phase component directly from the time-domain modality (high precision linearized quantum Fourier transform infrared spectrometer) and its injection into the mid-IR spectral-domain OCT signals (central wavelength of around 3770 nm) before the Fourier transform. The proposed method is compared with an alternative numerical technique. The results demonstrate a 2.2-fold improvement in axial resolution and outperform the alternative correction method in overall imaging performance.

We present a novel framework to infer the mass of clusters that host hierarchical binary black hole (BBH) mergers detected with gravitational-waves (GWs), on a single event basis. We show that the requirement that a second-generation (2G) remnant be retained, and subsequently undergo a dynamical encounter, places strong constraints on the mass of the cluster. Using a Plummer model as a readily interpretable baseline, we derive analytic scaling relations between the peak of the inferred host mass posterior, the GW-driven recoil velocity of the remnant, and the parameters that determine the structure of the host. We then perform exact numerical marginalization over thermal and recoil velocities, angles, and cluster structure parameters, to infer the host-mass posterior. We apply our framework to putative hierarchical mergers GW241011 and GW241110, and infer the masses of their hosts on a single-event basis. We find that these are consistent with either heavy globular clusters or nuclear star clusters, with inferred masses spanning $10^{5.7 - 7.7} M_{\odot}$ at $68\%$ confidence depending on the 2G recoil velocity distribution used.

Light-matter interfaces are pivotal for quantum computation and communication. While typically analyzed using single-mode or open-quantum-system approximations, these models often neglect multi-mode field states and light-matter entanglement, hindering exact protocol modeling. Here, we solve the full Hamiltonian dynamics of a solid-state spin-photon interface for three key protocols: the generation of photon-number superpositions, a controlled photon-photon gate, and the production of photonic cluster states. By deriving exact fidelities, we identify fundamental performance limits. Our results reveal that while realistic imperfections severely limit photon-photon gates, they only slightly affect linear photonic clusters and are nearly harmless for photon-number state superpositions.

Quantum processes, including quantum gates and channels, are integral to various quantum information tasks, making the efficient characterization of these processes and their underlying noise critically important. Here, we propose a framework for quantum process tomography in the presence of corrupted noise that is able to simultaneously reconstruct the process and corrupted noise. Firstly, within the Choi-state representation, we derive the corresponding generalized restricted isometry property and demonstrate the simultaneous reconstruction of various quantum gates under sparse noise. Moreover, in comparison with the Choi-state scheme, the process-matrix representation is employed to simultaneously reconstruct sparse noise and a broader range of target quantum gates. Our results demonstrate that significant reduction in experimental configurations is achievable even under corrupted noise.

As a quantum-inspired, non-traditional analog solver architecture, the analog Ising machine (AIM) has emerged as a distinctive computational paradigm to address the rapidly growing demand for computational power. However, the mathematical understanding of its principles, as well as the optimization of its solution speed and accuracy, remain unclear. In this work, we for the first time systematically discuss multiple implementations of AIM and establish a unified mathematical formulation. On this basis, by treating the binarization constraint of AIM (such as injection locking) as a Lagrange multiplier in optimization theory and combining it with a Lyapunov analysis from dynamical systems theory, an analytical framework for evaluating solution speed and accuracy is constructed, and further demonstrate that conventional AIMs possess a theoretical performance upper bound. Subsequently, by elevating the binarization constraint to a control variable, we propose the controllable analog Ising machine (CAIM), which integrates control Lyapunov functions and momentum-based optimization algorithms to realize adaptive sampling-feedback control, thereby surpassing the performance limits of conventional AIMs. In a proof-of-concept CAIM demonstration implemented using an FPGA-controlled LC-oscillator Ising machine, CAIM achieves a twofold speedup and a 7\% improvement in accuracy over AIM on a 50-node all-to-all weighted MaxCut problem, validating both the effectiveness and interpretability of the proposed theoretical framework.

Modified gravity theories offer a well-motivated extension of General Relativity and provide a possible explanation for the late-time accelerated expansion of the Universe. Among them, $f(R)$ gravity represents a minimal and theoretically appealing class, characterized by the Compton wavelength parameter $B_0$, which quantifies deviations from General Relativity. In this work, we explore the capability of future neutral hydrogen (HI) 21 cm intensity mapping (IM) observations to constrain $f(R)$ gravity at low redshifts. We perform Fisher-matrix forecasts for $B_0$ and standard cosmological parameters using upcoming 21 cm IM experiments, including BINGO and SKA1-MID (Band 1 and Band 2), both individually and in combination with Planck cosmic microwave background (CMB) priors. We find that even near-term experiments such as BINGO are able to place nontrivial bounds on $B_0$, $σ(B_0)\simeq 2.27\times 10^{-6}$, while SKA1-MID yields substantially tighter constraints, with SKA Band 2 providing the strongest sensitivity among the considered configurations, $σ(B_0)\simeq 6.37\times 10^{-8}$. We further demonstrate that the combination of low-redshift 21 cm IM data with CMB observations efficiently breaks degeneracies with background cosmological parameters and leads to a significant improvement in the constraints on $B_0$. These results highlight the potential of future HI intensity mapping surveys, in combination with CMB measurements, to provide stringent tests of General Relativity on cosmological scales.

This work presents a theoretical model of boson sampling with optical feedback, in which a subset of the interferometer's output modes is looped back into the input modes. If the bosons are injected periodically into the input modes of the interferometer and optical feedback lines' length match the period of injection, it allows for interference between bosons injected at the consequent time iterations. We propose several methods methods for computing the output photon distributions in both output spacial and temporal modes, including not only standard spatiotemporal mode-unfolding technique, but also the Kraus-operator formalism, and a correlation-tensor-based approach. The two latter approaches help us to reveal that for random interferometers this system evolves to a unique stationary state over time. Because of the existence of the stationary state, we introduce new computational problem \textit{Stationary Distribution Boson Sampling} which appears to be harder than conventional boson sampling problem and contains it as a special case when there are no optical feedback lines.

Quantum field theories (QFTs) including fourth-derivative terms such as the Lee-Wick finite QED and quadratic gravity have a better ultra-violet behavior compared to standard theories with second-derivative ones, but the existence of ghost with negative norm endangers unitarity. Such a ghost in general acquires a pair of complex conjugate masses from radiative corrections whose features are concisely described by the so-called Lee model. Working with the canonical operator formalism of QFTs, we investigate the issue of bound states in the Lee model. We find that the bound states cannot be created from ghosts by contributions of a complex delta function, which is a complex generalization of the well-known Dirac delta function. Since the cause of unitarity violation in the Lee-Wick model is the existence of the complex delta function instead of the Dirac delta function, it is of interest to notice that the violation of the unitarity is also connected to the non-existence of bound states. Finally, the problem of amelioration of the unitarity in quadratic gravity is briefly discussed.

In this work, we investigate how cosmological correlators can be reconstructed by applying the momentum-space dispersion formula to their discontinuities, treating them as functions of momentum variables associated with the corresponding de Sitter Witten diagrams. We focus on conformally coupled and massless polynomial scalar interactions (both IR-divergent and IR-convergent), and consider tree-level de Sitter Witten diagrams. We explicitly utilize the single-cut discontinuity relations, or cutting rules, involving the cosmological correlators recently constructed in arXiv:2512.20720. For diagrams with multiple interaction vertices, we apply the dispersion formula by cutting all internal lines in the diagram one by one, successively, thereby allowing us to reconstruct the full correlator using only lower-point contact-level objects and their discontinuity data, up to contact diagram ambiguities. We also rediscover how the cosmological correlators on the late-time slice of de Sitter space can be obtained from flat-space Feynman diagrams via a set of dressing rules. Our starting point, being the cutting rules for the cosmological correlators, also emphasizes how basic principles, such as unitarity for in-in correlators, can lead us to the dressing rules, which were previously derived in literature following a different method.

Continuous variable quantum key distribution allows two legitimate parties to share a common secret key and encompasses reconciliation protocols. A relatively new reconciliation protocol, Arithmetic Reconciliation, presents low complexity and has increasing reconciliation efficiency with lower SNRs. In this paper, we obtain reconciliation efficiencies for this protocol in realistic scenarios, by means of estimation of mutual information, and we also present rates for sequence match of secret keys by Alice and Bob. Results show that this technique is feasible and promising to continuous variable quantum key distribution applications.

Among the intriguing features of quantum theory, the problem of distinguishing quantum channels is of fundamental interest. In this paper, we focus on the single-shot discrimination of two noisy quantum channels using two distinct classes of probes: single-system (product) probes and entangled probes. Our aim is to identify optimal probing state for specific discrimination tasks and to analyze the necessity and role of entanglement in enhancing channel distinguishability. We show that maximally entangled probes are optimal for discriminating two qubit depolarizing channels, with any nonzero entanglement providing an advantage over single-system probes. In contrast, for dephasing channels in arbitrary dimensions, we prove that single-system probe can be optimal and that entanglement offers no improvement, even when the dephasing unitary is generalized. For qubit amplitude-damping channels, we identify distinct noise-dependent regimes in which either single-system probe outperforms maximally entangled probes and vice-versa. Moreover, we demonstrate that non-maximally entangled probes can act as the optimum probe if the noise parameters restricted to certain values in this task. We also present examples of noisy unitary channels for which discrimination is possible using non-maximally entangled probe, while both single-systems and maximally entangled probes fail. We introduce another class of noisy unitary channels for which perfect discrimination is achievable with a single system, while maximally entangled probes are insufficient. Finally, we show that two erasure channels can be optimally discriminated using any pure single-system probe, with no advantage gained from entanglement.

The no-signalling principle is a fundamental assumption in Bell-inequality and quantum-steering experiments. Nonetheless, experimental imperfections can lead to apparent violations beyond those expected from finite-sample statistics. Here, we propose extensions of local hidden variable and local hidden state theories that allow for bounded, operationally quantifiable, amounts of signalling. We show how non-classicality tests can be developed for these models, both through exact methods based on the full set of observed statistics and through corrections to the standard Bell and steering inequalities. We demonstrate the applicability of these methods via two scenarios that feature apparent signalling: an IBM quantum processor and post-selected data from inefficient detectors.

We review the physical basis for the assumption of locality in relativistic physics and its connection with the aberration of starlight. As a consequence of the hypothesis of locality, the standard relativistic formulas for the Doppler effect and aberration are independent of the polarization of the incident electromagnetic or gravitational radiation. The modification of these formulas due to the helicity-rotation coupling are discussed. In connection with the aberration of polarized radiation, we note that the helicity of radiation incident on a rotating observer couples to its angular velocity of rotation resulting in a slight helicity-dependent modification of the standard aberration formula. We discuss the physical origin of this effect and estimate the magnitude of the helicity aberration.

While energy band topology in spatial photonic crystals (PCs) and momentum-band topology in temporal crystals have each served as powerful probes of topological phases in their respective domains, their unification in a static platform remains unexplored. In this Letter, we bridge this gap by introducing a waveguide assisted non-Hermitian SSH model, in which controlled tuning of loss and coupling drives PT-symmetry breaking and enables a continuous transition between energy- and momentum-gap regimes. This allows us to construct a complete spatiotemporal topological phase diagram in a unified parameter space. By mapping this phase diagram onto a spatially graded PC, we experimentally observe multiple Bloch momentum-band gaps and a continuous spatiotemporal topological transition via translating across the static sample, enabling real-time control over the evolution pathway of the band topology. Our work creates a versatile, bias-free platform for exploring synthetic spacetime physics and opens new avenues for controlling light via non-Hermitian band engineering.

Noisy quantum annealing experiments on Rydberg atom arrays produce measurement outcomes that deviate from ideal distributions, complicating performance evaluation. To enable a data-driven benchmarking methodology for quantum devices that accounts for both solution quality and the classical computational cost of inference from noisy measurements, we introduce deterministic error mitigation (DEM), a shot-level inference procedure informed by experimentally characterized noise. We demonstrate this approach using the decision version of the $k$-independent set problem. Within a Hamming-shell framework, the DEM candidate volume is governed by the binary entropy of the bit-flip error rate, yielding an entropy-controlled classical postprocessing cost. Using experimental measurement data, DEM reduces postprocessing overhead relative to classical inference baselines. Numerical simulations and experimental results from neutral atom devices validate the predicted scaling with system size and error rate. These scalings indicate that one hour of classical computation on an Intel i9 processor corresponds to neutral atom experiments with up to $N=250-450$ atoms at effective error rates, enabling a direct, cost-based comparison between noisy quantum experiments and classical algorithms.

Matchgate unitaries are ubiquitous in quantum computation due to their relation to non-interacting fermions and because they can be used to benchmark quantum computers. Implementing such unitaries on fault-tolerant devices requires first compiling them into a discrete universal gate set, typically Clifford$+T$. Here, we propose a different approach for their synthesis: compile matchgate unitaries using only matchgate gates. To this end, we first show that the matchgate-Clifford group (the intersection of the matchgate and Clifford groups) plus the $\overline{T}$ gate (a $T$ unitary up to a phase) is universal for the matchgate group. Our approach leverages the connection between $n$-qubit matchgate circuits and the standard representation of $\mathbb{SO}(2n)$, which reduces the compilation from $2^n\times 2^n$ unitaries to $2n\times2n$ ones, thus reducing exponentially the size of the target matrix. Moreover, we rigorously show that this scheme is efficient, as an approximation error $\varepsilon_{\mathbb{SO}(2n)}$ incurred in this smaller-dimensional representation translates at most into an $O(n \,\varepsilon_{\mathbb{SO}(2n)})$ error in the exponentially large unitary. In addition, we study the exact version of the matchgate synthesis problem, and we prove that all matchgate unitaries $U$ such that $U\otimes U^*$ has entries in the ring $\mathbb{Z}\big[1/\sqrt 2,i\big]$ can be exactly synthesized by a finite sequence of gates from the matchgate-Clifford$+\overline{T}$ set, without ancillas. We then use this insight to map optimal exact matchgate synthesis to Boolean satisfiability, and compile the circuits that diagonalize the free-fermionic $XX$ Hamiltonian on $n=4,\,8$ qubits.

We review late-time cosmic acceleration after DESI Data Release~2 (DR2), emphasizing the interplay between Type~Ia supernovae (SNe~Ia), anisotropic BAO, CMB calibration, and perturbation-sensitive probes (RSD and weak lensing). DESI DR2 delivers percent-level BAO distance ratios over $0\lesssim z\lesssim2.5$, including a Ly$α$-forest anchor at $z_{\rm eff}=2.33$. In CMB-calibrated combinations, flat $Λ$CDM exhibits a mild parameter mismatch, while allowing evolving dark energy (e.g.\ CPL $w_0$--$w_a$) can improve the fit; the preference is dataset-dependent and is particularly sensitive to redshift-dependent SN calibration/selection residuals at the few$\times10^{-2}$\,mag level. To sharpen likelihood-level interpretation, we provide two diagnostics: (i) an $r_d$-independent BAO-shape observable, $F_{\rm AP}(z)\equiv D_{\rm M}(z)/D_{\rm H}(z)$, constructed directly from published $(D_{\rm M}/r_d,\,D_{\rm H}/r_d)$ with covariance propagation; and (ii) a linear-response map from SN Hubble-diagram systematics $δμ(z)$ to induced biases in $(w_0,w_a)$, yielding an explicit calibration requirement for DESI-era claims of evolving $w(z)$. We synthesize parametric and non-parametric reconstructions of $w(z)$ and $ρ_{\rm DE}(z)$ and map the resulting phenomenology to microphysical dark-energy and modified-gravity models subject to perturbation stability and gravitational-wave propagation constraints.

A multi-objective logistics optimization problem from a real-world supply chain is formulated as a Quadratic Unconstrained Binary Optimization Problem (QUBO) that minimizes cost, emissions, and delivery time, while maintaining target distributions of supplier workshare. The model incorporates realistic constraints, including part dependencies, double sourcing, and multimodal transport. Two hybrid quantum-classical solvers are proposed: a structure-aware informed tree search (IQTS) and a modular bilevel framework (HBS), combining quantum subroutines with classical heuristics. Experimental results on IonQ's Aria-1 hardware demonstrate a methodology to map real-world logistics problems onto emerging combinatorial optimization-specialized hardware, yielding high-quality, Pareto-optimal solutions.

We introduce a high-order dynamical decoupling (DD) scheme for arbitrary system-bath interactions in the weak-coupling regime. Given any decoupling group $\mathcal G$ that averages the interaction to zero, our construction yields pulse sequences whose length scales as $\mathcal{O}(|\mathcal G| K)$, while canceling all error terms linear in the system-bath coupling strength up to order $K$ in the total evolution time. As a corollary, for an $n$-qubit system with $k$-local system-bath interactions, we obtain an $\mathcal{O}(n^{k-1}K)$-pulse sequence, a significant improvement over existing schemes with $\mathcal{O}(\exp(n))$ pulses (for $k=\mathcal{O}(1)$). The construction is obtained via a mapping to the continuous necklace-splitting problem, which asks how to cut a multi-colored interval into pieces that give each party the same share of every color. We provide explicit pulse sequences for suppressing general single-qubit decoherence, prove that the pulse count is asymptotically optimal, and verify the predicted error scaling in numerical simulations. For the same number of pulses, we observe that our sequences outperform the state-of-the-art Quadratic DD in the weak-coupling regime. We also note that the same construction extends to suppress slow, time-dependent Hamiltonian noise.

Quantum metrology harnesses quantum entanglement to improve measurement precision beyond standard quantum limit. Although nonlinear interaction is essential for generating entanglement, during signal accumulation, it becomes detrimental and therefore must be suppressed. To address this challenge, we propose an alternating in-phase and quadrature modulation (AIQM) scheme, designed to operate under a fixed nonlinear interaction. During signal accumulation, our time-interleaved approach sequentially applies the in-phase and quadrature driving fields, thereby eliminating the effects of nonlinear interaction on signal accumulation. Our AIQM scheme achieves better metrological performance than conventional schemes, particularly under strong nonlinear interaction and prolonged signal accumulation, with pronounced robustness against parameter variations. By selectively eliminating and utilizing nonlinear interactions via AIQM, our work enables high-precision and high-accuracy entanglement-enhanced sensing without the need for active control of the nonlinear interaction.

The emergence of the Barren Plateau phenomenon poses a significant challenge to quantum machine learning. While most Barren Plateau analyses focus on single-qubit rotation gates, the gradient behavior of Parameterized Quantum Circuits built from multi-qubit gates remains largely unexplored. In this work, we present a general theoretical framework for analyzing the gradient properties of Parameterized Quantum Circuits with multi-qubit gates. Our method generalizes the direct computation framework, bypassing the Haar random assumption on parameters and enabling the calculation of the gradient expectation and variance. We apply this framework to single-layer and deep-layer circuits, deriving analytical results that quantify how gradient variance is co-determined by the size of the multi-qubit gate and the number of qubits, layers, and effective parameters. Numerical simulations validate our findings. Our study provides a refined framework for analyzing and optimizing Parameterized Quantum Circuits with complex multi-qubit gates.

A Reissner-Nordström black hole (RNBH) enclosed in a cavity is known to be superradiantly unstable to charged scalar perturbations below a critical frequency. Inspired by the emergence of the QCD axion as a prominent dark matter candidate, we construct a model featuring an axion field coupled to an electromagnetic field that undergoes superradiant growth around an RNBH. Utilizing numerical relativity, we achieve stable, long-term evolution of this system and perform a comparative analysis across various parameter spaces. Our comprehensive investigation reveals the formation of a hairy black hole, whose final state is governed by a diverse set of physical parameters. Notably, the decay constant in the axion potential, representing nonlinear interactions, bifurcates the superradiant instability into two distinct behavioral regimes, leading to more significant dynamical shifts than previously reported. Furthermore, we examine the influence of the scalar field's charge and mass, as well as the mirror's position. We investigate the axionic bosenova process and observe a long-term beating pattern of the axion field induced by nonlinear interactions. By fine-tuning these parameter combinations, we demonstrate that the system can evolve toward a variety of distinct physical endpoints.

Semiconductor quantum dots (QDs) have emerged as a premier solid-state platform for the deterministic generation of nonclassical light, offering a compelling pathway toward scalable quantum photonic systems. While single-photon emission from QDs has reached a high level of maturity, the realization of high-fidelity entangled photon-pair sources remains an active and rapidly evolving frontier. In this review, we survey the recent progress in QD-based entangled photon sources, highlighting the conceptual evolution from the established biexciton-exciton cascade to the emerging paradigm of spontaneous two-photon emission. We further examine how advances in nanophotonic architectures and coherent control strategies are redefining fundamental performance limits, enabling concurrent improvements in source brightness, coherence, and entanglement fidelity. Finally, we discuss the key physical and technological challenges that must be addressed to bridge the gap between laboratory demonstrations and large-scale deployment. We conclude by outlining the future opportunities for integrating QD-based entangled photon sources into practical quantum communication, computation, and sensing platforms.

In this paper, we apply the thermodynamic topology framework to ultraspinning Kerr-AdS black holes in arbitrary spacetime dimensions. By constructing the off-shell Helmholtz free energy and the associated vector field, black hole states are characterized as topological defects, and their phase structures are described through zero points, winding numbers, and asymptotic thermodynamic behavior. Analyses of the four- and five-dimensional cases highlight the differences between even- and odd-dimensional configurations, while representative higher-dimensional cases confirm that no additional topological classes or subclasses emerge. We find that only two thermodynamic topological structures appear: the standard class $W^{1+}$ for most configurations, and the distinct subclass $\tilde{W}^{1+}$ for odd-dimensional black holes with maximal rotations. These results establish a unified, dimension-independent classification scheme for ultraspinning Kerr-AdS black holes.

We investigate the scattering of a charged quantum particle in a helically twisted background that induces an effective Coulomb-like interaction, in the presence of an Aharonov-Bohm (AB) flux. Starting from the nonrelativistic Schrödinger equation in the twisted metric, we derive the radial equation and show that, after including the AB potential, it can be mapped onto the same Kummer-type differential equation that governs the planar $2D$ Coulomb $+$ AB problem, with a geometry-induced Coulomb strength and the azimuthal quantum number shifted as $m\to m-λ$. We construct the exact scattering solutions, obtain closed expressions for the partial-wave $S$ matrix and phase shifts, and derive the corresponding scattering amplitude, differential cross section, and total cross section. We also show that the pole structure of the $S$ matrix is consistent with the bound-state quantization previously obtained for the helically twisted Coulomb-like problem.

We quantify the quantum-to-classical transition of the single-mode Kerr nonlinear dynamics in the presence of loss. We establish three time scales that govern the dynamics, each with distinct characteristics. For times short compared to the Ehrenfest time, the evolution is classical, characterized by Gaussian dynamics. For sufficiently long times, as we increase the initial photon number, unitary Kerr evolution would generate macroscopic superpositions of coherent states (so-called kitten states), but this is severely restricted in the presence of small photon loss so that expectation values of observables coincide with their classical values. The intermediate time scale, however, shows resilient quantum behavior in the macroscopic limit. We show that in the mean-field non-Gaussian regime, the Kerr Hamiltonian (with small photon loss) generates a significant amount of Wigner-negativity, and classical flow is recovered only if the loss rate grows with system size. Our results broaden the usual understanding of quantum-to-classical transitions and demonstrate the potential for creating robust nonclassical resources for continuous-variable quantum information processing in the presence of loss.

Determination of the leptonic CP-violating phase $δ_{\mathrm{CP}}$, the atmospheric mixing angle $θ_{23}$, and the mass-squared difference $Δm_{31}^{2}$ constitutes a primary objective of current and next-generation long-baseline neutrino experiments. We employ QFI (QFI) to establish fundamental precision bounds on single-parameter estimation in three-flavor $ν_μ\to ν_e$ oscillations, treating the neutrino as an evolving pure quantum state. Computing QFI as a function of the baseline-to-energy ratio $L/E$ for benchmark parameter sets from NuFit-6.0, we find distinct sensitivity hierarchies and $L/E$-dependent structures. Specifically, $δ_{\mathrm{CP}}$ and $θ_{23}$ exhibit bimodal QFI profiles with peaks at $L/E \sim 500$ and $1500~\mathrm{km/GeV}$, corresponding to the first and second oscillation maxima, reaching $F_Q(δ_{\mathrm{CP}}) \sim 0.15$ and $F_Q(θ_{23}) \sim 15$, respectively. In contrast, $Δm_{31}^{2}$ displays a unimodal structure peaking at $L/E \sim 1000$--$1200~\mathrm{km/GeV}$ with $F_Q(Δm_{31}^{2}) \sim 3 \times 10^{6}$, reflecting its role in setting the oscillation length scale.

Digital signal processing technology has paved the way for the realization of high-speed continuous-variable quantum key distribution systems. However, existing security proofs are limited to static digital signal processing algorithms, while practical systems rely on dynamic multiple-input multiple-output algorithms to compensate for time-varying channel impairments. Our analysis reveals that the conventional dynamic algorithm, due to its non-unitary nature, systematically underestimates the excess noise, which in turn leads to security issues and the generation of insecure keys. To close this gap, we propose a secure algorithm model, mapping the dynamic algorithm to an equivalent physical optical model whose security can be rigorously assessed. Simulations illustrate the algorithm's non-unitary property and provide a quantitative analysis of the excess noise underestimation caused by the conventional algorithm. We further experimentally validate the necessity of the proposed modeling for dynamic digital signal processing, achieving a secret key rate of 14.4 Mbps based on estimated excess noise of 0.07 shot noise unit; whereas the conventional algorithm would have dangerously overestimated the key rate to 28.2 Mbps with noise of 0.008 shot noise unit. This work provides the essential security framework for dynamic digital signal processing, overcoming a critical impediment for the development of high-performance continuous-variable quantum key distribution systems.

High-fidelity coherent population transfer plays a vital role in the realization of quantum memories. However, population transfer with high performance across a broad frequency range is still challenging due to the finite Rabi coupling strength limited by laser powers. Here we propose a novel population-transfer scheme by suturing adiabatic control pulses with each pulse covering certain frequency interval, which are connected in a way that neighboring adiabatic pulses have opposite chirping directions. Taking the widely utilized hyperbolic-square-hyperbolic pulse as an example, we demonstrate that rapid and robust population transfer can be achieved. The transfer bandwidth scales linearly with the number of suture pulses while maintaining high fidelity, even at the suture points where adiabaticity breaks down. Crucially, these pulses can be realized by a single laser by means of temporal multiplexing. For a given bandwidth, this strategy substantially reduces the operational time which is necessary for on demand read-out and suppressing decoherence effects. Our scheme enables a dramatic increase in multimode storage capacity and paves the way for realizing practical quantum networks.

It was recently pointed out that the fundamental mode of the Regge-Wheeler effective potential is unstable against an insignificant Gaussian metric perturbation, which, in turn, might substantially challenge the black hole spectroscopy. This intriguing result has been interpreted by some authors as arising from essentially replacing the black hole's effective potential and its perturbation with two disjoint potential barriers. We argue that such an analysis may have oversimplified the real physical scenario. To be more precise, a metric perturbation planted farther away from the black hole horizon might not always be appropriately approximated by a disjoint minor barrier. Particularly, for the perturbed Pöschl-Teller potential, joint and disjoint metric perturbations might lead to drastically different stability properties for the low-lying modes. Following this line of thought, this study conducts a refined analysis of the stability of the fundamental mode of the Regge-Wheeler effective potential by closely examining a few physically relevant ingredients. While our analysis qualitatively confirms the main findings of previous studies, as the stability of the fundamental mode is primarily determined by the imaginary part of the quasinormal frequency, we show that specific features of both the effective potential at spatial infinity and the metric perturbation can have a sizable impact on the instability. In contrast, the spiral period, governed by the real part of the quasinormal frequency, appears largely insensitive to the details of the black hole metric or its perturbations. The analytic estimates are in reasonable agreement with the numerical results.

We show that a three-dimensional Manin gauge theory with gauge group $\operatorname{SL}(2;\mathbb R)$ (i.e. Yang-Mills theory, the third-way theory, or the imaginary third-way theory) minimally coupled to Einstein gravity admits a dual interpretation as Einstein gravity with an exotic coupling to a Manin gauge theory, where the roles of dreibein/spin connection and field strength/gauge potential are interchanged. The dual, or ersatz, gravitational metric $\hat g_{μν}\sim\operatorname{tr}((\star F)_μ(\star F)_ν)$ is a classical double copy of the gauge field strength $F_{μν}$ (as opposed to the usual double copy of the gauge potential $A_μ$). If matter exclusively couples to $\hat g$ (for example, in a gravitational decoupling limit), then one can formulate black-hole thermodynamics with regards to the ersatz metric. In particular, a black-hole solution for the ersatz metric $\hat g_{μν}$ (made of Yang-Mills fields) radiates ersatz Hawking radiation and obeys the laws of black-hole thermodynamics.

Neutron star binary mergers are often associated with short gamma-ray bursts (GRBs), but the recent detection of kilonovae coincident with long GRBs suggest that some mergers may produce long GRBs. Motivated by these developments, we perform a search for binary neutron star and neutron star-black hole gravitational-wave signals coincident with long GRBs using data from the third LIGO--Virgo--KAGRA (LVK) observing run. We analyze LVK data coincident with long GRBs detected by Fermi's GRB Monitor and Swift's Burst Alert Telescope when at least two gravitational-wave observatories were running. We find no evidence of a coincident gravitational-wave signal and set limits on the luminosity distance to each of these long GRBs under the assumption that they were powered by binary mergers.

This work introduces and characterizes quantum sequential circuits (QSCs) as a hardware-oriented paradigm for quantum computing, built upon a novel foundational element termed the quantum transistor. Unlike conventional qubit-based architectures, QSCs employ symmetry-protected topological junctions where quantum gates are encoded as Choi states via channel-state duality and activated through bulk measurements, utilizing ebits to realize the functional analog of feedback loops in classical sequential circuits. This framework establishes a universal model for quantum computation that inherently incorporates memory and temporal sequencing, complementing existing combinational quantum circuit model. Our work advances the conceptual bridge towards a quantum von Neumann architecture, underscoring the potential of hybrid and modular design principles for the development of large-scale, integrated quantum information processors.

QASMTrans is a lightweight, high-performance, C++-based quantum compiler that bridges abstract quantum algorithms to device-level control and is designed for just-in-time (JIT) deployment on QPU testbeds with tightly integrated FPGAs or CPUs. We focus on achieving fast transpilation times on circuits of interest, we find more than 100x faster compilation than Qiskit in some circuits with similar circuit quality, enabling transpilation of large, high-depth circuits in seconds. Unlike existing tools, QASMTrans offers end-to-end device-pulse compilation and direct quantum control integration with QICK, closing the gap between logical circuits and hardware control enabling closed-loop optimization. QASMTrans supports latency-aware Application-tailored Gate Sets (AGS) at the pulse level, identifying high-impact gate sequences on the circuit critical path and synthesizing optimized pulse schedules using pre-defined robust circuit ansatz. Validated through integrated QuTiP pulse-level simulation, this is found to significantly reduce execution latency and can improve final-state fidelity by up to 12% in some tested circuits. QASMTrans further implements device-aware, noise-adaptive transpilation that uses device calibration data for circuit placement on high-quality qubits and can focus on the circuit critical path to reduce transpilation-pass time while maintaining comparable fidelity. Additionally, it introduces circuit space sharing via calibration-aware device partitioning, enabling concurrent execution of multiple circuits or shots on a single QPU. Moreover, QASMTrans is entirely self-contained and has no external library dependencies, making it easy for practical deployment. By combining fast compilation, pulse-level control, and noise-aware optimization, QASMTrans enables real-time adaptive algorithms such as ADAPT-VQE and ADAPT-QAOA.

We advocate for a holographic definition of thermodynamic pressure and volume for black holes based on quasi-local gravitational thermodynamics. When a black hole is enclosed by a finite timelike boundary, York's quasi-local first law includes a surface pressure conjugate to the boundary area. Assuming the existence of a holographically dual theory living on this boundary, these geometric quantities correspond to the pressure and volume of the dual thermal system. In this work we focus on static, spherically symmetric black holes, for which these quantities reduce to global thermodynamic variables. The holographic volume provides a notion of system size, allowing extensivity to be defined in standard thermodynamic terms, and it yields a definition of the large-system limit. For the asymptotically flat case, we show that, in the canonical thermodynamic representation, small Schwarzschild black holes are non-extensive, whereas large black holes become extensive in the large-system limit. A similar conclusion applies to Anti-de-Sitter Schwarzschild black holes, with the difference that the quasi-local energy of the large black hole also becomes extensive in the large-system limit. Before this limit, the energy decomposes into subextensive and extensive contributions, and we derive an explicit expression for the extensive part as a function of the finite volume and entropy.

We present a new framework for imaging and sensing based on utilizing a quantum computer to coherently process quantum information in an electromagnetic field. We describe the framework, its potential to provide improvements in imaging and sensing performance and present an example application, the design of coherent receivers for optical communication. Finally, we go over the improvements in quantum technologies required to fully realize quantum computational imaging and sensing.

Nonlinear spectroscopic techniques using entangled photon pairs provide an opportunity to exploit non-classical correlations encoded in two-photon wavefunctions to manipulate two-exciton wavefunctions. We propose an entangled photon pair-enhanced multidimensional spectroscopic technique which is sensitive to exciton-exciton interactions and correlations at the ultrafast timescale. Simulations for a dissipative photosynthetic aggregate reveal the superior ability of entangled photon pairs, compared to both transform-limited and frequency-chirped laser pulses, to manipulate excited-state absorption pathways. The corresponding spectral features in the two-dimensional spectrogram are interpreted in terms of one- and two-exciton resonances. The signal scales linearly with the incoming intensity of the photon sources. It is argued that classifying these resonances using entangled photon source at the perturbative limit allow for probing exciton correlations at the natural energy scale. These insights can be used to explore multi-exciton dynamics using multiphoton entanglement.

Hamiltonian representations based on the sum-of-squares (SOS) hierarchy provide rigorous lower bounds on ground-state energies and facilitate the design of efficient classical and quantum simulation algorithms. This work presents a unified framework connecting SOS decompositions with variational two-particle reduced density matrix (v2RDM) theory. We demonstrate that the ``weighted'' SOS ansatz naturally recovers the dual of the v2RDM program, enabling the strict enforcement of symmetry constraints such as particle number and spin. We provide explicit SOS constructions for the Hubbard model and electronic structure Hamiltonians, ranging from spin-free approximations to full rank-2 expansions. We also highlight theoretical connections to block-invariant symmetry shifts. Numerical benchmarks on molecular systems and Iron-Sulfur clusters validate these near frustration-free representations, demonstrating their utility in improving spectral gap amplification and reducing block encoding costs in quantum algorithms.

We analyze in some detail the years leading up to Albert Einstein's visit to Argentina, from the moment he rose to fame after the publication of the results of the 1919 solar eclipse, until his arrival at the port of Buenos Aires in March 1925. We will see that this major event in the country's cultural life did not arise from a vacuum but was the product of diverse interests and unique international and local circumstances. The social and political situation surrounding the eminent scientist and the impetus provided by individuals and institutions in Buenos Aires played a decisive role, ultimately converging in the "successful" reception of this extraordinary figure in a country on the scientific periphery of the first quarter of the 20th century. We also review the written references and knowledge circulating in the local milieu on topics related to the theory of relativity in the years prior to his arrival in the country.

In this paper we analyze the conformal Einstein equations to all orders at null infinity without imposing any restriction on the spacetime dimension, the topology of $\mathscr{I}$, or fall-off conditions for the Weyl tensor. In particular, we study how the equations constrain the geometry of null infinity when it is assumed to be foliated by cross-sections, not necessarily spheres. Our approach is coordinate-free and treats the conformal factor $Ω$ as a dynamical variable. After identifying the free data at $\mathscr{I}$, we show that any two asymptotically flat spacetimes sharing the same free data at null infinity are necessarily isometric to infinite order. In addition, we provide a detached definition of null infinity and prove an existence theorem for asymptotically flat spacetimes solving the field equations to infinite order at $\mathscr{I}$ realizing the prescribed initial data.

In contrast to classical cryptography, where the security of encoded messages typically relies on the inability of standard algorithms to overcome computational complexity assumptions, Quantum Key Distribution (QKD) can enable two spatially separated parties to establish an information-theoretically secure encryption, provided that the QKD protocol is underpinned by a security proof. In the last decades, security proofs robust against a wide range of eavesdropping strategies have established the theoretical soundness of several QKD protocols. However, most proofs are based on idealized models of the physical systems involved in such protocols and often include assumptions that are not satisfied in practical implementations. This mismatch creates a gap between theoretical security guarantees and actual experimental realizations, making QKD protocols vulnerable to attacks. To ensure the security of real-world QKD systems, it is therefore essential to account for imperfections in security analyses. In this article, we present an overview of recent analytical and numerical developments in QKD security proofs, which provide a versatile approach for incorporating imperfections and re-establishing the security of quantum communication protocols under realistic conditions.

Recent advances in neural rendering, particularly 3D Gaussian Splatting (3DGS), have enabled real-time rendering of complex scenes. However, standard 3DGS relies on spherical harmonics, which often struggle to accurately capture high-frequency view-dependent effects such as sharp reflections and transparency. While hybrid approaches like Viewing Direction Gaussian Splatting (VDGS) mitigate this limitation using classical Multi-Layer Perceptrons (MLPs), they remain limited by the expressivity of classical networks in low-parameter regimes. In this paper, we introduce QuantumGS, a novel hybrid framework that integrates Variational Quantum Circuits (VQC) into the Gaussian Splatting pipeline. We propose a unique encoding strategy that maps the viewing direction directly onto the Bloch sphere, leveraging the natural geometry of qubits to represent 3D directional data. By replacing classical color-modulating networks with quantum circuits generated via a hypernetwork or conditioning mechanism, we achieve higher expressivity and better generalization. Source code is available in the supplementary material. Code is available at https://github.com/gwilczynski95/QuantumGS

Searches for late--time gravitational--wave echoes following compact binary mergers are often interpreted as probes of exotic near--horizon or horizonless physics. We show that a well--motivated class of ultraviolet--finite quantum gravity and modified gravity theories, those based on diffeomorphism--invariant, analytic, entire--function nonlocality, generically suppress observable echo signals. The suppression follows from Paley--Wiener bounds associated with the analyticity of the nonlocal regulator, which enforce exponential damping of high--frequency reflection amplitudes and smear sharp effective potentials in the black--hole interior. We demonstrate that for regular black holes with horizons, the standard ingoing boundary condition eliminates the cavity required for echoes, while for regular horizonless compact objects the nonlocal kernel strongly attenuates both inner and photon--sphere reflections. As a result, repeated reflections during the ringdown phase are exponentially suppressed in frequency space and washed out in the time domain. Our results imply that the absence of echoes in current gravitational--wave data is consistent with covariant nonlocal gravity theories and does not, by itself, favor classical horizons over regular or horizonless ultraviolet completions. The analyticity and Paley--Wiener bounds that enforce echo suppression are required to ensure unitarity, ghost freedom, and ultraviolet finiteness of the underlying quantum theory. As a result, the classical ringdown phenomenology already encodes quantum--gravity consistency conditions. We emphasize that the absence of echoes in such theories does not rely on quantum decoherence or stochastic effects, but instead reflects the analytic structure imposed by quantum gravity on the classical limit.

We implement a Markov Chain Monte Carlo method to obtain posterior probability distributions for the parameters of the S2 star orbit around Sagittarius A*, for seven representative non-rotating black hole space-time solutions. In particular, we consider the Schwarzschild, Reissner-Nordström, Janis-Newman-Winicour, and Bardeen black hole space-times from General Relativity, as well as a black hole solution from Einstein-Maxwell-dilaton gravity, a hairy black hole solution from Horndeski theory, and a Yukawa-like black hole from $f(\mathscr{R})$ gravity. To constrain model parameters, we use the most recent publicly available observational data of the S2 star orbit, namely astrometric measurements, spectroscopic data, and the pericentre advance measured by the GRAVITY Collaboration. We further perform a consistent Bayesian comparison of models, calculating the log-Bayes factor of each space-time with respect to the Schwarzschild solution. Our results show that the currently available data indicate no statistically significant preference among the space-times considered. The Bardeen and Yukawa-like models are indistinguishable from Schwarzschild within current uncertainties, while the Reissner-Nordström, Janis-Newman-Winicour, Horndeski and Einstein-Maxwell-dilaton geometries show at most weak and non-decisive strength of evidence under the adopted priors and likelihood choices.

We analyze cosmological particle production driven by spacetime expansion in the early universe for homogeneous and isotropic cosmologies with positive, negative, and zero spatial curvature. We prioritize analytical results to gain a deeper understanding of curvature-induced effects. Specifically, for a conformally coupled scalar field, we model the inflationary epoch as an exact de Sitter phase followed by a transition to a static universe. Both instantaneous and smooth exits from inflation are considered, the latter being implemented via the adiabatic vacuum prescription. Starting from an initial Bunch-Davies vacuum, we derive the associated mode functions carefully adapted to each curvature sign. Using the Bogoliubov formalism, we non-perturbatively compute the number density of produced scalar particles. Our results demonstrate that spatial curvature significantly impacts the resulting particle spectra, particularly for light fields, where the deviation from the flat-space scenario is most prominent and can reach several orders of magnitude

Decoy-state QKD with phase-randomized weak coherent pulses is typically analyzed assuming independent, precisely prepared intensities. Real sources, however, can exhibit correlated intensity drift across rounds, potentially leaking intensity information and breaking the standard decoy-state reduction to linear programs. Cauchy--Schwarz (CS) constraints can restore security by coupling $n$-photon yields across intensities, but they introduce nonlinear square-root constraints that are commonly handled via outer linearisation around channel-model-based reference points. We propose a reproducible alternative: first solve the full CS-constrained parameter-estimation problems using the interior-point nonlinear solver IPOPT, then use the resulting candidate solution as the linearisation point for the outer optimisation that certifies a valid lower bound on the asymptotic key rate. Simulations for both coarse-grained model-independent correlations and fine-grained truncated-Gaussian models show consistently tighter key-rate bounds than canonical reference points, and in some cases allow certifying optimality when both optimisation stages coincide.

Although in general boundary conditions do not affect the bulk properties of a system, some of them are special and defy such expectation. This is the case, for instance, of those inducing geometrical frustration in a classical magnet. Recently, the study of such settings in quantum systems (dubbed topological frustration) has uncovered peculiar features, interesting both from a fundamental and technological point of view. In this work, we present and discuss the behavior of several quantum resources in presence of TF, namely the (disconnected) entanglement entropy and the non-stabilizerness Renyi entropy. We will show that, compared to their non-frustrated counterparts, TF adds a distinct contribution to these resources, due to a stable, delocalized, topological excitation. Remarkably, this contribution can be calculated analytically, due to its similarities with that of a W-state.

Simulating quantum many-body systems (QMBS) is one of the long-standing, highly non-trivial challenges in condensed matter physics and quantum information due to the exponentially growing size of the system's Hilbert space. To date, tensor networks have been an essential tool for studying such quantum systems, owing to their ability to efficiently capture the entanglement properties of the systems they represent. One of the well-known tensor network architectures, namely matrix product states (MPS), is the standard method for simulating one-dimensional QMBS. Here, we propose a simple, yet efficient, method to augment the already available MPS algorithms to simulate the dynamics of time-dependent Hamiltonians with better accuracy and a faster convergence rate, giving a second-order convergence compared to the first-order convergence of the standard method. We apply our proposed method to simulate the dynamics of a chain of single spins associated with nitrogen-vacancy color centers in diamonds, which has potential applications for practical and scalable quantum technologies, and find that our method improves the average error for a system of few NV centers by a factor of about 1000 for moderate step sizes. Our work paves the way for efficient simulation of QMBS under the influence of time-dependent Hamiltonians.

We study Oppenheimer-Snyder (OS) gravitational collapse matched to a general static, spherically symmetric exterior spacetime. Unlike the Schwarzschild case, two new features can arise in black holes with two horizons: an apparent-horizon left vertex, a temporary minimum in the apparent-horizon radius during collapse, and a bounce, where the star surface stops collapsing at a nonzero radius and reverses into expansion. We identify the conditions that lead to these two features. For two-horizon exteriors, trapped-region consistency requires that the apparent-horizon turning point occurs no earlier than the surface crossing of the inner horizon. As a concrete example, the OS collapse of the Reissner-Nordström (RN) spacetime shows both effects. In contrast, regular black holes with de Sitter cores show neither: their collapse is smooth and monotonic, and the surface approaches the center only as the proper time goes to infinity. These results naturally classify the OS collapses into three categories: singular, which ends at the center in finite time; bouncing, which reverses at a finite radius; and soft-landing, which reaches the center only asymptotically. We argue that these features are consistent with Penrose's strong cosmic censorship conjecture.

In thermodynamics, a Legendre transformation of the free energy provides a mapping between different statistical ensembles. In this work, we demonstrate that performing a Legendre transformation of the black hole on-shell action is equivalent to imposing different boundary conditions on the fields. Consequently, the choice of ensemble must be consistent with, and cannot contradict, the imposed boundary conditions. From this perspective, it follows that for four-dimensional dyonic black holes, the on-shell action can only be expressed either as a function of the electric charge and the magnetic potential, or alternatively as a function of the magnetic charge and the electric potential. Inspired by the Legendre transformation of the Maxwell field, we argue that for purely gravitational theories whose metric geometries admit a \(U(1)\) fiber bundle structure, i.e.\ rotating, boosted, or Kaluza-Klein monopole configurations, one can similarly introduce appropriate Legendre terms, in the sense of dimensional reduction, to modify the thermodynamic ensemble of the black hole. Within the dimensional reduction framework, we study the on-shell action of black holes in five-dimensional minimal supergravity with a Chern-Simons term, analyze the corresponding Legendre transformation procedure, and show how the resulting formulation remains consistent with the Wald formalism.

In this work we study the relationship between quantum random walks on graphs and Krylov/spread complexity. We show that the latter's definition naturally emerges through a canonical method of reducing a graph to a chain, on which we can identify the usual Krylov structure. We use this identification to construct families of graphs corresponding to special classes of systems with known complexity features and conversely, to compute Krylov complexity for graphs of physical interest. The two main outcomes are the analytic computation of the Lanczos coefficients for the SYK model for an arbitrary number $q$ of interacting fermions and the complete characterization of Krylov complexity for the hypercube graph in any number of dimensions. The latter serves as the starting point for an in-depth comparison between Krylov and circuit complexities as they purportedly arise in the context of black holes. We find that while under certain conditions Krylov complexity follows the growth and saturation pattern ascribed to such systems, the timescale at which saturation happens can generally be shorter than what is predicted by random unitary circuits, due to the effects of quantum speed-ups commonly occurring when comparing quantum and classical random walks.

Tidal interaction is a unique, detectable signature in gravitational wave signals from inspiraling binary neutron stars (BNSs), which can be used to constrain the neutron star (NS) equation of state (EoS). The tidal interaction is resonantly amplified as the orbital frequency approaches the NS fundamental mode (f-mode) frequency. It has been shown that the exclusion of tidal resonance in parameter estimation leads to a significant bias in the inferred NS tidal deformability and hence the NS EoS [Pratten et al. PRL 129, 081102 (2022)]. The strength and location of tidal resonance depend sensitively on the f-mode frequency, which is typically modeled using its universal relation with the tidal deformability that is derived for an isolated, non-spinning NS assuming only linear fluid perturbations. In a BNS inspiral, the f-mode frequency can be corrected by at least three known effects: nonlinear hydrodynamics, background spin, and relativity. We use Hamiltonian Monte Carlo simulations to estimate the systematic bias on tidal deformability when each frequency correction is ignored. Our study considers both loud, individual events and the stacking of a population of detections. Both scenarios are expected when the next-generation detectors are available with a sensitivity level increased by about an order of magnitude.

Cold atoms have become a powerful platform for quantum-simulating lattice gauge theories in higher spatial dimensions. However, such realizations have been restricted to the lowest possible truncations of the gauge field, which limit the connections one can make to lattice quantum electrodynamics. Here, we propose a feasible cold-atom quantum simulator of a $(2+1)$-dimensional U$(1)$ lattice gauge theory in a spin $S=1$ truncation, featuring dynamical matter and gauge fields. We derive a mapping of this theory onto a bosonic computational basis, stabilized by an emergent gauge-protection mechanism through quantum Zeno dynamics. The implementation is based on a single-species Bose--Hubbard model realized in a tilted optical superlattice. This approach requires only moderate experimental resources already available in current ultracold-atom platforms. Using infinite matrix product state simulations, we benchmark real-time dynamics under global quenches. The results demonstrate faithful evolution of the target gauge theory and robust preservation of the gauge constraints. Our work significantly advances the experimental prospects for simulating higher-dimensional lattice gauge theories using larger gauge-field truncations.

The quantum Fisher information of a quantum state with respect to a certain parameter quantifies the sensitivity of the quantum state to changes in that parameter. Maximizing the quantum Fisher information is essential for achieving the optimal estimation precision of quantum sensors. A typical quantum sensor involves a qubit(e.g. a spin-1/2) probe undergoing an unknown rotation, here the unknown rotation angle is the parameter to be estimated. A well known limitation is that if the rotation axis is unknown, the maximal quantum Fisher information is impossible to attain. This limitation has been lifted recently by leveraging entanglement between the probe qubit and an ancilla qubit. Namely, through measurement of the ancilla after the axis is revealed, one can prepare the probe that is optimal for any unknown rotation axis. This proposal, however, works only for a spin-1/2. Considering large spin probes can achieve a larger quantum Fisher information, offering enhanced metrological advantage, we here utilize the entanglement between a large spin probe and an ancilla to achieve optimal quantum Fisher information for estimating the rotation angle, without prior knowledge of the rotation axis. Different from the previous spin-1/2 case, achieving the optimal precision with large spins generally requires post-selection, resulting in a success probability dependent on the dimension of the Hilbert space. Furthermore, we extend the encoding state from the maximally entangled case to general entangled states, showing that optimal metrology can still be achieved with a certain success probability.

The standard method for benchmarking quantum error-correction is randomized fault-injection testing. The state-of-the-art tool \stim is efficient for error correction implementations with distances of up to 10, but scales poorly to larger distances for low physical error rates. In this paper, we present a scalable approach that combines stratified fault injection with extrapolation. Our insight is that some of the fault space can be sampled efficiently, after which extrapolation is sufficient to complete the testing task. As a result, our tool scales to distance 17 for a physical error rate of 0.0005 with a two-hour time budget on a desktop. For this case, it estimated a logical error rate of $1.51 \times 10^{-11}$ with high confidence.

Large language models (LLMs) can generate structured artifacts, but using them as dependable optimizers for scientific design requires a mechanism for iterative improvement under black-box evaluation. Here, we cast quantum circuit synthesis as a closed-loop, test-time optimization problem: an LLM proposes edits to a fixed-length gate list, and an external simulator evaluates the resulting state with the Meyer-Wallach (MW) global entanglement measure. We introduce a lightweight test-time learning recipe that can reuse prior high-performing candidates as an explicit memory trace, augments prompts with a score-difference feedback, and applies restart-from-the-best sampling to escape potential plateaus. Across fixed 20-qubit settings, the loop without feedback and restart-from-the-best improves random initial circuits over a range of gate budgets. To lift up this performance and success rate, we use the full learning strategy. For the 25-qubit, it mitigates a pronounced performance plateau when naive querying is used. Beyond raw scores, we analyze the structure of synthesized states and find that high MW solutions can correspond to stabilizer or graph-state-like constructions, but full connectivity is not guaranteed due to the metric property and prompt design. These results illustrate both the promise and the pitfalls of memory evaluator-guided LLM optimization for circuit synthesis, highlighting the critical role of prior human-made theoretical theorems to optimally design a custom tool in support of research.

Quantum generative models offer a novel approach to exploring high-dimensional Hilbert spaces but face significant challenges in scalability and expressibility when applied to multi-modal distributions. In this study, we explore a Hybrid Quantum-Classical U-Net architecture integrated with Adaptive Non-Local Observables (ANO) as a potential solution to these hurdles. By compressing classical data into a dense quantum latent space and utilizing trainable observables, our model aims to extract non-local features that complement classical processing. We also investigate the role of Skip Connections in preserving semantic information during the reverse diffusion process. Experimental results on the full MNIST dataset (digits 0-9) demonstrate that the proposed architecture is capable of generating structurally coherent and recognizable images for all digit classes. While hardware constraints still impose limitations on resolution, our findings suggest that hybrid architectures with adaptive measurements provide a feasible pathway for mitigating mode collapse and enhancing generative capabilities in the NISQ era.

Quantum Error Correction (QEC) is essential for building robust, fault-tolerant quantum computers; however, the decoding process often presents a significant computational bottleneck. Tesseract is a novel Most-Likely-Error (MLE) decoder for QEC that employs the A* search algorithm to explore an exponentially large graph of error hypotheses, achieving high decoding speed and accuracy. This paper presents a systematic approach to optimizing the Tesseract decoder through low-level performance enhancements. Based on extensive profiling, we implemented four targeted optimization strategies, including the replacement of inefficient data structures, reorganization of memory layouts to improve cache hit rates, and the use of hardware-accelerated bit-wise operations. We achieved significant decoding speedups across a wide range of code families and configurations, including Color Codes, Bivariate-Bicycle Codes, Surface Codes, and Transversal CNOT Protocols. Our results demonstrate consistent speedups of approximately 2x for most code families, often exceeding 2.5x. Notably, we achieved a peak performance gain of over 5x for the most computationally demanding configurations of Bivariate-Bicycle Codes. These improvements make the Tesseract decoder more efficient and scalable, serving as a practical case study that highlights the importance of high-performance software engineering in QEC and providing a strong foundation for future research.

The Euclidean Gravitational Path Integral has proven remarkably effective in the quantum regime of black hole physics. In this work, we examine the applicability of the Kontsevich-Segal-Witten (KSW) criterion for admissible complex metrics in the context of the Euclidean Gravitational Path Integral. We find that, for the super-conformal index of ${\cal N}=4$ SYM with unequal angular momenta, the black hole saddle points violate the KSW criterion precisely where the statistical description of the index breaks down. The corresponding critical point coincides with a phase transition into two-component ``grey galaxy'' configurations in the micro-canonical ensemble.

We present a new Monte-Carlo generator for the simulation of di-jet production in polarized proton-proton collisions at the next-to-leading order in QCD matched to parton showers using the framework of the POWHEG BOX. With this program we compute a variety of observables of immediate relevance for the spin program of the Relativistic Heavy Ion Collider at Brookhaven National Laboratory. While parton-shower effects are generally small, we find that in some search regions their inclusion improves agreement of predictions with data. Moreover, we provide a critical assessment of selection criteria applied in experiment in the light of perturbative stability.

In this work, we discuss recent results obtained with the application of the medium separation scheme (MSS) in different contexts where a clear violation of the conformal limit for the speed of sound at finite density has been observed in Quantum Chromodynamics (QCD). We analyze several scenarios, including QCD at finite isospin density, two-color QCD, and two-flavor color superconductivity. Whenever possible, we compare our findings with lattice QCD (LQCD) results, showing that the Nambu--Jona-Lasinio (NJL) model combined with the MSS provides a consistent description across different regimes of the QCD phase diagram. Our analysis highlights how effective models, when properly regularized, can capture essential nonperturbative features of dense QCD matter, offering complementary insights to lattice simulations.

We study the correlation functions of a conformally coupled $φ^4$-interacting theory in AdS$_3$ and its dual CFT$_2$. The one-loop diagram is not expressible in terms of known transcendental functions, but is shown to be expressible as an infinite sum of previously well-studied tree-level diagrams, and we compute this sum using several number-theoretic conjectures. This enables us to extract recursively, the analytic expressions of anomalous dimensions of all dual double-trace operators. In the $s$-channel various consistency checks were performed against established bootstrap method, while our results in the $t$- and $u$-channel are not available in any previous literature to our knowledge.

We give a physical interpretation for the analytic continuation of the partition function of superconformal SU$(2)$ $\mathcal{N}=2$ gauge theory on the four-sphere to all values of the Yang-Mills coupling. We show that a well-motivated 2d construction associates two-dimensional unstable instantons to the 4d complex saddles which appear as singularities in the integrand of the supersymmetric localization expression. The construction is based on the chiral algebra subsector, and aligns with the alternative Higgs branch localization.

Here we add a U(1) gauge theory to the simple EP exotic invariant model in the paper E4. This paper E5 is the fifth in a series of papers En.

We investigate the impact of a nonstandard electron mass $m_e$ on early-Universe thermal history, focusing on neutrino decoupling and Big Bang Nucleosynthesis (BBN). In the standard cosmology, neutrino--electron interactions keep neutrinos in thermal contact with the electromagnetic plasma until shortly before $e^\pm$ annihilation. Varying $m_e$ shifts the decoupling epoch and the entropy transfer from $e^\pm$ annihilation, thereby modifying the neutrino energy density and the inferred effective number of relativistic species, $N_{\mathrm{eff}}$. Independently, during BBN the rates of charged-current weak processes, and hence the neutron-to-proton ratio, depend on $m_e$. By confronting BBN predictions for the primordial light-element abundances with observations and imposing cosmological constraints on $N_{\mathrm{eff}}$, we obtain a bound on $m_e$ in the early Universe of $m_e = 0.504^{+0.007}_{-0.006}$ MeV or $m_e=0.510\pm0.007$ MeV ($1σ$), depending on the considered nuclear reaction network (NACRE II or PRIMAT, respectively). The allowed range is close to the present laboratory value at the level of 1.4\%, thus supporting the constancy of the electron mass over cosmological timescales.

In this paper we compute explicitly the Unitary Irreducible Representations of the Poincaré and Euclidean groups in dimension D=2, following, step by step, Mackey's theorem of induced representations.

The difference between the charged and neutral pion masses can be predicted from a well-known dispersion relation involving an infinite-energy integral over experimental data, the pion sum rule. This relation, however, holds only in the chiral limit. Here we combine several nonperturbative techniques to determine the form of the physical finite-energy integral, thereby generalizing this sum rule to include effects linear in the light-quark masses. We test the dispersion relation using hadronic tau-decay data.

We formulate within Palatini Supergravity a model of induced-gravity inflation excellently consistent with ACT DR6. The inflaton belongs in the decomposition of a conjugate pair of Higgs superfields which lead to the spontaneous breaking of a U(1)B-L symmetry at a scale close to the range (0.102-5.85)x10^16 GeV. The inflaton field is canonically normalized thanks to a real and shift-symmetric contribution into the Kaehler potential. It also includes two separate holomorphic and antiholomorphic logarithmic terms, the argument of which can be interpreted as the coupling of the inflaton to the Ricci scalar. The attainment of inflation allows for subplanckian inflaton values and energy scales below the cut-off scale of the corresponding effective theory. Embedding the model in a B-L extension of the MSSM we show how the mu parameter can be generated and non-thermal leptogenesis can be successfully realized. An outcome of our scheme is split SUSY with gravitino mass in the range (40-60) PeV, which is consistent with the results of LHC on the Higgs boson mass.

Within the framework of nonrelativistic QCD (NRQCD) factorization,and based on the Collins--Soper operator definition of fragmentation functions, we present a systematic calculation of the fragmentation functions for a heavy quark fragmenting into color-singlet $P$-wave quarkonium states. After reproducing and confirming the known leading-order results, we further compute the relativistic corrections up to order $\mathcal{O}(v^{2})$. Our analysis applies both to quarkonium systems composed of heavy quarks with the same flavor and to $B_c$-type mesons formed by heavy quarks of different flavors. Numerical results show that, for all color-singlet $P$-wave channels, the $\mathcal{O}(v^{2})$ relativistic corrections give sizable negative contributions over most of the momentum-fraction $z$ region. We further compute inclusive cross sections for $P$-wave quarkonium plus charmed hadrons in $e^+e^-$ annihilation via the single photon process up to $\mathcal{O}(v^{2})$ by applying our obtained fragmentation functions, and the resulting predictions are consistent with the full fixed-order results in the high-energy region.

We uncover surprising hidden simplicity in Mellin amplitudes for tree-level AdS holographic correlators for spinning operators, such as AdS "gluons" and "gravitons" (spin 1 and 2). We define Mellin amplitudes with $n$ spinning operators via the so-called "scaffolding" of $2n$-scalar ones with specific projection operators for each spin state, which are rational functions of Mellin variables of $2n$ scalars generalizing flat-space scaffolding amplitudes. We classify possible three-point structures with spin 1 and 2 which take the same form as massive three-point amplitudes in flat space, and match with special solutions such as those extracted from 6-scalar ones in $\mathrm{AdS}_5\times S^3$ or $\mathrm{AdS}_5\times S^5$. Focusing on $\mathrm{AdS}_5$ gluons, we directly bootstrap spinning amplitudes in scaffolding form up to $n=6$ gluons (which amounts to $2n=12$ scalars) using factorizations, multi-linearity and flat-space limit. The results take a remarkably simple form in analogy with flat-space amplitudes, which can be constructed from familiar 3- and 4-vertices as well as propagators of massive spin-1 particles. Surprisingly, we find that vertices with any descendant levels are proportional to the primary ones with nice combinatorial coefficients, which makes manifest the correct flat-space limit in the simplest possible way.

The determination of neutrino mass ordering (NMO) is the primary goal of the currently running JUNO reactor experiment. We show that the measurement of NMO at JUNO may severely deteriorate in the presence of non-standard neutrino interactions mediated by a beyond standard model scalar (SNSI). Taking inverted ordering and the lightest neutrino mass at $m_l=0.01$ eV, the NMO sensitivity falls below $2σ$ for SNSI parameter values in the range $η_{ee}< -7.1\times 10^{-3}$ and $η_{ee} > 3.3\times 10^{-3}$. More importantly, for $η_{ee} \gtrsim 5.7\times 10^{-3}$ the NMO sensitivity in JUNO is completely lost. We show that this is due to the presence of a hitherto unrecognized resonant enhancement of the mixing angle $θ_{12}$, which gives rise to a mass ordering degeneracy.

We investigate the finite-temperature SU(2) Savvidy model under an imaginary angular velocity. Employing the background-field method, we derive the one-loop effective potential and analyze both its real and imaginary parts. We demonstrate that imaginary rotation modifies the chromomagnetic condensate and the Polyakov loop, and can partially suppress the Nielsen-Olesen instability of the chromomagnetic background. Moreover, a high-temperature expansion shows that imaginary rotation strengthens the effective coupling and that the chromomagnetic field induces a negative contribution to the moment of inertia.

We perturbatively investigate the rotation effect on the Polyakov loop potential in SU(2) gauge theroy within a chromomagnetic background. It is observed that the imaginary rotation spontaneously induces both confinement and chromomagnetic condensation at high temperatures, thereby provides a perturbative window to explore non-perturbative dynamics. Compared to the case without including the induced chromomagnetic field, the perturbative confinement transition becomes first-order, with a temperature-dependent phase boundary that asymptotically approaches $\tildeΩ_c = π/\sqrt{3}$ at high temperatures. This leads to a significantly enriched $\tildeΩ$-$T$ phase diagram characterized by an expanded deconfined region. For real angular velocities, we find that the chromomagnetic condensate decreases with increasing rotation, and that the coupling between rotation, spin, and the chromomagnetic background leads to a cusp in the Polyakov loop potential, suggesting that the underlying dynamics could be more intricate.

We investigate the electromagnetic polarizabilities of singly heavy mesons and doubly heavy baryons within the framework of heavy hadron chiral perturbation theory up to $\mathcal{O}(p^3)$. We estimate the low-energy constants using the non-relativistic constituent quark model. A striking prediction of our study is the giant electric polarizabilities of the $D^*$ mesons: $α_E(\bar{D}^{*0}) \approx 291.4 \times 10^{-4} \text{fm}^3$ and $α_E(D^{*-}) \approx -0.4-64.4 i \times 10^{-4} \text{fm}^3$. These anomalously large values arise from the near-degenerate mass between $D^*$ and $D π$, which are orders of magnitude larger than those of their bottom counterparts. This kinematic coincidence induces a pronounced cusp structure in the chiral loops, reflecting the long-range dynamics of a pion cloud. For doubly heavy baryons, polarizabilities depend strongly on heavy-flavor composition: the $bcq$ system differs markedly from $ccq$ and $bbq$ due to mixing with scalar heavy-diquark states. Using heavy diquark-antiquark symmetry (HDAS), we unify the chiral dynamics of singly heavy mesons and doubly heavy baryons in the heavy-quark limit. The pion-loop contributions dominate the electromagnetic structure of heavy hadrons and provide essential benchmarks for future lattice QCD simulations.

Ending the inflation by vacuum decay is considered infeasible due to the graceful exit problem. Even if considering an alternative field other than the inflaton to realize a first-order phase transition (FoPT) during inflation, it is usually challenging for concrete model building, as bubble nucleations might not be fast and dense enough to successfully end the inflation. In this work, we propose a FoPT at the grand-unification-theory (GUT) scale within the Starobinsky inflation. The key construction is an exponentially evolving potential barrier dynamically controlled by the rolling inflaton, so that almost no bubble is nucleated during the early inflationary era, but with massive bubble nucleations near the end of inflation. With lattice numerical simulations, we have successfully tested this GUT-FoPT during Starobinsky inflation, and the resulting gravitational-wave energy density spectrum reproduces previous analytical estimation with a distinctive oscillation feature at high frequencies.

In this work, we investigate the drag and diffusion coefficients of various hadrons propagating through a hadronic thermal bath by employing the Fokker Planck equation within the framework of Tsallis nonextensive statistics. The nonextensive parameter $q$ accounts for the deviation from equilibrium and provides a more realistic description of the medium that is not perfectly thermalized. The hadronic bath, consisting of various mesonic and baryonic species, is characterized by different mass cutoffs that control the spectral composition of the medium. Our analysis shows that both the drag $F$ and momentum diffusion coefficients $Γ$ increases exponentially with temperature and increases systematically with increasing $q$ and mass cutoff. The spatial diffusion coefficient $D_x$ exhibits a decreasing trend with temperature $T$, $q$ and mass cutoff which highlights the significant influence of nonequilibrium effects and hadronic composition on the transport behaviour of hadrons, offering valuable insights into the thermal and dynamical properties of the hadronic phase preceding freezeout in heavy ion collisions. Additionally, we have studied the relaxation time of heavy mesons such as $D_0$, $J/ψ$ and $Υ$. We found that the heavier mesons relaxed later in comparison to the lighter mesons in the hadronic medium.

The Standard Model Effective Field Theory (SMEFT) provides a powerful, model-independent framework to explore deviations from the Standard Model (SM) by parametrising potential new physics through higher-dimensional operators. This thesis investigates the renormalisation structure of SMEFT, focusing on dimension-eight operators, which are increasingly relevant in precision analyses and in models where dimension-six effects are suppressed. We review renormalisation in quantum field theory, emphasising dimensional regularisation and the $\overline{\text{MS}}$ scheme, and outline the conceptual foundations of EFTs. One of the central results of this work is the systematic construction and classification of bosonic operators in SMEFT at dimension eight, employing Group Theory techniques and removing redundancies by working in momentum space. Building on this operator basis, we compute the complete one-loop renormalisation group equations (RGEs) involving insertions of dimension-eight-or-lower operators. This includes pure dimension-eight effects, pairs of dimension-six operators and lepton-number-violating sectors. Our calculations use an off-shell Green's function basis and leverage algebraic simplifications derived from symmetry and gauge invariance. These results are applied to positivity bounds and oblique parameters, providing essential tools for consistent SMEFT analyses across energy scales. The findings extend SMEFT's theoretical reach and support its use in high-precision phenomenology.

Hydrogen electrocatalysis (HER/HOR) exhibits an anomalous, non-Nernstian pH dependence that has motivated competing mechanistic narratives yet still lacks a unified, transferable, and quantitatively predictive description across conditions. Here, we introduce a theory-neutral waveguide kinetics framework that reinterprets the polarization curve as a power-flow-like response, enabling a compact modal representation of interfacial kinetics. Without presuming any specific mechanism, the model quantitatively fits four representative polarization datasets historically explained by divergent theories. From each fit, we extract interpretable diagnostics, including a reflection amplitude and a useful-output density, that provide transferable metrics for interfacial efficiency. The framework thus establishes a computational experiment platform for mechanistic triangulation, operating-regime diagnosis, and the rational design of hydrogen electrocatalysts across the pH spectrum.

We propose a method for computing universal (in Vogel's sense) quantum dimension formulae for universal multiplets whose associated $sl$, $so$, and $sp$ representations are nonzero. The method uses the relation between $sl$ and $so$ representations given by the vertical-sum operation, and the dual relation between $sl$ and $sp$ representations given by the horizontal-sum operation on the corresponding Young diagrams. The usual quantum dimensions of these three representations, together with subtleties related to the invariance of universal formulae under automorphisms of the $sl$ Dynkin diagram, allow one to determine the $γ$-independent factors of a universal quantum dimension (note that $γ$ is the only parameter for classical algebras, depending on their rank). Using this approach, we compute the $γ$-independent factors for (known) adjoints' universal quantum dimension, and also obtain such a factor in one new case. We discuss how to extend this approach to the $γ$-dependent factors in the quantum dimension formulae, and other issues. This is another instance in which calculations purely within the classical algebras predict the answers for the exceptional cases, due to the hidden universality structure of the theory of simple Lie algebras.

We report a follow-up axion haloscope search near 1.036 GHz that completes and extends our previous work [Phys. Rev. X 14, 031023 (2024)], in which a portion of the HEMT-based data could not be analyzed due to unrecorded experimental information. While recovering this dataset, we identified an excess near 1.036 GHz that satisfied our candidate-selection criteria, motivating dedicated validation studies, including independent cross-checks and re-examination with the original apparatus. The excess did not persist under these investigations and was not confirmed as an axion dark-matter signal. We subsequently extended the search over a 20-MHz band surrounding the candidate using a quantum-noise-limited amplifier, achieving sensitivity close to the Dine-Fischler-Srednicki-Zhitnitsky benchmark. In the absence of a confirmed signal, we set improved 90% confidence-level upper limits on the axion-photon coupling over the frequency range 1.026-1.045 GHz. This work highlights the importance of robust candidate-validation strategies as haloscope searches approach discovery-level sensitivity.

We present a systematic study of $B_c$ meson photoproduction at electron--proton colliders within the framework of nonrelativistic QCD (NRQCD) factorization. In addition to the dominant direct channel $γ+g\to B_c+X$, we include resolved contributions initiated by $g+g$ and $q+\bar q (q=u,d,s)$ subprocesses. Total cross sections and transverse-momentum distributions are calculated for several collider configurations, including HERA, LHeC, FCC-$ep$, and EIC. The numerical results show that the $γ+g$ channel provides the leading contribution to the total cross section, while the resolved $g+g$ channel yields a non-negligible correction at the level of ${\cal O}(10\%)$ in the low-$p_T$ region where most events are produced. The $q+\bar q$ channel is found to be numerically insignificant.

The subject matter of these lecture notes is the open bosonic critical string, its perturbative spectrum and interactions. We begin with a brief review of classical string propagation, quantization, as well as the level-by-level construction of physical string states. We then review a new, covariant and efficient technology of constructing entire trajectories of infinitely many physical states deeper in the string spectrum. Finally, elements of the calculation of string scattering amplitudes, including aspects of the application of the technology for the efficient calculation of tree-level amplitudes of deeper trajectories, are also covered. The material is based on three invited lectures delivered by the author at the 2024 Modave Summer School in Mathematical Physics.

The LiF Experiment for keV-scale Sterile Neutrino Search (LiFE-SNS) aims to probe sterile neutrinos through precision measurements of the tritium $β$ spectrum. Tritium nuclei are produced and embedded in LiF crystals via the ${}^{6}\mathrm{Li}(n,α){}^{3}\mathrm{H}$ reaction, allowing thermal calorimetric detection of $β$ decays with magnetic microcalorimeters (MMCs) operated at millikelvin temperatures. We present the detector configuration, background studies, and calibration method, including modeling of position-dependent response and characterization of detector nonlinearity. We also discuss potential sources of systematic uncertainty relevant to the sterile-neutrino search. While the first phase of LiFE-SNS has been completed, this paper focuses on calibration and detector characterization. The achieved performance enables precision $β$-spectrum measurements, and projected sensitivities indicate competitive reach in the keV mass region.

The Higgs-strahlung process $\ell^+\ell^-\to Z h$ is one of the most important production channels of the standard model Higgs boson $h$ at the lepton colliders. The cross section reaches the maximum value slightly above the threshold $\sqrt{s}\sim m_Z+m_h$, and decreases as $\sim 1/s$ at high energies. In the gauged extension models, the new gauge boson $Z'$ and heavy Higgs boson $H$ exist after the symmetry breaking. The heavy Higgs-strahlung process $\ell^+\ell^-\to Z' H$ would also reach the maximum cross section around the threshold $\sqrt{s}\sim m_{Z'}+m_H$. Therefore, the future high energy lepton colliders, such as the TeV scale muon collider, are promising to probe this new process. If heavy neutral lepton $N$ is introduced to generate the tiny neutrino masses via seesaw mechanism, novel signatures could arise from $μ^+μ^-\to Z' H \to NN +NN \to 4 μ^\pm +4J$ and $μ^+μ^-\to Z' H \to μ^+μ^- +NN \to 3 μ^\pm μ^\mp +2J$, where the fat-jets $J$ come from the hadronic decay of $W$ bosons. In this paper, we investigate the same-sign tetralepton signature $4μ^\pm+4J$ and the same-sign trilepton signature $3μ^\pm μ^\mp + 2J$ at the 3 TeV and 10 TeV muon collider.

Resonant Spin-Flavor Precession (RSFP) of solar neutrinos is studied using the quantum density matrix formalism, explicitly taking into account collisional decoherence and solar matter density profiles. The transition probabilities for standard $^8$B solar neutrinos ($E \approx 10$ MeV) and ultra-high-energy flare neutrinos ($E \gtrsim 1$ GeV) under three magnetic field hypotheses: core-concentrated (Wood-Saxon), tachocline-confined (Gaussian), and turbulent convective (Power Law) are compared. For standard LMA parameters, we show the resonance for 10 MeV neutrinos is strictly confined to the deep solar core ($r < 0.2 R_\odot$), rendering standard solar neutrinos insensitive to outer magnetic fields. Conversely, for 1 GeV flare neutrinos, the resonance shifts to the tachocline and convective zones, where strong fields ($B \sim 50$ kG) drive efficient spin conversion. We apply this effect to compute the difference between Dirac or Majorana neutrino scattering cross section as electron-neutrino scattering and Coherent Elastic Neutrino-Nucleus Scattering (CE$ν$NS). We show that significant asymmetry in these cross section are possible allowing in case of detection to distinguish between Dirac or Majorana neutrinos. In case of null observation, we show that this method can potentially improved the limit on the neutrino magnetic moment by one order to magnitude compared to current limits.

Bayesian analyses of the convergence pattern of Effective Field Theories (EFTs) enable estimation of the uncertainty induced by a truncated expansion. When an EFT that has been calibrated to data is used to make a prediction this truncation uncertainty enters the posterior predictive distribution twice: directly from the finite-order calculation of the predicted quantity and indirectly through the posterior probability distributions of the EFT low-energy constants (LECs) determined by the calibration. In this work, we focus on the interplay of these two sources of uncertainty. We do this in the context of a toy EFT that we fit to pseudodata and use to make predictions. Direct EFT truncation uncertainty and LEC uncertainty are correlated in predictions when the predicted quantity is correlated with the observables used to fit the LECs. Here this results in the overall theoretical uncertainty in the EFT prediction being smaller than either the uncertainty induced by the truncation error or that stemming from the LECs alone.

The Resonant Spin-Flavor Precession (RSFP) of core-collapse supernova neutrinos within the framework of the quantum density matrix formalism is studied. The cooling duration of SN1987A severely constraints standard RSFP models for Dirac Neutrinos. Using the properties of the outer stellar envelope where resonant conversion could occur after thermal decoupling, we show that for neutrino magnetic moments in the range $μ_ν\sim 10^{-14} - 10^{-12} μ_B$, adiabatic conversion in the envelope ($R > 1000$ km) leads to macroscopic helicity inversion without violating cooling bounds.This RSFP neutrino helicity change induces different signatures for Dirac or Majorana neutrinos in Coherent Elastic Neutrino-Nucleus Scattering (CE$ν$NS) detectors. For Dirac neutrino, a massive flux deficit for Dirac neutrinos due to sterile conversion should be seen. For Majorana neutrinos, the flux will not change but a modification to its spectral decomposition of the cross section should be seen due to the transition from left-handed electron neutrinos to right-handed $μ$ or $τ$ anti-neutrinos. An experimental strategy is proposed to minimize the astrophysical uncertainties using the high-energy neutrino tail ($E \approx 1$ GeV) which evades RSFP to normalize the signal. This ratio-based approach effectively cancels astrophysical uncertainties, allowing future detectors to distinguish the fundamental nature of the neutrino and probe magnetic moments down to $10^{-14} μ_B$, two orders of magnitude beyond current solar limits.

We study integrated correlation functions of half-BPS operators in $SU(N)$ $\mathcal{N} = 4$ supersymmetric Yang-Mills theory (SYM) involving two superconformal primary operators in the stress-tensor multiplet and two identical maximal-trace operators of arbitrary $R$-charge $p$. Thanks to $\mathcal{N}=4$ SYM electro-magnetic duality these integrated correlators have recently been computed as exact functions of $N$, $p$, and of the Yang-Mills complexified coupling $τ$. Using a combination of tools from ${\rm SL}(2,\mathbb{Z})$ spectral theory and resurgence analysis, we study the landscape of large-$N$ and/or large-charge expansions for these correlators. In particular, we find novel non-perturbative effects in the limit where $N\rightarrow \infty$ with $p/N^2$ fixed. From a holographic point of view this double-scaling regime is deeply connected with a second family of correlators which we analyse. Using the results for the maximal-trace operators, we derive an exact expression for a new integrated correlator involving two coherent-state operators, defined via an exponential generating function of multi-graviton states. At large-$N$ this correlator admits a holographic dual description in terms of a back-reacted geometry known as the AdS bubble. First, we show that the leading supergravity contribution to the integrated correlator agrees with a direct explicit integration of the correlator itself. Secondly, we derive predictions for the integrated version of the Virasoro-Shapiro amplitude evaluated on the AdS bubble background. Lastly, we demonstrate that the large-$N$ non-perturbative contributions to this integrated correlator emerge from giant-magnon configurations in the dual AdS bubble.

Four-dimensional $\mathcal{N}=2$ supergravity theories become rigid in gravity-decoupling limits. We study this effect for type II string compactifications on general Calabi--Yau manifolds, focusing on vector-multiplet trajectories whose endpoints exhibit axionic shift symmetries. This comprises field excursions of both finite- and infinite distance, but the latter display specific features due to the appearance of light towers of extremal BPS states, in agreement with Swampland principles. We show that vector multiplets split into two sets: those with gravitational and with rigid mutual interactions, and that only a subset of the latter -- dubbed core RFT -- can fully decouple from gravity. We characterise the core RFT in terms of the axionic shift symmetry, and derive decoupling criteria based on kinetic and Pauli interaction mixing. Our framework is illustrated in large complex structure, conifold-like, and Seiberg--Witten limits. In the last case, Pauli mixing obstructs decoupling whenever the dyonic and extremal BPS towers appear at the same scale. Across all these examples, the decoupling from gravity is signalled by a divergent moduli-space scalar curvature.

We conjecture that decoupling relations in the operator product expansion of a 4d $\mathcal{N}=2$ superconformal field theory (SCFT) are encoded by an algebro-geometric object: a bifiltered affine scheme. We demonstrate how this scheme reproduces the Macdonald index (thus the Schur index) as well as the Higgs branch. Although the associated scheme typically admits continuous deformations, we find that a geometric extremization principle uniquely fixes these moduli, thereby providing a possible geometric route toward a classification of 4d $\mathcal{N}=2$ SCFTs.

We compute Mellin moments of the heavy-quark structure functions in deep-inelastic scattering at next-to-leading order in quantum chromodynamics, retaining their full dependence on the heavy-quark mass. Using the optical theorem and the operator product expansion, we derive analytic results for fixed Mellin moments $N = 2$ to $22$ of the structure functions $F_2$ and $F_L$. Our results reproduce the known expressions in the relevant asymptotic limits, in particular for virtualities of the exchanged photon $Q^2$ much larger than the heavy-quark mass squared $m^2$, and are in agreement with existing parametrisations of the next-to-leading-order coefficient functions. The computational set-up developed in this work also provides a direct pathway toward extending these calculations to next-to-next-to-leading order.

The realization that primordial black holes (PBHs) might be some fraction of the dark matter begged the question, how often do PBHs enter the solar system? For a Neptune radius solar system the answer is, rarely. For an Oort cloud sized system the answer is different. Simulations of bombardment of the Oort cloud by dark matter suggest that dislodgement of protocomets and their entry into the inner solar system can match the observed frequency of comets, if that PBH fraction is high enough. Comets were traditionally considered as messengers, usually omens. After 50 years of puzzlement regarding dark matter, we need a hint from the dark universe about the size and nature of dark matter particles.

The impact of the most recent cosmic microwave background (CMB) data from the Atacama Cosmology Telescope (ACT) is studied for a model of cosmic inflation which predicts a non-zero vacuum expectation value (VEV) $M$ for a large-field regime. Since lower values of $M$ are compatible with the higher spectral index $n_s$ provided by the ACT+Planck joint analysis, we establish new limits on this parameter while also considering further CMB data from the latest BICEP/Keck Array release for CMB polarization modes. We find $\log_{10}M/M_{Pl}=-2.5^{+1.1}_{-1.3}$ at 68\% confidence level, compatible with $M/M_{Pl}\simeq 0.003$, which is interesting for post-inflationary processes, such as preheating. We conduct a lattice simulation for the inflaton field for the first few e-folds, as the model is compatible with the production of relics such as oscillons, which are possible candidates as sources of gravitational waves and primordial black holes. We find that the model indeed produces localized, quasi-spherical structures compatible with oscillons, which might lead to signatures detectable by future experiments. However, in agreement with recent works, we find that although the abundance of gravitational waves that could be generated in this regime has an amplitude well within the sensitivities of these detectors, the frequency range is on the GHz limit, away from the expected frequencies. Finally, we estimate the impact of a coupling of the type $yφσ^2$ to the inflaton, in the realization of perturbative reheating, directly impacting the predictions of the model, as lower values of $M$ are consistent with both the entire allowed temperature range, and the limits imposed by BICEP/Keck Array+Planck+ACT.

Little red dots (LRDs) are an abundant population of compact high-redshift sources with red rest-frame optical continua, discovered by the James Webb Space Telescope (JWST). Their red colors and power sources have been attributed either to dust reddening of standard hot accretion disks or to intrinsically cool thermal emission from dense hydrogen envelopes, in both cases surrounding accreting supermassive black holes. These scenarios predict order-of-magnitude differences in emission temperature but have lacked decisive temperature diagnostics. Here we report a prominent absorption feature at rest-frame $\sim 1.4 \, μ\mathrm{m}$ in two out of four LRDs at $z \sim 2$ with high signal-to-noise JWST spectra, among the coolest from a large LRD sample. The feature matches the shape and wavelength of the water absorption band seen in cool stars. Atmosphere models require $T \lesssim 3000\, \mathrm{K}$ to reproduce it, confirming unambiguously the presence of a cool, dense gas component contributing $20-30\%$ to the emergent continuum. A composite model reproduces both the absorption and the rest-frame optical-to-infrared continuum shape and suggests a temperature range ($\sim2000\, \mathrm{K} - 4000 \, \mathrm{K}$) rather than a single blackbody predicted by some gas envelope models. Molecular absorption demonstrates that the red continua of some LRDs are intrinsic rather than dust-reddened, implying order-of-magnitude lower bolometric luminosities and black-hole masses, and providing a new diagnostic of the emitting gas.

We test claims of large-scale anisotropy in the local expansion rate using cluster scaling relations as distance indicators. Using a Bayesian forward model, we jointly fit the X-ray luminosity--temperature (LT) and thermal Sunyaev-Zel'dovich--temperature (YT) relations, marginalising over the latent cluster distances and modelling selection effects as well as peculiar velocities. The latter are modelled using reconstructions of the local peculiar velocity field where we self-consistently account for possible anisotropic redshift--distance relations via an approximate scheme. This treatment proves crucial to the inferred anisotropy and breaks the degeneracy between anisotropy in scaling relation normalisations and underlying cosmological anisotropy. We apply our method to 312 clusters at $z \lesssim 0.2$, testing dipolar, quadrupolar and general (pixelised) anisotropy models. Bayesian model selection finds no more than weak evidence for any anisotropic model. For dipole models, we obtain upper limits of $δH_0 / H_0 < 3.2\%$ and bulk flow magnitude $< 1300\,\mathrm{km\,s^{-1}}$. Our results contrast with previous claims of statistically significant anisotropy from the same data, which we attribute to our principled forward modelling of both redshifts and scaling relation observables through latent distances and our treatment of the impact of anisotropic redshift--distance relations when modelling the local peculiar velocity field. Our work highlights the importance of accurately modelling peculiar velocities when testing isotropy with distance indicators, and motivates the further development of reconstructions that self-consistently treat large-scale deviations from the Hubble flow.

Studying the cross-correlation function between the soft and hard X-ray emission in Neutron Star Low Mass X-ray Binaries provides crucial insight into the structure and dynamics of the innermost accretion regions. In this work, we investigate the CCF of the Z-source GX 349+2 using an XMM-Newton observation. We noted that asymmetric CCFs with lags of a few hundred secondsbetween soft and hard band LCs in the horizontal branch, whereas CCFs remained symmetric in normal and flaring branches. We also performed a CCF study during the flux transition duration and observed lags of the order of a few tens to hundreds of seconds. Monte Carlo simulations were performed to assess the robustness of these CCFs, confirming their significance at a 95% confidence level. Spectral analysis during the flux transitions further suggests that the inner accretion disk extends close to the last stable orbit. We propose that the observed hard lags arise from the readjustment of the boundary layer/coronal region located near the inner edge of the accretion disk. From the measured lags, we estimate the characteristic size of the boundary layer. We show that the observed lags could also be associated with the depletion timescale of the boundary layer with low viscosity.

Understanding dark matter's elusive nature is crucial for the framework of particle physics and expanding the Standard Model. This analysis utilizes the High Altitude Water Cherenkov (HAWC) gamma ray Observatory to indirectly search for dark matter (DM) by studying gamma ray emission from dwarf spheroidal galaxies (dSphs). Selected for their high ratio of dark matter to baryonic matter, dSphs are useful for this type of search owing to the low background emission. In comparison to previous HAWC studies, we significantly improve our sensitivity to DM from dSphs due to improvements to our event reconstruction and reduced hadronic contamination. We expanded the number of dSphs studied, DM annihilation channels into the Standard Model (SM), and the amount of data collected on each previously studied dSph. We searched for DM signals in each dSph using the latest version of the algorithms used to reconstruct data from the primary detector of the HAWC instrument. We report that we do not detect evidence of DM from dSphs, so we place upper limits for the velocity-weighted DM annihilation cross-section ($\langleσv \rangle$) on the order of $10^{-23}~\text{cm}^3\text{s}^{-1}$ for a DM mass range of $1-10^4$ TeV.

Winds launched at the scale of the accretion disc or dusty torus in Active Galactic Nuclei (AGN) are thought to drive energy-conserving outflows that shape galaxy evolution. The key signature of such outflows, the presence of a hot ($T \gtrsim 10^9 \, \rm K$), shocked wind component, is hard to detect directly. Observations of AGN outflows typically probe a separate outflow phase: cool/warm gas with $T \lesssim 10^5 \, \rm K$. Here, we show that the density of cool outflowing gas scales with AGN luminosity, serving as an indirect diagnostic of the elusive hot, shocked wind. We use hydrodynamic simulations with the moving-mesh code AREPO to target the interaction between a small-scale AGN wind of speed $\approx 10^4 \, \rm km \, s^{-1}$ and galactic discs containing an idealised, clumpy interstellar medium (ISM). Through a new refinement scheme targeting rapidly-cooling, fast-moving gas, our simulations reach a resolution of $\lesssim 0.1 \, \rm pc$ in the cool, outflowing phase. We extract an ensemble of cool clouds from the AGN-driven outflows produced in our simulations, finding that their densities increase systematically with AGN wind power and AGN luminosity. Moreover, the mass distribution and internal properties of these cloudlets appear to be insensitive to the initial properties of the ISM, and shaped mainly by the dynamics of radiative, turbulent mixing layers. The increase in cool outflow density with kinetic wind power and AGN luminosity has profound implications for observational estimates of outflow rates and their scaling with AGN luminosity. Depending on the available outflow and density tracers, observationally-derived outflow rates may be overestimated by orders of magnitude.

We conducted a timing analysis of over 18 years of data on the young ($τ_{\rm c} = 112$ kyr, $P = 114$\,ms) relativistic binary PSR~J1906+0746, using six radio telescopes: Arecibo, FAST, Green Bank, Lovell, MeerKAT, and Nançay. This pulsar is known to orbit a compact high-mass companion with a period of 3.98\,hrs in a mildly eccentric orbit ($e = 0.085$). By combining all data and maintaining a coherent timing solution over the full span, we obtained a more precise measurement of the advance of periastron, $\dotω = 7.5841(2)$\,$°$\,yr$^{-1}$, the Einstein delay, $γ= 4.59(2) \times 10^{-4}$\,s, and the secular change in orbital period, $\dot{P}_{\rm b} = -5.65(2) \times 10^{-13}$\,s\,s$^{-1}$. Assuming the validity of general relativity, we obtain a total mass of 2.6133(1)\,$M_{\odot}$ and component masses of 1.316(5) $M_{\odot}$ for the pulsar and 1.297(5) $M_{\odot}$ for the companion, consistent with a double neutron star system. However, when fitting for the secular change in the projected semi-major axis we obtain $\dot{x} = -1.8(6) \times 10^{-13}$\,s\,s$^{-1}$, the component masses are shifted by $\sim 3.5σ$, which is expected from the correlation of $\dot{x}$ and $γ$. The $\dot{x}$ has a similar magnitude to that observed in PSR~J1141$-$6545, which is due to spin-orbit coupling; if confirmed, it would indicate that, as in the latter system, the companion of the PSR~J1906+0746 system is a massive fast-rotating white dwarf formed before the pulsar. Additionally, we report and characterize a large glitch near MJD 56664, with a fractional frequency increase comparable to those observed in the Vela pulsar.

We present deep, medium-resolution $λ=1-5\,μ$m JWST/NIRSpec spectroscopy for 14 quiescent galaxies at $3<z<5$ with $\log_{10}(M_*/\mathrm{M_\odot}){\,>\,}10$, obtained as part of the EXCELS survey. We perform a complete re-reduction of these data, including a custom optimal-extraction approach to combat the spectral "wiggles" that result from undersampling of the NIRSpec spatial PSF. We constrain the star-formation histories and stellar metallicities of these objects via full-spectral fitting, finding a clear stellar age vs stellar mass correlation, in which more massive galaxies assembled their stellar mass at earlier times. This confirms spectroscopically that the archaeological "downsizing" trend was already in place by $z\simeq4$. The slope of our measured relation ($\simeq1.5$ Gyr per dex in stellar mass) is consistent with literature results at $0 < z < 3$. We do not observe objects with $\log_{10}(M_*/\mathrm{M_\odot})\lesssim10.5$ and ages of more than a few hundred Myr at this epoch, suggesting that recently reported examples of higher-redshift quiescent galaxies at these masses are likely to soon rejuvenate. We measure relatively high stellar metallicities for the majority of our sample, consistent with similar objects at $0 < z < 3$. Finally, we explore evidence for $α$-enhancement in six older and more luminous galaxies within our sample, finding considerable disagreements in the chemical abundances measured using different stellar population models, different fitted rest-frame wavelength ranges, star-formation history models and fitting codes. We therefore conclude that inferring detailed stellar chemical abundances for the earliest quiescent galaxies remains challenging, and higher signal-to-noise spectra are required (SNR per resolution element $>100$ for $R\simeq1000$).

Understanding how galaxies trace the underlying matter density field is essential for characterizing the influence of the large-scale structure on galaxy formation, being therefore a key ingredient in observational cosmology. This connection, commonly described through the galaxy bias, $b$, can be studied effectively using machine learning (ML) techniques, which offer strong predictive capabilities and can capture non-linear relationships. We aim to incorporate the linear bias parameter assigned to individual galaxies into a ML framework, quantify its dependence on various halo and environmental properties, and evaluate whether different algorithms can accurately predict this parameter and reproduce the scatter in several bias relations. We use data from the IllustrisTNG300 simulation, including the distance to different cosmic-web structures computed with DisPerSE. These data are complemented with an object-by-object estimator of the large-scale linear bias ($b_i$), providing the individual contribution of each galaxy to the bias of the entire population. Our ML framework uses three models to predict $b_i$: a Random Forest Regressor, a Neural Network and a probabilistic method (Normalizing Flows). We recover the full hierarchy of galaxy bias dependencies, showing that the most informative features are the overdensities, particularly $δ_8$, followed by the distances to cosmic-web structures and selected internal halo properties, most notably $z_{1/2}$. We also demonstrate that Normalizing Flows clearly outperform deterministic methods in predicting galaxy bias, including its joint distributions with galaxy properties, owing to their ability to capture the intrinsic variance associated with the stochastic nature of the matter-halo-galaxy connection. Our ML framework provides a foundation for future efforts to measure individual bias with upcoming spectroscopic surveys.

We present updated microlensing analysis results based on high-cadence ($\sim$2~min) Subaru Hyper Suprime-Cam (HSC) observations of the Andromeda Galaxy (M31) in 2014, 2017, and 2020, yielding a total of 39.3 hours of data. We use a point-lens finite-source model for the microlensing light curve model and employ multi-stage selection procedures to identify microlensing candidates. From more than 25,000 variable candidates detected across all nights, we identify 12 microlensing candidates with light-curve timescales shorter than 5~hours, and among them, 4 secure candidates with high-significance detections. We estimate detection efficiencies using light-curve-level simulations that account for observational conditions and finite-source effects. Using a hierarchical Bayesian framework that combines the light-curve fitting information for each candidate with the Poisson statistics of the number of candidates, we derive constraints on parameters that characterize the abundance and mass scale of primordial black hole (PBH) dark matter. First, we derive upper limits on the PBH abundance under the null hypothesis that all events are assumed to be false detections. Next, employing the PBH hypothesis in which all (or only secure) candidates are assumed to be due to PBH microlensing, we derive the allowed region of the PBH parameters; the inferred mass scale is $M_{\rm PBH}\sim10^{-7}$--$10^{-6}M_\odot$, and the PBH abundance to the total dark matter is $f_{\rm PBH}\sim \mathcal{O}(10^{-2}{\rm -}10^{-1})$. Our results demonstrate that HSC-M31 monitoring remains a uniquely powerful probe of PBHs, and highlight the need for further studies for example, using Rubin Observatory LSST observations of the Large Magellanic Cloud.

Finite nuclei constrain the dense-matter equation of state (EOS), yet they are self-bound quantum droplets far from the thermodynamic limit. Motivated by an analogy to quantum dots, we show that the nuclear chart nevertheless defines a mesoscopic regime in which mesoscopic chemical-potential analogs $\{μ_B,μ_Q,μ_S\}$ can be extracted directly from nuclear and hypernuclear binding energies after consistent Coulomb subtraction. These are discrete finite-difference response functions -- local slopes of the strong-interaction energy landscape -- not equilibrium grand-canonical chemical potentials. The nuclear chart itself supplies an "ensemble of nearby droplets": finite differences across neighboring nuclei suppress shell- and pairing-scale oscillations while retaining the smooth bulk trend, producing robust slopes without a macroscopic limit. Thus, the data provide empirical local derivatives that any strangeness-enabled EOS must reproduce near saturation. Mapping the measured (hyper)nuclear landscape at $T\simeq 0$, we find smooth, numerically stable responses, including a large, negative strangeness chemical-potential analog, and we identify specific hypernuclear measurements that can directly test and sharpen these EOS constraints.

Radio observations offer a dust-independent probe of star formation and active galactic nucleus (AGN) activity, but sufficiently deep data are required to access the crossover luminosity between these processes at high redshift ($z>4.5$). We present three spectroscopically confirmed high-redshift radio sources (HzRSs) detected at 1.3 GHz at $z=4.9$-$5.6$, with radio luminosities spanning $L_{\rm 1.3 \, GHz}\approx2$-$5\times10^{24} \, \rm W \, Hz^{-1}$. These sources were first identified as high-redshift candidates through spectral energy distribution (SED) fitting of archival Hubble, JWST NIRCam+MIRI, and ground-based photometry, and then spectroscopically confirmed via the $\rm H\,α$ emission line using wide-field slitless spectroscopy from JWST COSMOS-3D. The star formation rates (SFRs) measured from SED fitting, the $\rm H\,α$ flux, and the 1.3 GHz luminosity, span $\sim100$-$1800\, M_{\odot} \, \rm yr^{-1}$, demonstrating broad agreement between these SFR tracers. We find that these three sources lie either on or $0.5$-1.0 dex above the star-forming main sequence at $z=4$-6 and have undergone a recent burst of star formation. The sources have extended rest-UV/optical morphologies with no evidence for a dominant point source component, indicating that an AGN is unlikely to dominate their rest-UV and optical emission. Two of the sources have complex, multi-component rest-frame UV/optical morphologies, suggesting that their starbursts may be triggered by merging activity. These HzRSs open up a new window towards probing radio emission powered by star formation alone at $z> 4.5$, representing a remarkable opportunity to begin tracing star formation, independent of dust, in the early Universe.

The cosmic dipole anomaly, currently detected at a significance exceeding 5$σ$ in several independent survey poses a significant challenge to the standard model of cosmology. The Ellis & Baldwin formula provides a theoretical link between the intrinsic dipole anisotropy in the sky distribution of extragalactic light sources and the observer's velocity relative to the cosmic rest frame, under the assumptions that the sources follow a power-law luminosity function and exhibit power-law spectral energy distributions. In this work, we demonstrate that this relation can be generalized to arbitrary luminosity distributions and spectral profiles. We derive the corresponding expression for the effective spectral index and apply it to a sample of quasars observed in the W1 band of the CatWISE survey. We show that the anomalous cosmic dipole persists beyond the power-law assumption. These results provide a more general and robust framework to interpret measurements of the cosmic dipole in future large-scale surveys.

Galaxy groups are affected by outflows from central Active Galactic Nuclei due to the shallower gravitational potential compared to galaxy clusters. The group binding energy is comparable to the energy output from AGN, making it an important factor in mutual evolution. We present a multi-wavelength analysis of three dynamically active groups: SDSSTG8102, SDSSTG16393, and SDSSTG28674, which are part of the XMM-Newton Group AGN Project X-GAP sample, a statistically complete sample of 49 galaxy groups. We combine uGMRT 400 MHz observations with 144 MHz LOFAR and XMM-Newton observations to study the radio sources associated with the Brightest Group Galaxies (BGGs). The BGGs in SDSSTG8102 and SDSSTG16393 have extended radio emission with asymmetric distortions in their morphologies. SDSSTG28674 has a compact flat-spectrum radio source associated with the BGG and an extended lobe on one side, connected by a faint bridge detected with LOFAR. Integrated spectral indices of the three BGGs are $-0.96\pm0.09$ (SDSSTG8102), $-1.35\pm0.09$ (SDSSTG16393), and $-1.6\pm0.02$ (SDSSTG28674). X-ray images reveal elongated morphologies in all three groups, with SDSSTG28674 showing evidence of a binary merger, while thermodynamical maps highlight temperature variations. In SDSSTG8102, lobes are bent and displaced by IGrM flows, while SDSSTG16393 hosts steep-spectrum relic-like plasma coinciding with X-ray emission. SDSSTG28674, with its ultra-steep spectrum lobe and disturbed morphology, likely traces merger-driven activity, consistent with a remnant or revived radio phoenix. The spectral diversity across the systems reflects different stages of AGN fading governed by duty cycle, source age, and confinement by the hot IGrM.

The study of multiwavelength flux and spectral variations during rapid flares from blazars provides strong constraints on the physical parameters of the compact emission regions responsible for these still poorly understood events. Although a full description of the continuous and transient emission from blazars seems to require more sophisticated scenarios, particle acceleration and loss mechanisms can be approximately described within the simple leptonic one-zone framework, enabling a systematic study of their impact on the observable properties of multiwavelength flare light curves. Our goal is to identify characteristic signatures in these light curve profiles that permit one to discriminate between the main physical processes situated inside the relativistic jet and commonly invoked to explain blazar flares. The present study exclusively focuses on modeling rapid flares from BL Lac type objects, which can be described within the synchrotron self-Compton (SSC) emission scenario. Combinations of several commonly employed mechanisms to describe the gain and loss of energetic particles in onezone models during flaring events are studied in a systematic way: particle injection; diffusive shock and stochastic acceleration and reacceleration; particle escape; adiabatic losses; radiative losses through synchrotron and inverse-Compton radiation. The current study is limited to the case of "hard-sphere" scattering. A large variety of light curve shapes arises from the different scenarios under study. Characteristic signatures, in particular energy-dependent time delays and differences in the shapes of the rising part of the flare, should allow the distinction to be made between different injection and acceleration scenarios, given the availability of sufficiently high-quality multiwavelength data sets. This is illustrated with a simplified application to a flare event from the blazar Mrk 421.

Hyperaccreting stellar-mass black hole systems are leading candidates for the central engines of gamma-ray bursts (GRBs). Their jets are thought to be powered by either the Blandford-Znajek (BZ) process or neutrino-dominated accretion flows (NDAFs), but discriminating between these mechanisms remains challenging. To address this, we propose using the luminosity decay slope (parameter d) of GRB light curves to distinguish between the BZ and NDAF mechanisms, thereby linking the light-curve morphology to the central engine physics. By analysing 85 single-peaked GRBs with fast-rise, exponential-decay (FRED) profiles observed by Swift/BAT using 64 ms background-subtracted light curves, we fit the decay slope (parameter d) with the empirical Kocevski-Ryde-Liang (KRL) function and compare the results with theoretical predictions for the BZ (d approximately 1.67) and the NDAF (d approximately 3.7 to 7.8) mechanisms. We find that the decay slope (parameter d) can differentiate central engine mechanisms, with 15 GRBs consistent with the BZ mechanism and 22 supporting the NDAF mechanism. However, most events exhibit slopes within the range between 2 and 4, suggesting a hybrid of mechanisms, with NDAF being dominant.

Wide-orbit binaries serve as crucial laboratories for understanding stellar evolution and identifying quiescent compact objects. In this work, we search for compact objects in wide-orbit binaries by merging the LAMOST multi-epoch catalogs from LRS and MRS in the 12th data release. We specifically focus on sources with at least 20 observation epochs that clearly exhibit long-term radial velocity (RV) variations while remaining stable over short time scales. By constraining the mass function with Lomb-Scargle periods and RV ranges, we identified 76 single-lined spectroscopic binary candidates harboring potential compact objects with robust orbital solutions. These systems exhibit orbital periods ranging from 10 to 1000 days, with semi-amplitudes of velocity $K_1 \lesssim 50$ km/s and mass functions $f(M_2)$ between 0.05 and 0.6 $M_{\odot}$. Combining $f(M_2)$ with SED-derived stellar parameters, we identify 6 strong compact object candidates with main-sequence companions (Class A), 24 systems likely consisting of either compact objects with giant/subgiant companions or mass-inverted Algol-type binaries (Class B), and 46 candidates with relatively lower mass ratios (Class C). Cross-matching with the Gaia DR3 nss_two_star_orbit catalog yields 16 sources, all of which exhibit orbital solutions consistent with our results. This study demonstrates the essential role of long-term spectroscopic monitoring in searching for compact objects in wide-orbit binaries and validating orbital solutions. The strategy of leveraging extended time baselines will be increasingly effective as spectroscopic databases continue to grow, enabling the systematic discovery of compact objects in wide orbits across the Galaxy.

Recent observations with the James Webb Space Telescope (JWST) reveal that the merger fraction among Ly$α$ emitters (LAEs) at redshifts $z > 3$ is significantly higher than previously estimated. In this study, we focus on three high signal-to-noise merging LAE systems at $3 < z < 4$, selected from the VLT/MUSE-Deep survey in the GOODS-S field. We combine new \textit{JWST}/NIRCam broadband and medium-band imaging with archival \textit{HST}/ACS data to perform spatially resolved spectral energy distribution (SED) fitting using the \textsc{Bagpipes} software package. Our analysis reveals that two of the systems are minor mergers, while the third is a major merger. The close agreement between spatially resolved and integrated stellar mass estimates indicates that recent star formation does not significantly outshine the light from older stellar populations in these systems. Moreover, both the individual components and the systems as a whole lie on the star-forming main sequence, further supporting the conclusion that these mergers have not yet triggered substantial starburst activity. Furthermore, we detect prominent color gradients and disturbed dust distributions in these merging systems, indicating that the mergers have already induced significant internal structural perturbations. These morphological and dust-related changes may facilitate the escape of Ly$α$ photons -- potentially through mechanisms such as gas redistribution or a reduced covering fraction of neutral hydrogen -- thereby playing a key role in shaping the observed properties of LAEs.

We present a comprehensive temporal and spectral analysis of the long-duration gamma-ray burst GRB 110801A, utilizing multi-band data from the Neil Gehrels Swift Observatory and ground-based telescopes. The $γ$-ray emission exhibits a distinct two-episode (``double-burst'') structure. Rapid follow-up observations in the optical and X-ray bands provide full coverage of the second burst. The optical light curve begins to rise approximately 135 s after the trigger, significantly preceding the second emission episode observed in X-rays and $γ$-rays at $\sim 320$ s. This chromatic behavior suggests different physical origins for the optical and high-energy emissions. Joint broadband spectral fitting (optical to $γ$-rays) during the second episode reveals that a two-component model, consisting of a power-law plus a Band function, provides a superior fit compared to single-component models. We interpret the power-law component as the afterglow of the first burst (dominating the optical band), while the Band component is attributed to the prompt emission of the second burst (dominating the high-energy bands). A physical synchrotron model is also found to be a viable candidate to explain the high-energy emission. Regarding the afterglow, the early optical light curve displays a sharp transition from a rise of $\sim t^{2.5}$ to $\sim t^{6.5}$, which is well-explained by a scenario involving both reverse shock (RS) and forward shock (FS) components. We constrain the key physical parameters of the burst, deriving an initial Lorentz factor $Γ_0 \sim 60$, a jet half-opening angle $θ_j \sim 0.09$, and an isotropic kinetic energy $E_{\rm k,iso} \sim 10^{54.8}$ erg.

Integral field spectroscopy (IFS) provides spatially resolved spectra, enabling detailed studies that address the physical and kinematic properties of the interstellar medium. A critical step in analyzing IFS data is the decomposition of emission lines, where different velocity components are often modeled with Gaussian profiles. However, conventional fitting methods that treat each spectrum independently often yield spatial discontinuities in the fitting results. Here, we present Emission Line Fitting Optimization (ELFO), a Python package for IFS spectral fitting. ELFO uses the results of neighboring spectra to determine multiple initial guesses and selects the result that exhibits spatial smoothness. We tested ELFO on IFS data of two quasars obtained from the Multi-Unit Spectroscopic Explorer, where it successfully corrected anomalous fits, revealed previously unresolved substructures, and made large-scale kinematic structures more evident. With minor modifications, this method can also be easily adapted to other IFS data and different emission lines.

The chemical evolution in star forming regions is driven by the interplay between gas and ice mantles. Identifying the ice compositions at the early stage of star formation thus provides constraints on the chemical processes inaccessible from gas-phase characterizations. As part of the CORINOS program, spectra from the James Webb Space Telescope (JWST) MIRI MRS were taken toward four Class 0 protostars: IRAS 15398-3359, Ser-emb7, L483, and B335. The spectra were processed with simultaneous fitting of a continuum and silicate absorption to produce optical depth mid-infrared spectra of the ices at 5-28 $μ$m (360-2000 cm$^{-1}$) toward these four sources. Simple molecules such as water (H$_2$O), carbon dioxide (CO$_2$), methanol (CH$_3$OH), formic acid/formate (HCOOH/HCOO$^-$), ammonia/ammonium (NH$_3$/NH$_4$$^+$), and formaldehyde (H$_2$CO) are the most abundant features in these ices, while complex organic molecules (COMs) represent a smaller contribution. Likely COMs include hydroxylamine (NH$_2$OH), methylamine (CH$_3$NH$_2$), and ethanol (CH$_3$CH$_2$OH). Absorption features belonging to functional groups such as -CH$_3$ and -OH suggest that additional COMs are present, but these cannot be unambiguously assigned due to overlapping bands. Formation pathways toward these COMs utilizing radical-radical combination reactions based on laboratory simulation experiments is presented. By extension, COMs predicted by these reactions, but absent from the spectra, are discussed. The results provide insight into the chemical environment of these ices and also highlight the critical need for caution and sufficient evidence in order to confidently identify COMs in ice.

Active Galactic Nuclei (AGN) outflows can regulate host galaxy evolution via AGN feedback. Ionised gas outflows have been linked to enhanced radio emission. In the first paper of this series, AGN detected with the International LOFAR Telescope (ILT) at 6" were more likely to host an [O III] 5007 outflow than AGN not detected, although only high-powered jets were ruled out as the origin of radio emission. New wide-field, sub-arcsecond resolution imaging at 144 MHz with the ILT now enables a resolved morphological study of this sample. We present the first wide-field, sub-arcsecond images of the Boötes Deep Field at 144 MHz, detecting 4074 sources in the $\sim$0.3" image with a central sensitivity of 33.8 uJy $\mathrm{beam^{-1}}$. For 47 AGN matched in AGN luminosity, we probe radio emission on kiloparsec-scales to investigate correlations with [O III] outflows. This sample spans $z<0.83$, $10^{40}<L_{\mathrm{[O III]}}<10^{43}~\mathrm{erg~s^{-1}}$, and $10^{21}<L_{\mathrm{144MHz}}<10^{24.5}~\mathrm{W~Hz^{-1}}$. We find that if we detect an AGN on both large-scales (6") and small-scales (0.3"), 90$\pm$7 per cent have an [O III] outflow, compared to 63$\pm$9 per cent of sources detected on large-scales, but undetected on small-scales. Furthermore, 17$\pm$6 per cent of sources without an [O III] outflow are detected on kiloparsec-scales, compared to 51$\pm$12 per cent of sources with an [O III] outflow. This implies a connection between [O III] outflows and kiloparsec-scale radio emission, which is likely AGN-driven. In contrast, AGN without an [O III] outflow are dominated by diffuse radio emission, likely to be associated with star formation.

We investigate the relative alignment between density structures and magnetic fields in eight young protostars from the ALMA B-field Orion Protostellar Survey. Column density maps are derived from 870 $μ$m dust continuum emission, and the Histogram of Relative Orientations (HRO) method is applied to quantify the correlation between magnetic field orientations and density structures on envelope scales ($\sim$10$^{3}$~au). We find that the relative alignment shows overall weak evidence of systematic evolution with column density, suggesting that column density alone does not fully determine the alignment. The magnetization level also plays a crucial role, with weakly magnetized envelopes exhibiting predominantly parallel or random alignment, whereas strongly magnetized ones show perpendicular configurations even at moderate densities. These results reveal that density and magnetization jointly shape the morphology of protostellar envelopes and the coupling between gravity and magnetic fields during early stages of star formation.

The fractal dimension $D$ is used to map the large-scale galaxy distribution in the Universe by color types: blue, green and red. Using a $NUVrK$-complete COSMOS2020 subsample of 618,952 galaxies observed up to $z=4$, number densities were derived and plotted against two cosmological distance measures, the luminosity and comoving (galaxy area) distances, in order to estimate $D$ for each galaxy color type in two redshift intervals: $z\gtrless1$. We found a general gradient $D_{\mathrm{blue}}> D_{\mathrm{red}}>D_{\mathrm{green}}$ with $D=1.40-2.03$ for $z<1$. For $1<z\leq4$, the gradient changes to $D_{\mathrm{blue}}>D_{\mathrm{green}}D_{\mathrm{red}}$, and the fractal dimension values are lower, $D=0.03-0.44$. These results suggest that the fractal dimension is a sensitive diagnostic for how galaxy populations trace the evolving cosmic web, and confirm the fractal dimension as a useful tool for observational mapping of large-scale structure by galaxy color.

Prior to its infall, the Sagittarius (Sgr) dwarf galaxy was a major satellite with a mass of $M_{\rm sgr}\sim 10^{11}$ M$_\odot$. For the past $5-6~\mathrm{Gyr}$ , it has been heavily stripped by the Milky Way (MW), losing most of its mass while crossing the MW disc multiple times. Recent models of Milky Way disc perturbations $-$ including the spiral arms, the stellar bar, the Gaia phase spiral, and stellar and gaseous disc corrugations $-$ have identified these crossings as possible formation triggers, but have generally treated each perturbation in isolation. Here, we adopt a holistic perspective and ask whether a single disc-crossing impulse can simultaneously account for these features as observed today. We focus on simulations of single disc-crossing events by a Sgr-like perturber, and present a forensic analysis of the role of the powerful impulse in forming spiral arms, disc corrugations, the phase spiral and the `$L_{z}-\bar{V}_{R}$ wave', determined from a star's angular momentum and radial velocity, respectively. We find that a single disc crossing can reproduce reasonably well (e.g. structure, amplitude, phase) the observed local disc corrugation, and the Outer, Local and Sagittarius-Carina arm segments, implying that the last significant impulse due to a transit took place $700-1200~\mathrm{Myr}$ ago. Moreover, the $L_{z}-\bar{V}_{R}$ wave and phase spiral appear within the simulations over the same epoch and their general structure is reasonably well replicated, but not in detail. We conclude that Sgr's last significant crossing roughly a Gyr ago could be the primary cause of large-scale MW disc perturbations, but it cannot fully account for the $L_{z}-\bar{V}_{R}$ wave. Consequently, other triggers, possibly the Galactic bar or interactions with other satellites, must be considered in order to fully explain the current dynamical state of the MW's disc.

We introduce the Accretion-Modulated Internal Shock model (AMIS) as a possible framework for explaining the observational properties of long gamma-ray burst (GRB) prompt emission. In this scenario, the envelope of the prompt light curve follows the time-dependent mass-supply history to the central engine, associated with stellar collapse and, where applicable, fallback accretion, whose early-time onset can be approximated by $\dot{M}\propto t^{0-1/2}$ and which subsequently may decay as $\dot{M}\propto t^{-5/3}$, producing a photon count rate with a single fast-rise-exponential-decay (FRED)-like profile. In general, the prompt-emission envelope is regulated by a time-dependent mass supply to the central engine, while internal shocks produce the rapid variability. Since we only aim to introduce this framework here, we focus on the simplest single-FRED shape of the prompt emission profiles, while more complex cases involving multiple episodes and interacting shocks will be explored in forthcoming studies. The model indicates correlations between spectral evolution, FRED-pulse narrowing at high energies, and the mass-supply-controlled envelope. Stochastic Lorentz factor variations of ejected mass- or rate-driven shells, superimposed on the Accretion-Modulated envelope, explain the coexistence of smooth global trends and irregular short-timescale features, such as the widths of individual pulses in long GRB light curves, offering diagnostic tools for probing the inner engine activity.

This work presents integrated flux and velocity channel maps of the planetary nebula Abell 30 (A30) inner knot system. The observations were taken with the INTEGRAL spectrograph at the William Herschel Telescope (WHT). Our IFU data cube partially covers knots J1, J2, and completely covers knots J3, J4 in the system. Optical Recombination Lines of C II, He I, He II, N III, O II and Collisionally Excited Lines of [Ar IV], [Ar V], [N II], [Ne III], [Ne IV], and [O III] were detected. Our integrated flux maps visualise the ionisation structure and the chemical inhomogeneity in the system previously reported by other groups. We find that ORLs are concentrated in the polar region (J1, J3), whereas the equatorial knots (J2, J4) are dominated by CELs. The flux ratio map of the diagnostic [O III] lambda 5007/4363 Angstrom lines reveals the electron temperature distribution, which shows cold cores of 15,000 K in knots J3 and J4 surrounded by a hot outer layer of above 20,000 K. Our channel maps show positive and negative velocity excursions from the systemic value among the ions. Several ions show variation in their velocity structures from their lower-energy-level counterparts, including [Ar IV] and [Ar V], [Ne III] and [Ne IV], and He I and He II. New recurrent velocity structures are identified in the low-density regions where the ions move much faster compared to their surrounding environments. The velocity dispersion measurements highlight extreme turbulence in some of the ions (sigma_vrad approx 140 km/s), consistent with supersonic/hypersonic motion driven by shocks. The forbidden line species [N II] exhibits lower turbulence (sigma_vrad approx 50-60 km/s), tracing denser, less-turbulent gases. Based on our data, we conclude both the ionisation and kinematic studies hint at shock heating and multiple ejection history in the evolutionary pathway of A30.

Among thousands of extragalactic $γ$-ray emitters, only a handful of distant ($z >$ 3) sources are detected, yet they are cruial probes shedding light on the cosmic evolution of jets of active galactic nuclei and the initial phase of mass growth of supermassive black holes. Here, we report on a multi-band study of a radio quasar B3 1239+376 with $z$ = 3.82. By analyzing the Fermi-LAT data, a significant (globally 7.7$σ$) $γ$-ray source in its direction, with an estimated association probability of 0.91, is observed in a half-year period of 2025. The analyses also reveal the emergence of co-spatial $γ$-ray residues in prior epochs. Moreover, the $γ$-ray and infrared light curves are likely correlated, particularly, the two emissions climb to the peaking values at the same time. The temporal coincidence establishes a firm association relationship between the $γ$-ray source and the quasar. Therefore, B3 1239+376 is proposed as the {\it third} most distant $γ$-ray-emitting blazar to date. Benefiting from the multi-wavelength observations, broadband spectral energy distributions in different flux states are drawn and reproduced by the classic one-zone leptonic radiation model to investigate the jet properties. Considering the recent brightening in $γ$ rays, prompt follow-up observations are encouraged, especially radio interferometry observations which may catch the potential ejection of a new jet blob.

We investigate the structure of strange dwarfs, modeled as hybrid compact stars composed of a self bound strange quark matter core surrounded by a white dwarf like crust, within a fully relativistic framework. Static configurations are constructed by solving the Tolman Oppenheimer Volkoff equations, and uniformly rotating configurations are modeled within the Hartle Thorne slow rotation expansion (to ${\cal O}(Ω^2)$). We therefore interpret results at large fractional spins conservatively, and use the Kepler frequency mainly as a reference scale for comparing different masses and models. The stellar matter is described using a hybrid equation of state, in which the crust is modeled by a degenerate electron ion system and the core by the MIT Bag Model. By comparing strange dwarfs with conventional white dwarfs across a range of rotation rates, we show that rotation inflates the radius and can reduce (in a quantifiable way) the separation between the two families in the $(M,R)$ plane, potentially masking structural signatures associated with the presence of a quark core. Our results highlight the importance of accounting for rotational effects when interpreting mass radius measurements and other global observables in the context of searches for exotic compact objects in current and future high precision surveys.

Quasars hosting $\gtrsim 10^{9}\,M_\odot$ black holes at $z>6$ challenge growth scenarios that start from light seeds and assume accretion within already formed galaxies. Motivated by the James Webb Space Telescope (JWST) discovery of Little Red Dots (LRDs), which suggests that $\sim 10^{6}\,M_\odot$ black holes can be active in compact, pre-galactic environments, we revisit early black hole growth with a minimal cosmology-based framework. We model the accretion history as the smaller of the Bondi inflow rate and the Eddington-limited rate, where the Bondi rate is set by the supply of overdense primordial gas whose density declines with cosmic expansion, and the Eddington rate captures regulation by radiative feedback. By fitting the observed masses and luminosities of J0313--1806 ($z=7.64$) and J0100+2802 ($z=6.30$) with Bayesian inference, we infer initial conditions that favor massive seed black holes with initial mass $M_0 \sim 10^{6}\,M_\odot$, formed at $z\sim20{-}30$ in environments with baryonic overdensity factors $f_ρ\gtrsim 50$ relative to the cosmic mean. The resulting growth histories include a prolonged supply-limited stage, and they reproduce the observed quasar masses without requiring sustained Eddington accretion or any super-Eddington episodes. The inferred seed mass scale is consistent with black holes produced by the collapse of quantum-degenerate fermion dark matter cores, providing a physically defined pathway to massive seeds at the redshifts implied by LRD phenomenology.

We present mid-infrared (MIR) observations of the Seyfert 1 galaxy NGC 4151 using the Large Binocular Telescope Interferometer (LBTI). We took open-loop Fizeau images with 66-104 mas (5.8-9.1 pc) resolution in the N-band (at $8.7$ and $10.5~μm$), using the full resolution of the LBTI -- equivalent to that of a 28.8 m telescope. These images were complemented by AO imaging in the LM-bands ($3.7$ and $4.8~μm$), with 50-62 mas (4.4-5.4 pc) resolution. These images bridge the scales between previous Very Large Telescope Interferometer (VLTI)/MIDI and VLT/VISIR data, delivering ELT-like imaging resolution in the N-band. We resolve a dusty torus, (diameter 32 pc, PA$=125^{\circ}$), and detect dusty clouds within the narrow line region. Matching the resolution across four bands, we measured spatially-resolved SEDs of the central $\sim 100$ pc. Modified blackbody fitting revealed dust temperature and extinction profiles, indicating both heating from the accretion disk and additional shock heating due to the radio jet. The spatial coincidence of ionized emission (e.g., [Fe II] and [O III]), extended MIR structures, and radio features further supports the interpretation of shock heating. Comparison with NGC 1068 tests the Unified Model of Active Galactic Nuclei (Unified Model of AGN): Structures are similar, despite differences in orientation and Eddington ratio. NGC 4151's torus is smaller than NGC 1068's following a $r\propto L^{0.5}$ scaling. These thirty-meter-telescope class observations of NGC 4151 and NGC 1068 highlight the need to revise MIR radiative transfer models of AGN to account for jet-related heating.

We apply Simulation-Based Inference ('SBI') to the cosmic dipole problem for the first time, measuring the distribution of quasar counts over the sky in the CatWISE2020 ('CatWISE') sample. We show that the quadrupole anisotropy in CatWISE can be attributed to the correlation between WISE's scanning law and photometric uncertainty in the $W1$ and $W2$ magnitudes, inducing an Eddington bias which varies with sky position. After explicitly modelling this with SBI, we use a neural likelihood estimator to find the posterior distribution for CatWISE's dipole, confirming the presence of a dipole twice as large as the CMB expectation but more seriously misaligned with the CMB direction ($\approx 3 σ$). We also use our learned likelihood to infer the Bayesian evidence, learning that models which increase the scale of CatWISE's photometric errors are most favoured. This is strong evidence that the sample's errors are underestimated or that there is an additional, unresolved systematic producing the same effect as Eddington bias. While our results indicate that the cosmic dipole excess is a persistent issue for $Λ$CDM, we showcase that SBI can untangle the subtle and complex systematic issues affecting any sample derived from real astronomical data.

We present the first detailed numerical modeling of the radio emission from MACS J0018.5+1626 as part of the Improved Constraints on Mergers with SZ, Hydrodynamical simulations, Optical, and X-ray (ICM-SHOX) project. By matching X-ray, thermal and kinetic Sunyaev--Zel'dovich, optical and lensing observables to simulations, the ICM-SHOX pipeline indicates that MACS J0018.5+1626 is undergoing a binary merger close to pericenter passage and is observed along a line of sight nearly aligned with the merger axis. We perform three-dimensional magnetohydrodynamic simulations of binary cluster mergers coupled to tracer particles and a Fokker--Planck solver to model the radio emission. Exploring variations in the most likely initial conditions within the ICM-SHOX parameter space, such as the relative cluster velocity and impact parameter, we find that the resulting merger configuration consistently produces two merger-driven shocks with typical average Mach numbers $\mathcal{M}_s \sim 2$--$3$ with corresponding standard deviations of $σ_{\mathcal{M}} \sim 0.5$--$1.5$. Within this framework, we examine the cluster conditions under which standard diffusive shock acceleration can reproduce LOFAR observations. In particular, we discuss the possibility that the apparent radio halo seen by LOFAR arises from the superposition of two radio relics viewed nearly face-on.

The Large and Small Magellanic Clouds (LMC and SMC, respectively) are the largest satellite galaxies of the Milky Way (MW) and their interactions with each other have given rise to multiple stellar substructures in their periphery as well as the gaseous Magellanic Stream. To better understand the origin of the stellar substructures and constrain their past orbit, we model the past 2.5 Gyr of the interactions between the MW and the LMC and SMC using N-body simulations. Due to the strong interactions, analytical orbit integrations are insufficient to analyze the past galaxy orbits accurately. Therefore, we use a genetic algorithm in combination with N-body simulations to determine the LMC and SMC initial positions and velocities 2.5 Gyr ago that result in the Magellanic Clouds (MCs) arriving near their observed locations and velocities at the current time. After running ~8,000 simulations, our best matching model includes two close interactions between the MCs (940 Myr and 140 Myr ago) and reproduces some observed features of the MCs, including the LMC disc warp, a ring-shaped overdensity in the LMC, the tidal expansion of the SMC, and a greater distance dispersion on the eastern side of the SMC. The LMC disc warp is caused by the most recent interaction with the SMC, which occurred ~140 Myr before the present. The interaction causes global ripples in the LMC disc with a mean amplitude of 1.3 kpc.

ALMA continuum measurements of the local ULIRG IRAS 07251-0248 E at 667$μ$m reveal an extremely compact (R < 27 pc) and bright ($T_B$ >200 K) nucleus with an absorbing foreground envelope and a surrounding (R ~ 75 pc) disk or torus seen nearly face-on. The bright and unresolved nuclear emission implies large optical depths ($τ_{667μm}$ >0.5, corresponding to $N_H > 10^{25}$ cm^-2) of hot dust at >500 K. In addition, JWST observations of the source show strong mid-infrared (mid-IR) absorption in the ro-vibrational bands of H2O nu_2=1-0 (5-7 $μ$m) and of other species including CO, HCN, C2H2, CH4, and CO2, and Herschel/PACS observations exhibit strong and saturated absorption due to OH, H2O, CH^+, and CH. We propose a model in which the unresolved ALMA submillimeter and JWST mid-IR continua trace the same nuclear source, the former penetrating deep into the nucleus and the latter probing the nuclear photosphere. The continuum model, which includes trapping of photons (the "greenhouse" effect), indicates that the nuclear ($R_h$ ~ 13 pc) luminosity and luminosity surface density are ~10$^{12}$ Lsun and $Σ_{bol}$~ 5e8 Lsun pc^-2, arising from an active galactic nucleus (AGN) so buried that high-ionization lines are completely obscured. The observed mid-IR gas-phase molecular bands probe outflowing gas with velocities of ~160 km s^-1 and are reproduced with the predicted $T_{dust}$ profile, while the far-IR molecular absorption lines are generated in the surrounding thick disk or torus with $τ_{100 μm}$~10. We conclude that IRAS 07251-0248 harbors a compact obscured nucleus (CON) that hides an AGN currently emitting at quasar luminosity. While the observed outflow could be driven by radiation pressure, we favor the scenario of a (partially) energy-conserving hot bubble caught in a very early phase of the expulsion of the highly concentrated gas at the galactic nucleus.

Large Gaia data set shows substantial misfit between models and observation for cool white dwarfs with $T_{\rm eff}<6000\,\rm K$, resulting in severe underestimation of masses of these stars. We aim to understand the underlying modelling issues. State of the art atmosphere models have been applied to analyse the Gaia DR3 sample of white dwarfs as well as quantum mechanical calculations to quantify formation and stability of different hydrogen species in the atmospheres of these stars. We reconcile the models and observations when we artificially suppress formation of $\rm H_3^+$ species, a process which substantially alters the chemical equilibrium at $T_{\rm eff}<6000\,\rm K$, resulting in an overabundance of free electrons and $\rm H^-$, and strengthening of $\rm H^-$ bound-free absorption. Removing the $\rm H_3^+$ species from chemical equilibrium consideration makes ionization of hydrogen atoms the main source of free electrons, with the resulting models reproducing well the Gaia white dwarfs cooling branch. Because $\rm H_3^+$ must form under the considered conditions, likely it is the overestimation of its partition function and resulting abundance or the formation of $\rm H_3^-$ or another anionic species that suppresses the formation of $\rm H^-$ as a countercharge for $\rm H_3^+$ in current models. Chemical equilibrium in cool, hydrogen atmospheres white dwarfs must be reconsidered in respect to the abundance of $\rm H_3^+$ species and presence of unaccounted charge species.

The parameter $f_{\textrm{NL}}$ measures the local non-Gaussianity in the primordial energy fluctuations of the Universe, with any deviation from $f_{\textrm{NL}}=0$ providing key constraints on inflationary models. Galaxy clustering is sensitive to $f_{\textrm{NL}}$ at large scale modes and the next generation of galaxy surveys will approach a statistical error of $σ_{f_{\textrm{NL}}}\sim1$. However, the systematic errors on these constraints are dominated by the degeneracy of $f_{\textrm{NL}}$ with the galaxy bias parameters $b_1$ (galaxy overdensities caused by mass perturbations) and $b_φ$ (galaxy overdensities caused by primordial potential perturbations). It has been shown that the assumed scaling of $b_φ(z)=2δ_c (b_1(z)-1)$ is not accurate for realistically simulated galaxies, and depends both on the galaxy selection and the way that galaxies are modeled. To address this, we leverage the CAMELS-SAM pipeline to explore how varying parameters of galaxy formation affects $b_φ$ and $b_1$ for various galaxy selections. We run separate-universe N-body simulations of $L=205 h^{-1}$ cMpc and $N=1280^3$ to measure $b_φ$, and run 55 unique instances of the Santa Cruz semi-analytic model with varying parameters of stellar and AGN feedback. We find the behavior and evolution of a SC-SAM model's stellar-, SFR- and sSFR- to halo mass relationships track well with how $b_1$ and $b_φ(b_1)$ change across redshift and selection for the SC-SAM. We find our variations of the SC-SAM encapsulate the $b_φ$ behavior previously measured in IllustrisTNG, the Munich SAM, and Galacticus.Finally, we identify sSFR selections as particularly robust to varied galaxy modeling.

We present a systematic study of the environments of 25 luminous quasars at $z > 6.5$ from the ASPIRE program. Using JWST/NIRCam WFSS data, we identified 487 galaxies at $5.3 \lesssim z \lesssim 7.0$ exhibiting [OIII] emission. Among these, 122 [OIII] emitters lie within $|Δv_{\rm los}| < 1000~{\rm km~s^{-1}}$ of the quasars, corresponding to a $\sim9.4$-fold enhancement relative to the average galaxy density at other redshifts. Furthermore, we identified 16 [CII]-emitting galaxies at the quasar redshifts from ALMA mosaic observations. A cross-correlation function (CCF) analysis between quasars and [OIII]+[CII] emitters yields a cross-correlation length of $r_0^{\rm QG} = 8.68^{+0.51}_{-0.55}~h^{-1}~\mathrm{cMpc}$ and a auto-correlation of $r_0^{\rm{QQ}}=15.76^{+2.48}_{-2.70}~h^{-1}~{\rm cMpc}$, indicating that $z \sim 7$ quasars reside in dark matter halos with $M_{\rm halo} = 10^{12.27^{+0.21}_{-0.26}}~M_\odot$. Notably, the number of [OIII]-emitting galaxies at quasar redshifts varies significantly from field to field, ranging from zero to twenty, highlighting a diverse quasar environment. Remarkably, seven quasars trace significant galaxy overdensities (i.e., protoclusters), with $δ_{\rm gal} > 5$ within a volume of $V \sim 500~{\rm cMpc^3}$. We also find that $|Δv_{\rm los}|$ increases rapidly toward smaller galaxy-quasar separations in protocluster fields, consistent with galaxy kinematics around extremely massive halos in cosmological simulations. By combining JWST and ALMA data, we reveal the complex and diverse environments of these early quasars, providing robust evidence that the earliest luminous quasars are effective tracers of galaxy overdensities, albeit with substantial field-to-field variation.

We present diffHOD-IA, a fully differentiable implementation of a halo occupation distribution (HOD) model that incorporates galaxy intrinsic alignments (IA). Motivated by the diffHOD framework, we create a new implementation that extends differentiable galaxy population modeling to include orientation-dependent statistics crucial for weak gravitational lensing analyses. Our implementation combines this HOD formulation with an IA model, enabling end-to-end automatic differentiation from HOD and IA parameters through to the galaxy field. We additionally extend this framework to differentiably model two-point correlation functions, including galaxy clustering and IA statistics. We validate diffHOD-IA against the reference halotools-IA implementation using the Bolshoi-Planck simulation, demonstrating excellent agreement across both one-point and two-point statistics. We verify the accuracy of gradients computed via automatic differentiation by comparison with finite-difference estimates for both HOD and IA parameters. We present science use cases leveraging gradients in the simulations to recover the IA parameters of a galaxy field representative of the TNG300 simulation. Finally, we apply diffHOD-IA in a Hamiltonian Monte Carlo analysis and compare its performance with halotools-IA and a neural-network-based emulator, IAEmu. Unlike emulator-based approaches, diffHOD-IA provides differentiability at the catalog level, enabling integration into field-level inference pipelines and extension to arbitrary summary statistics for next-generation weak-lensing analyses. Our code is publicly available.

We use data from the JWST ASPIRE Wide Field Slitless Spectroscopy (WFSS) program to measure the auto-correlation function of [O,{\sc iii}] emitters at 5.3$<z<$7.0 and the quasar--[O,{\sc iii}] emitter cross-correlation around 25 ASPIRE quasars (6.51$<z<$6.82; $\langle z\rangle=6.6$). We use synthetic source injection to calibrate the selection function, which we combine with the large-volume FLAMINGO-10k simulation (2.8,cGpc box) to construct realistic mock observations. Our simulation-based approach captures nonlinear structure growth and scale-dependent bias on small scales and derives covariance matrices that include cosmic variance. The clustering yields correlation lengths of $r_0^{\rm GG}=4.7^{+0.4}{-0.5},h^{-1}$,cMpc for the [O,{\sc iii}] auto-correlation with fixed slope $γ{\rm GG}=1.8$, and $r_0^{\rm QG}=8.7^{+0.8}{-0.9},h^{-1}$,cMpc for the quasar--[O,{\sc iii}] cross-correlation with $γ{\rm QG}=2.0$. We infer $\log(M_{h,{\rm min}}^{[{\rm O,III}]}/M_\odot)=10.5^{+0.1}{-0.1}$ for [O,{\sc iii}] emitters and $\log(M{h,{\rm min}}^{\rm QSO}/M_\odot)=12.1^{+0.3}{-0.4}$ for quasars. These imply duty cycles of $2.5^{+1.0}{-0.8}$,per,cent for [O,{\sc iii}] emitters and $0.3^{+4.0}{-0.3}$,per,cent for quasars, corresponding to UV-bright lifetimes of $t{\rm Q}=2.6^{+30}_{-2.5}$,Myr (less than 10,per,cent of a Salpeter $e$-folding time). The results indicate that the observed UV-luminous phase contributes little to total SMBH growth, placing tight constraints on early black-hole formation.

(Abridged) We present stellar kinematics of the MUSE Most Massive Galaxies (M3G) Survey, comprising 25 galaxies brighter than -25.7 mag in the Ks-band and stellar mass above ~6x10^11 Msun. Galaxies are divided between the brightest cluster galaxies (BCGs) and lower-ranked (in brightness) galaxies (non-BCGs) in three rich galaxy clusters within the core of the Shapley super cluster. We find several velocity maps with rich kinematic structure, including multiple spin reversals within the region encompassing central two effective radii, typically associated with BCGs. The majority of BCGs show rotation around the major-axis, at least in one of the visible velocity components. These kinematic structures are possible only if galaxies have non-axisymmetric shapes and contain several orbital families with both prograde and retrograde rotations. There are six fast rotators in the M3G sample, all among non-BCGs, and typically ranked below the 3rd brightest galaxy. Based on the properties of the h3 Gauss-Hermite moment, fast rotation can be linked to the dominance of prograde rotating short-axis tubes in the orbital distribution. Slow rotators are BCGs or the second and sometimes third brightest galaxies, indicating that the galaxy mass (brightness) is not the only driver of low spin, but that the location within the local environment also plays a role. Slow rotators, as evidenced from their multi-spin velocity maps, require more complex orbital structures. Furthermore, some BCGs show kinematic evidence for a secondary component at larger radii, likely not in equilibrium with the main galaxy and possibly made of stars accreted from other cluster galaxies. Multi-spin velocity maps, low angular momentum and additional kinematic components highlight the difference in the evolutionary histories of BCGs (including 2nd ranked galaxies) and non-BCGs.

Hydrocarbons play a key role in shaping the chemistry of the interstellar medium (ISM), but their enrichment and relationship with carbonaceous grains and polycyclic aromatic hydrocarbons (PAHs) still lack clear observational constraints. We report JWST NIRSpec+MIRI/MRS infrared (IR; 3-28 micron) observations of the local ultra-luminous IR galaxy (ULIRG) IRAS 07251-0248, revealing the extragalactic detection of small gas-phase hydrocarbons such as benzene (C$_6$H$_6$), triacetylene (C$_6$H$_2$), diacetylene (C$_4$H$_2$), acetylene (C$_2$H$_2$), methane (CH$_4$), and methyl radical (CH$_3$) as well as deep amorphous C-H absorptions in the solid phase. The unexpectedly high abundance of these molecules indicates an extremely rich hydrocarbon chemistry, not explained by high-temperature gas-phase chemistry, ice desorption or oxygen depletion. Instead, the most plausible explanation is the erosion and fragmentation of carbonaceous grains and PAHs. This scenario is supported by the correlation between the abundance of one of their main fragmentation products, C$_2$H$_2$, and cosmic ray (CR) ionization rate for a sample of local ULIRGs. These hydrocarbons are outflowing at $\sim$160 km/s, which may represent a potential formation pathway for hydrogenated amorphous grains. Our results suggest that IRAS 07251-0248 might not be unique but represents an extreme example of the commonly rich hydrocarbon chemistry prevalent in deeply obscured galactic nuclei.

With tens to hundreds of spectra of white dwarfs being taken each night from multi-object spectroscopic surveys, automated spectral classification is essential as part of efficient data processing. In this study, we design a neural network to classify the spectral type of white dwarfs using a combination of spectra from the Dark Energy Spectroscopic Instrument (DESI) data release~1 and imaging from Pan-STARRS photometry. The trained network has a near 100% accuracy at identifying DA and DB white dwarf spectral types, while having an 85-95% accuracy for identifying all other primary types, including metal pollution. Distinct spectral or photometric features map into separate structures when performing a Uniform Manifold Approximation and Projection (UMAP) dimensionality reduction. Investigating further and looking at multiple epoch spectra, we performed a separate search for objects that have strongly changing spectral signatures using UMAP, discovering 3 new inhomogeneous surface composition ('double-faced') white dwarfs in the process. We lastly show how machine learning has the potential to separate single white dwarfs from double white dwarf binary star systems in a large dataset, ideal for isolating a single star population. The results from all of these techniques show a compelling use of machine learning to boost efficiency in analysing white dwarfs observed in multi-object spectroscopy surveys, at times replacing the need for human-driven spectral classifications. This demonstrates our techniques as powerful tools for batch population analyses, finding outliers as a form of rare subclass detection, and in conducting multi-epoch spectral analyses.

Measurements of the cosmic 21 cm signal need to achieve a high dynamic range to isolate it from bright foreground emissions. Calibration inaccuracies can compromise the spectral fidelity of the smooth foreground continuum, thereby limiting the dynamic range and potentially precluding the detection of the cosmic line signal. In light of this challenge, recent work has proposed using the calibration-independent closure phase to search for the spectral fluctuations of the cosmic 21 cm signal. However, so far there has been only a heuristic understanding of how closure phases map to the cosmological 21 cm power spectrum. This work aims to establish a more accurate mathematical relationship between closure phases and the cosmological power spectrum of the background line signal. Building on previous work, we treat the cosmic signal component as a perturbation to the closure phase and use a delay spectrum approach to estimate its power. We establish the relationship between this estimate and the cosmological power spectrum using standard Fourier transform techniques and validate it using simulated HERA observations. We find that, statistically, the power spectrum estimate from closure phases is approximately equal to the cosmological power spectrum convolved with a foreground-dependent window function, provided that the signal-to-foreground ratio is small. Compared with standard approaches, the foreground dependence of the window function results in an increased amount of mode-mixing and a more pronounced proliferation of foreground power along the line-of-sight dimension of the cylindrical power spectrum. These effects can be mitigated by flagging instances where the window function is broad. Crucial to gaining the necessary sensitivity, this mapping will allow us to average the measurements of closure triads of different shapes based on their imprint in cylindrical Fourier space.

Tailed radio galaxies are powerful tracers of interactions between active galactic nuclei (AGN) and the intracluster medium (ICM), providing unique insights into cluster dynamics. We present LOw Frequency ARray (LOFAR) 144 MHz and uGMRT 400 MHz observations of the cluster MACS J1354.6+7715 (z = 0.3967) to investigate the radio emission associated with its member galaxies and the cluster environment. The dominant tailed radio galaxy in the cluster exhibits a sharply bent tail extending over approximately 300 kpc, with the spectral index steepening from approximately -0.46 +/- 0.21 near the AGN core to approximately -2.43 +/- 0.30 in the outermost regions. Synchrotron modelling of the tail yields a radiative age of 150 +/- 10 Myr, implying a galaxy velocity of 1956 +/- 130 km s^-1, which is of order ~ 0.9 times the escape velocity. We find no evidence of relics or halos in our radio images, and the X-ray morphology from Chandra appears relatively undisturbed, suggesting that the system is a pre-merging candidate. Our results indicate that the radio galaxy is undergoing its first infall into the cluster, providing an excellent laboratory for studying the impact of the ICM on AGN activity and galaxy evolution, and demonstrating how multi-frequency radio observations of tailed galaxies can uniquely probe both AGN lifecycles and the early stages of cluster assembly.

The distribution of dark matter in the inner regions of galaxies poses a key challenge for small-scale ΛCDM cosmology. While cold dark matter simulations predict cuspy inner density profiles, observations of low surface brightness (LSB) and dwarf galaxies often favour cored profiles, an issue known as the cusp-core problem. We investigate this problem by comparing four dark matter halo profiles: NFW (cuspy), Einasto (intermediate), Burkert (cored), and pseudo-isothermal (pISO) (cored) in a pilot sample of $11$ galaxies from the GMRT archive atomic gas survey (GARCIA). We have performed mass modelling using Markov Chain Monte Carlo (MCMC) techniques, utilising rotation curves derived from robust 3D Kinematic modelling. Baryonic contributions from stars derived using stellar kinematics based on $3.6μ\mathrm{m}$ or $r$-band photometry via Multi-Gaussian Expansion (MGE) combined with Jeans Anisotropic Model (JAM) and from gas, calculated directly from the gas surface density (HI + He) without assuming any predefined functional form, are included. Our mass modelling shows that all halo profiles provide statistically good fits, yielding consistent estimates of halo mass and stellar mass-to-light ratio. To validate our analysis, we examine the stellar-to-halo mass relation and find broad agreement with empirical models. Non-parametric density profiles derived from baryon-subtracted rotation curves show that NFW fits the inner regions best, while all profiles converge in the outskirts. Future studies with a larger sample from GARCIA will be helpful in refining this trend and addressing the cusp-core issue in greater depth.

In both observed and simulated galaxies, disk morphologies become more prevalent at higher masses and lower redshifts. To elucidate the physical origin of this trend, we develop a simple analytical model in which galaxy morphology is governed by the competition between rotational support and turbulence in a gravitational potential of a dark matter halo and the galaxy itself, and a disk forms when the potential steepens due to the accumulation of baryons in the halo center. The minimum galaxy mass required for this transition decreases with an increasing dark matter contribution within the galaxy, making more concentrated halos more prone to forming disks. Our model predicts that galaxy sizes behave qualitatively differently before and after disk formation: after disks form, sizes are governed by the halo spin, in agreement with classical models, whereas before disk formation, sizes are larger and set by the scale on which turbulent motions, which dominate over rotation, can be contained. We validate our model against the results of the TNG50 cosmological simulation and, despite the simplicity of the model, find remarkable agreement. In particular, our model explains the increase with redshift in the critical halo mass for disk formation, reported in both simulations and observations, as a consequence of the evolution of the halo mass-concentration and baryonic mass-halo mass relations. This redshift trend therefore supports the recent proposal that it is the steepening of the gravitational potential that causes disk formation, while other effects discussed in the literature, such as potential deepening and hot gaseous halo formation, can still play important roles in the transition from early turbulent to dynamically cold disks.

Measuring the mass distribution of supermassive black holes (SMBHs) over cosmic time remains particularly challenging for the low mass ($M_{\bullet}\lesssim10^8~M_\odot$) population at $z>1$. This population is also the most sensitive to SMBH seeding and early growth models. In this work we construct a semi-empirical model for the redshift evolution of the TDE rate under multiple SMBH mass function prescriptions, and show that the observed redshift-dependent rate of TDEs is very sensitive to the SMBH mass function and its evolution with redshift. We further incorporate galaxy-scale processes that evolve with redshift -- namely, increasing galaxy nuclear stellar densities, enhanced galaxy-galaxy merger rates, dust obscuration, and a possible top-heavy IMF at early cosmic times -- and quantify their combined impact on the TDE rate. We find that including these effects generally results in a volumetric TDE rate that increases with redshift until a maximum near cosmic noon, before declining at higher redshift where SMBHs that can disrupt stars become increasingly scarce. We forecast TDE rates in the Rubin LSST and the Roman High Latitude Time Domain Survey, alongside expectations for serendipitous TDE rates in the JWST COSMOS-Web survey. Finally, we provide a methodology for using a flux-limited survey of TDEs in LSST to directly constrain the redshift evolution of the SMBH mass function.

Measurements of the mean free path of ionizing photons from high-redshift quasar spectra at $z \sim 5$-$6$ constrain the reionization history, but interpreting them requires modeling the kiloparsec-scale clumping that large-volume reionization simulations cannot resolve. We present a deep learning emulator for the mean free path (MFP) trained on high-resolution cosmological radiative transfer simulations of ionization fronts sweeping through small 2 comoving~Mpc/h volumes. Using a residual multi-layer perceptron neural network, we predict the MFP at a given redshift as a function of the reionization redshift, photoionization rate, wavelength, and box-scale density, achieving a median relative error of 1.6\% across nearly four orders of magnitude in MFP. Integrating its predictions over box-scale overdensity and an extended reionization history allows the emulator to predict the global MFP. We apply the emulator to extended reionization histories constrained by observed photoionization rates, finding that models prefer late reionization with substantial neutral fractions persisting at $z \lesssim 6$. Fitting a parametric ionization history yields a midpoint of reionization of $z_{\rm re} = 6.8\pm 1.2$ for reionization durations consistent with Planck and kinetic Sunyaev-Zeldovich constraints, and the universe being $10\%$ neutral still at $z_{\rm re} < 5.8 ~(6.3)$ at 1~(2)$σ$. Global ionizing emissivity inferences using measurements of the photoionization rate and MFP plus our emulator, which avoids common power-law assumptions, suggest a factor of $2-3$ decline between $z = 6$ and $4.8$, in agreement with previous studies. Our method provides an efficient (and more converged) alternative to large-volume radiative-hydrodynamic simulations of reionization for interpreting MFP measurements, and can also serve as a subgrid prescription for the ionizing opacity within such simulations.

Baryonic feedback processes associated with galaxy formation directly influence the large-scale structure by redistributing gas. Recent measurements of the kinetic Sunyaev-Zel'dovich effect and stacks of X-ray emission from optically selected galaxy clusters suggest that feedback from Active Galactic Nuclei (AGN) is more efficient at expelling gas from low-mass clusters than previously thought. The measurement of the cross-correlation between cosmic shear and diffuse X-ray emission provides a new probe of the distribution of gas in groups and clusters. We use the FLAMINGO cosmological, hydrodynamical simulations to examine the X-ray--cosmic-shear cross-correlation. The cross-correlation is most sensitive to the distribution of gas in haloes with masses $10^{14}\leq M_{200\mathrm{c}}/\mathrm{M}_{\odot}\leq10^{15}$. It is sensitive to the strength of feedback, but the effects of variations in cosmology and baryonic physics are largely degenerate. We compare the FLAMINGO predictions with the cross-correlation between cosmic shear from the Dark Energy Survey and ROSAT all-sky X-ray maps. We find that, if we neglect the X-ray emission from AGN that would remain unresolved by ROSAT, then the fiducial FLAMINGO model is in excellent agreement with the data, while models with stronger or weaker feedback are ruled out. However, if we account for unresolved AGN, either using the direct FLAMINGO predictions or by abundance matching to the observed (extrapolated) AGN luminosity function, then models with stronger feedback are preferred. We conclude that to exploit the potential of the X-ray--lensing cross-correlation, it will be necessary to resolve fainter AGN, and to use external constraints to break the degeneracy between baryonic feedback and cosmology.