arXiv Daily Digest - 2026-02-02

Large language models (LLMs) often make accurate next token predictions but their confidence in these predictions can be poorly calibrated: high-confidence predictions are frequently wrong, and low-confidence predictions may be correct. This miscalibration is exacerbated by preference-based alignment methods breaking the link between predictive probability and correctness. We introduce a Calibration Aware Token-level Training Objective (CATTO), a calibration-aware objective that aligns predicted confidence with empirical prediction correctness, which can be combined with the original preference optimization objectives. Empirically, CATTO reduces Expected Calibration Error (ECE) by 2.22%-7.61% in-distribution and 1.46%-10.44% out-of-distribution compared to direct preference optimization (DPO), and by 0.22%-1.24% in-distribution and 1.23%-5.07% out-of-distribution compared to the strongest DPO baseline. This improvement in confidence does not come at a cost of losing task accuracy, where CATTO maintains or slightly improves multiple-choice question-answering accuracy on five datasets. We also introduce Confidence@k, a test-time scaling mechanism leveraging calibrated token probabilities for Bayes-optimal selection of output tokens.

Humans develop a series of cognitive defenses, known as epistemic vigilance, to combat risks of deception and misinformation from everyday interactions. Developing safeguards for LLMs inspired by this mechanism might be particularly helpful for their application in high-stakes tasks such as automating compliance with data privacy laws. In this paper, we introduce Dynamic Epistemic Fallback (DEF), a dynamic safety protocol for improving an LLM's inference-time defenses against deceptive attacks that make use of maliciously perturbed policy texts. Through various levels of one-sentence textual cues, DEF nudges LLMs to flag inconsistencies, refuse compliance, and fallback to their parametric knowledge upon encountering perturbed policy texts. Using globally recognized legal policies such as HIPAA and GDPR, our empirical evaluations report that DEF effectively improves the capability of frontier LLMs to detect and refuse perturbed versions of policies, with DeepSeek-R1 achieving a 100% detection rate in one setting. This work encourages further efforts to develop cognitively inspired defenses to improve LLM robustness against forms of harm and deception that exploit legal artifacts.

Wireless ethical hacking relies heavily on skilled practitioners manually interpreting reconnaissance results and executing complex, time-sensitive sequences of commands to identify vulnerable targets, capture authentication handshakes, and assess password resilience; a process that is inherently labour-intensive, difficult to scale, and prone to subjective judgement and human error. To help address these limitations, we propose WiFiPenTester, an experimental, governed, and reproducible system for GenAI-enabled wireless ethical hacking. The system integrates large language models into the reconnaissance and decision-support phases of wireless security assessment, enabling intelligent target ranking, attack feasibility estimation, and strategy recommendation, while preserving strict human-in-the-loop control and budget-aware execution. We describe the system architecture, threat model, governance mechanisms, and prompt-engineering methodology, and empirical experiments conducted across multiple wireless environments. The results demonstrate that GenAI assistance improves target selection accuracy and overall assessment efficiency, while maintaining auditability and ethical safeguards. This indicates that WiFiPenTester is a meaningful step toward practical, safe, and scalable GenAI-assisted wireless penetration testing, while reinforcing the necessity of bounded autonomy, human oversight, and rigorous governance mechanisms when deploying GenAI in ethical hacking.

Semantic caching has emerged as a pivotal technique for scaling LLM applications, widely adopted by major providers including AWS and Microsoft. By utilizing semantic embedding vectors as cache keys, this mechanism effectively minimizes latency and redundant computation for semantically similar queries. In this work, we conceptualize semantic cache keys as a form of fuzzy hashes. We demonstrate that the locality required to maximize cache hit rates fundamentally conflicts with the cryptographic avalanche effect necessary for collision resistance. Our conceptual analysis formalizes this inherent trade-off between performance (locality) and security (collision resilience), revealing that semantic caching is naturally vulnerable to key collision attacks. While prior research has focused on side-channel and privacy risks, we present the first systematic study of integrity risks arising from cache collisions. We introduce CacheAttack, an automated framework for launching black-box collision attacks. We evaluate CacheAttack in security-critical tasks and agentic workflows. It achieves a hit rate of 86\% in LLM response hijacking and can induce malicious behaviors in LLM agent, while preserving strong transferability across different embedding models. A case study on a financial agent further illustrates the real-world impact of these vulnerabilities. Finally, we discuss mitigation strategies.

Chain-of-thought (CoT) reasoning provides a significant performance uplift to LLMs by enabling planning, exploration, and deliberation of their actions. CoT is also a powerful tool for monitoring the behaviours of these agents: when faithful, they offer interpretations of the model's decision making process, and an early warning sign for dangerous behaviours. However, optimisation pressures placed on the CoT may cause the model to obfuscate reasoning traces, losing this beneficial property. We show that obfuscation can generalise across tasks; models that learn to obfuscate reasoning involving reward hacking (e.g. accessing and utilising leaked information) generalise both the reward hacking behaviour and its obfuscation in CoT to unseen reward hacking settings. Most worryingly, we show that obfuscation of CoT reasoning, and its generalisation across tasks, also follows when we penalise only the model's final actions after closing its CoT. Our findings suggest that current practices of penalising harmful generations may inadvertently lead to a reduction in the broader monitorability of LLMs in unpredictable ways.

Resolving complex information needs that come with multiple constraints should consider enforcing the logical operators encoded in the query (i.e., conjunction, disjunction, negation) on the candidate answer set. Current retrieval systems either ignore these constraints in neural embeddings or approximate them in a generative reasoning process that can be inconsistent and unreliable. Although well-suited to structured reasoning, existing neuro-symbolic approaches remain confined to formal logic or mathematics problems as they often assume unambiguous queries and access to complete evidence, conditions rarely met in information retrieval. To bridge this gap, we introduce OrLog, a neuro-symbolic retrieval framework that decouples predicate-level plausibility estimation from logical reasoning: a large language model (LLM) provides plausibility scores for atomic predicates in one decoding-free forward pass, from which a probabilistic reasoning engine derives the posterior probability of query satisfaction. We evaluate OrLog across multiple backbone LLMs, varying levels of access to external knowledge, and a range of logical constraints, and compare it against base retrievers and LLM-as-reasoner methods. Provided with entity descriptions, OrLog can significantly boost top-rank precision compared to LLM reasoning with larger gains on disjunctive queries. OrLog is also more efficient, cutting mean tokens by $\sim$90\% per query-entity pair. These results demonstrate that generation-free predicate plausibility estimation combined with probabilistic reasoning enables constraint-aware retrieval that outperforms monolithic reasoning while using far fewer tokens.

Emergent Misalignment refers to a failure mode in which fine-tuning large language models (LLMs) on narrowly scoped data induces broadly misaligned behavior. Prior explanations mainly attribute this phenomenon to the generalization of erroneous or unsafe content. In this work, we show that this view is incomplete. Across multiple domains and model families, we find that fine-tuning models on data exhibiting specific character-level dispositions induces substantially stronger and more transferable misalignment than incorrect-advice fine-tuning, while largely preserving general capabilities. This indicates that emergent misalignment arises from stable shifts in model behavior rather than from capability degradation or corrupted knowledge. We further show that such behavioral dispositions can be conditionally activated by both training-time triggers and inference-time persona-aligned prompts, revealing shared structure across emergent misalignment, backdoor activation, and jailbreak susceptibility. Overall, our results identify character formation as a central and underexplored alignment risk, suggesting that robust alignment must address behavioral dispositions rather than isolated errors or prompt-level defenses.

On-policy deep reinforcement learning remains a dominant paradigm for continuous control, yet standard implementations rely on Gaussian actors and relatively shallow MLP policies, often leading to brittle optimization when gradients are noisy and policy updates must be conservative. In this paper, we revisit policy representation as a first-class design choice for on-policy optimization. We study discretized categorical actors that represent each action dimension with a distribution over bins, yielding a policy objective that resembles a cross-entropy loss. Building on architectural advances from supervised learning, we further propose regularized actor networks, while keeping critic design fixed. Our results show that simply replacing the standard actor network with our discretized regularized actor yields consistent gains and achieve the state-of-the-art performance across diverse continuous-control benchmarks.

Flow matching is a scalable generative framework for characterizing continuous normalizing flows with wide-range applications. However, current state-of-the-art methods are not well-suited for modeling dynamical systems, as they construct conditional paths using linear interpolants that may not capture the underlying state evolution, especially when learning higher-order dynamics from irregular sampled observations. Constructing unified paths that satisfy multi-marginal constraints across observations is challenging, since naïve higher-order polynomials tend to be unstable and oscillatory. We introduce SplineFlow, a theoretically grounded flow matching algorithm that jointly models conditional paths across observations via B-spline interpolation. Specifically, SplineFlow exploits the smoothness and stability of B-spline bases to learn the complex underlying dynamics in a structured manner while ensuring the multi-marginal requirements are met. Comprehensive experiments across various deterministic and stochastic dynamical systems of varying complexity, as well as on cellular trajectory inference tasks, demonstrate the strong improvement of SplineFlow over existing baselines. Our code is available at: https://github.com/santanurathod/SplineFlow.

Computing the importance of features in supervised classification tasks is critical for model interpretability. Shapley values are a widely used approach for explaining model predictions, but require direct access to the underlying model, an assumption frequently violated in real-world deployments. Further, even when model access is possible, their exact computation may be prohibitively expensive. We investigate whether meaningful Shapley value estimations can be obtained in a zero-shot setting, using only the input data distribution and no evaluations of the target model. To this end, we introduce ExplainerPFN, a tabular foundation model built on TabPFN that is pretrained on synthetic datasets generated from random structural causal models and supervised using exact or near-exact Shapley values. Once trained, ExplainerPFN predicts feature attributions for unseen tabular datasets without model access, gradients, or example explanations. Our contributions are fourfold: (1) we show that few-shot learning-based explanations can achieve high fidelity to SHAP values with as few as two reference observations; (2) we propose ExplainerPFN, the first zero-shot method for estimating Shapley values without access to the underlying model or reference explanations; (3) we provide an open-source implementation of ExplainerPFN, including the full training pipeline and synthetic data generator; and (4) through extensive experiments on real and synthetic datasets, we show that ExplainerPFN achieves performance competitive with few-shot surrogate explainers that rely on 2-10 SHAP examples.

Speech deepfake detection (SDD) focuses on identifying whether a given speech signal is genuine or has been synthetically generated. Existing audio large language model (LLM)-based methods excel in content understanding; however, their predictions are often biased toward semantically correlated cues, which results in fine-grained acoustic artifacts being overlooked during the decisionmaking process. Consequently, fake speech with natural semantics can bypass detectors despite harboring subtle acoustic anomalies; this suggests that the challenge stems not from the absence of acoustic data, but from its inadequate accessibility when semantic-dominant reasoning prevails. To address this issue, we investigate SDD within the audio LLM paradigm and introduce SDD with Auditory Perception-enhanced Audio Large Language Model (SDD-APALLM), an acoustically enhanced framework designed to explicitly expose fine-grained time-frequency evidence as accessible acoustic cues. By combining raw audio with structured spectrograms, the proposed framework empowers audio LLMs to more effectively capture subtle acoustic inconsistencies without compromising their semantic understanding. Experimental results indicate consistent gains in detection accuracy and robustness, especially in cases where semantic cues are misleading. Further analysis reveals that these improvements stem from a coordinated utilization of semantic and acoustic information, as opposed to simple modality aggregation.

Visual geolocalization, the task of predicting where an image was taken, remains challenging due to global scale, visual ambiguity, and the inherently hierarchical structure of geography. Existing paradigms rely on either large-scale retrieval, which requires storing a large number of image embeddings, grid-based classifiers that ignore geographic continuity, or generative models that diffuse over space but struggle with fine detail. We introduce an entity-centric formulation of geolocation that replaces image-to-image retrieval with a compact hierarchy of geographic entities embedded in Hyperbolic space. Images are aligned directly to country, region, subregion, and city entities through Geo-Weighted Hyperbolic contrastive learning by directly incorporating haversine distance into the contrastive objective. This hierarchical design enables interpretable predictions and efficient inference with 240k entity embeddings instead of over 5 million image embeddings on the OSV5M benchmark, on which our method establishes a new state-of-the-art performance. Compared to the current methods in the literature, it reduces mean geodesic error by 19.5\%, while improving the fine-grained subregion accuracy by 43%. These results demonstrate that geometry-aware hierarchical embeddings provide a scalable and conceptually new alternative for global image geolocation.

We evaluate how effectively platform-level parental controls moderate a mainstream conversational assistant used by minors. Our two-phase protocol first builds a category-balanced conversation corpus via PAIR-style iterative prompt refinement over API, then has trained human agents replay/refine those prompts in the consumer UI using a designated child account while monitoring the linked parent inbox for alerts. We focus on seven risk areas -- physical harm, pornography, privacy violence, health consultation, fraud, hate speech, and malware and quantify four outcomes: Notification Rate (NR), Leak-Through (LR), Overblocking (OBR), and UI Intervention Rate (UIR). Using an automated judge (with targeted human audit) and comparing the current backend to legacy variants (GPT-4.1/4o), we find that notifications are selective rather than comprehensive: privacy violence, fraud, hate speech, and malware triggered no parental alerts in our runs, whereas physical harm (highest), pornography, and some health queries produced intermittent alerts. The current backend shows lower leak-through than legacy models, yet overblocking of benign, educational queries near sensitive topics remains common and is not surfaced to parents, revealing a policy-product gap between on-screen safeguards and parent-facing telemetry. We propose actionable fixes: broaden/configure the notification taxonomy, couple visible safeguards to privacy-preserving parent summaries, and prefer calibrated, age-appropriate safe rewrites over blanket refusals.

Large Language Models (LLMs) are increasingly relevant in Software Engineering research and practice, with Automated Bug Fixing (ABF) being one of their key applications. ABF involves transforming a buggy method into its fixed equivalent. A common preprocessing step in ABF involves removing comments from code prior to training. However, we hypothesize that comments may play a critical role in fixing certain types of bugs by providing valuable design and implementation insights. In this study, we investigate how the presence or absence of comments, both during training and at inference time, impacts the bug-fixing capabilities of LLMs. We conduct an empirical evaluation comparing two model families, each evaluated under all combinations of training and inference conditions (with and without comments), and thereby revisiting the common practice of removing comments during training. To address the limited availability of comments in state-of-the-art datasets, we use an LLM to automatically generate comments for methods lacking them. Our findings show that comments improve ABF accuracy by up to threefold when present in both phases, while training with comments does not degrade performance when instances lack them. Additionally, an interpretability analysis identifies that comments detailing method implementation are particularly effective in aiding LLMs to fix bugs accurately.

Reinforcement learning has become a cornerstone for enhancing the reasoning capabilities of Large Language Models, where group-based approaches such as GRPO have emerged as efficient paradigms that optimize policies by leveraging intra-group performance differences. However, these methods typically rely on absolute numerical rewards, introducing intrinsic limitations. In verifiable tasks, identical group evaluations often result in sparse supervision, while in open-ended scenarios, the score range instability of reward models undermines advantage estimation based on group means. To address these limitations, we propose Reinforcement Learning with Relative Rewards (RLRR), a framework that shifts reward shaping from absolute scoring to relative ranking. Complementing this framework, we introduce the Ranking Reward Model, a listwise preference model tailored for group-based optimization to directly generate relative rankings. By transforming raw evaluations into robust relative signals, RLRR effectively mitigates signal sparsity and reward instability. Experimental results demonstrate that RLRR yields consistent performance improvements over standard group-based baselines across reasoning benchmarks and open-ended generation tasks.

Generating realistic graph-structured data is challenging due to discrete structures, variable sizes, and class-specific connectivity patterns that resist conventional generative modelling. While recent graph generation methods employ generative adversarial network (GAN) frameworks to handle permutation invariance and irregular topologies, they typically rely on random edge sampling with fixed probabilities, limiting their capacity to capture complex structural dependencies between nodes. We propose a density-aware conditional graph generation framework using Wasserstein GANs (WGAN) that replaces random sampling with a learnable distance-based edge predictor. Our approach embeds nodes into a latent space where proximity correlates with edge likelihood, enabling the generator to learn meaningful connectivity patterns. A differentiable edge predictor determines pairwise relationships directly from node embeddings, while a density-aware selection mechanism adaptively controls edge density to match class-specific sparsity distributions observed in real graphs. We train the model using a WGAN with gradient penalty, employing a GCN-based critic to ensure generated graphs exhibit realistic topology and align with target class distributions. Experiments on benchmark datasets demonstrate that our method produces graphs with superior structural coherence and class-consistent connectivity compared to existing baselines. The learned edge predictor captures complex relational patterns beyond simple heuristics, generating graphs whose density and topology closely match real structural distributions. Our results show improved training stability and controllable synthesis, making the framework effective for realistic graph generation and data augmentation. Source code is publicly available at https://github.com/ava-12/Density_Aware_WGAN.git.

Medical calculators are fundamental to quantitative, evidence-based clinical practice. However, their real-world use is an adaptive, multi-stage process, requiring proactive EHR data acquisition, scenario-dependent calculator selection, and multi-step computation, whereas current benchmarks focus only on static single-step calculations with explicit instructions. To address these limitations, we introduce MedMCP-Calc, the first benchmark for evaluating LLMs in realistic medical calculator scenarios through Model Context Protocol (MCP) integration. MedMCP-Calc comprises 118 scenario tasks across 4 clinical domains, featuring fuzzy task descriptions mimicking natural queries, structured EHR database interaction, external reference retrieval, and process-level evaluation. Our evaluation of 23 leading models reveals critical limitations: even top performers like Claude Opus 4.5 exhibit substantial gaps, including difficulty selecting appropriate calculators for end-to-end workflows given fuzzy queries, poor performance in iterative SQL-based database interactions, and marked reluctance to leverage external tools for numerical computation. Performance also varies considerably across clinical domains. Building on these findings, we develop CalcMate, a fine-tuned model incorporating scenario planning and tool augmentation, achieving state-of-the-art performance among open-source models. Benchmark and Codes are available in https://github.com/SPIRAL-MED/MedMCP-Calc.

Large language models now solve many benchmark math problems at near-expert levels, yet this progress has not fully translated into reliable performance in real-world applications. We study this gap through contextual mathematical reasoning, where the mathematical core must be formulated from descriptive scenarios. We introduce ContextMATH, a benchmark that repurposes AIME and MATH-500 problems into two contextual settings: Scenario Grounding (SG), which embeds abstract problems into realistic narratives without increasing reasoning complexity, and Complexity Scaling (CS), which transforms explicit conditions into sub-problems to capture how constraints often appear in practice. Evaluating 61 proprietary and open-source models, we observe sharp drops: on average, open-source models decline by 13 and 34 points on SG and CS, while proprietary models drop by 13 and 20. Error analysis shows that errors are dominated by incorrect problem formulation, with formulation accuracy declining as original problem difficulty increases. Correct formulation emerges as a prerequisite for success, and its sufficiency improves with model scale, indicating that larger models advance in both understanding and reasoning. Nevertheless, formulation and reasoning remain two complementary bottlenecks that limit contextual mathematical problem solving. Finally, we find that fine-tuning with scenario data improves performance, whereas formulation-only training is ineffective. However, performance gaps are only partially alleviated, highlighting contextual mathematical reasoning as a central unsolved challenge for LLMs.

As AI becomes more capable, we entrust it with more general and consequential tasks. The risks from failure grow more severe with increasing task scope. It is therefore important to understand how extremely capable AI models will fail: Will they fail by systematically pursuing goals we do not intend? Or will they fail by being a hot mess, and taking nonsensical actions that do not further any goal? We operationalize this question using a bias-variance decomposition of the errors made by AI models: An AI's \emph{incoherence} on a task is measured over test-time randomness as the fraction of its error that stems from variance rather than bias in task outcome. Across all tasks and frontier models we measure, the longer models spend reasoning and taking actions, \emph{the more incoherent} their failures become. Incoherence changes with model scale in a way that is experiment dependent. However, in several settings, larger, more capable models are more incoherent than smaller models. Consequently, scale alone seems unlikely to eliminate incoherence. Instead, as more capable AIs pursue harder tasks, requiring more sequential action and thought, our results predict failures to be accompanied by more incoherent behavior. This suggests a future where AIs sometimes cause industrial accidents (due to unpredictable misbehavior), but are less likely to exhibit consistent pursuit of a misaligned goal. This increases the relative importance of alignment research targeting reward hacking or goal misspecification.

Differentiable matching layers, often implemented via entropy-regularized Optimal Transport, serve as a critical approximate inference mechanism in structural prediction. However, recovering discrete permutations via annealing $ε\to 0$ is notoriously unstable. We identify a fundamental mechanism for this failure: \textbf{Premature Mode Collapse}. By analyzing the non-normal dynamics of the Sinkhorn fixed-point map, we reveal a theoretical \textbf{thermodynamic speed limit}. Under standard exponential cooling, the shift in the target posterior ($O(1)$) outpaces the contraction rate of the inference operator, which degrades as $O(1/ε)$. This mismatch inevitably forces the inference trajectory into spurious local basins. To address this, we propose \textbf{Efficient PH-ASC}, an adaptive scheduling algorithm that monitors the stability of the inference process. By enforcing a linear stability law, we decouple expensive spectral diagnostics from the training loop, reducing overhead from $O(N^3)$ to amortized $O(1)$. Our implementation and interactive demo are available at https://github.com/xxx0438/torch-sinkhorn-asc and https://huggingface.co/spaces/leon0923/torch-sinkhorn-asc-demo. bounded away from zero in generic training dynamics unless the feature extractor converges unrealistically fast.

Tool-Integrated Reasoning (TIR) enables large language models (LLMs) to solve complex tasks by interacting with external tools, yet existing approaches depend on high-quality synthesized trajectories selected by scoring functions and sparse outcome-based rewards, providing limited and biased supervision for learning TIR. To address these challenges, in this paper, we propose AutoTraj, a two-stage framework that automatically learns TIR by repairing and rewarding tool-use trajectories. Specifically, in the supervised fine-tuning (SFT) stage, AutoTraj generates multiple candidate tool-use trajectories for each query and evaluates them along multiple dimensions. High-quality trajectories are directly retained, while low-quality ones are repaired using a LLM (i.e., LLM-as-Repairer). The resulting repaired and high-quality trajectories form a synthetic SFT dataset, while each repaired trajectory paired with its original low-quality counterpart constitutes a dataset for trajectory preference modeling. In the reinforcement learning (RL) stage, based on the preference dataset, we train a trajectory-level reward model to assess the quality of reasoning paths and combine it with outcome and format rewards, thereby explicitly guiding the optimization toward reliable TIR behaviors. Experiments on real-world benchmarks demonstrate the effectiveness of AutoTraj in TIR.

We study a class of iterated empirical risk minimization (ERM) procedures in which two successive ERMs are performed on the same dataset, and the predictions of the first estimator enter as an argument in the loss function of the second. This setting, which arises naturally in active learning and reweighting schemes, introduces intricate statistical dependencies across samples and fundamentally distinguishes the problem from classical single-stage ERM analyses. For linear models trained with a broad class of convex losses on Gaussian mixture data, we derive a sharp asymptotic characterization of the test error in the high-dimensional regime where the sample size and ambient dimension scale proportionally. Our results provide explicit, fully asymptotic predictions for the performance of the second-stage estimator despite the reuse of data and the presence of prediction-dependent losses. We apply this theory to revisit a well-studied pool-based active learning problem, removing oracle and sample-splitting assumptions made in prior work. We uncover a fundamental tradeoff in how the labeling budget should be allocated across stages, and demonstrate a double-descent behavior of the test error driven purely by data selection, rather than model size or sample count.

Inference in non-linear continuous stochastic processes on trees is challenging, particularly when observations are sparse (leaf-only) and the topology is complex. Exact smoothing via Doob's $h$-transform is intractable for general non-linear dynamics, while particle-based methods degrade in high dimensions. We propose Neural Backward Filtering Forward Guiding (NBFFG), a unified framework for both discrete transitions and continuous diffusions. Our method constructs a variational posterior by leveraging an auxiliary linear-Gaussian process. This auxiliary process yields a closed-form backward filter that serves as a ``guide'', steering the generative path toward high-likelihood regions. We then learn a neural residual--parameterized as a normalizing flow or a controlled SDE--to capture the non-linear discrepancies. This formulation allows for an unbiased path-wise subsampling scheme, reducing the training complexity from tree-size dependent to path-length dependent. Empirical results show that NBFFG outperforms baselines on synthetic benchmarks, and we demonstrate the method on a high-dimensional inference task in phylogenetic analysis with reconstruction of ancestral butterfly wing shapes.

Long chain-of-thought reasoning (Long CoT) is now fundamental to state-of-the-art LLMs, especially in mathematical reasoning. However, LLM generation is highly sequential, and long CoTs lead to a high latency. We propose to train Divide-and-Conquer CoT (DC-CoT) to reduce the latency. With DC-CoT, the model can act as a director that identifies distinct subtasks that can be performed in parallel in its reasoning process, and then spawns workers to execute the subtasks. Our goal is to achieve high accuracy, with a low longest path length, which is a theoretical measure of the latency needed for the response. We start with a long CoT base model (DeepScaleR-1.5B-Preview), and first use SFT with a small curated demonstration set to initialize its ability to spawn workers in a certain format. Because SFT degrades the accuracy significantly, we design a multi-stage RL algorithm, with various data filtering strategies, to recover the accuracy while decreasing the longest path length. Across several benchmarks including AIME 2024 and HMMT 2025, DC-CoT achieves similar accuracy as DeepScaleR-1.5B-Preview while decreasing longest path length by 35-40%. Our code, SFT dataset and models are publicly available at https://github.com/amahankali10/DC_CoT_RL_for_Low_Latency_CoT_with_Parallel_Reasoning.

Root cause analysis of anomalies aims to identify those features that cause the deviation from the normal process. Existing methods ignore, however, that anomalies can arise through two fundamentally different processes: measurement errors, where data was generated normally but one or more values were recorded incorrectly, and mechanism shifts, where the causal process generating the data changed. While measurement errors can often be safely corrected, mechanistic anomalies require careful consideration. We define a causal model that explicitly captures both types by treating outliers as latent interventions on latent ("true") and observed ("measured") variables. We show that they are identifiable, and propose a maximum likelihood estimation approach to put this to practice. Experiments show that our method matches state-of-the-art performance in root cause localization, while it additionally enables accurate classification of anomaly types, and remains robust even when the causal DAG is unknown.

Aspect-Based Sentiment Analysis (ABSA) focuses on extracting sentiment at a fine-grained aspect level and has been widely applied across real-world domains. However, existing ABSA research relies on coarse-grained categorical labels (e.g., positive, negative), which limits its ability to capture nuanced affective states. To address this limitation, we adopt a dimensional approach that represents sentiment with continuous valence-arousal (VA) scores, enabling fine-grained analysis at both the aspect and sentiment levels. To this end, we introduce DimABSA, the first multilingual, dimensional ABSA resource annotated with both traditional ABSA elements (aspect terms, aspect categories, and opinion terms) and newly introduced VA scores. This resource contains 76,958 aspect instances across 42,590 sentences, spanning six languages and four domains. We further introduce three subtasks that combine VA scores with different ABSA elements, providing a bridge from traditional ABSA to dimensional ABSA. Given that these subtasks involve both categorical and continuous outputs, we propose a new unified metric, continuous F1 (cF1), which incorporates VA prediction error into standard F1. We provide a comprehensive benchmark using both prompted and fine-tuned large language models across all subtasks. Our results show that DimABSA is a challenging benchmark and provides a foundation for advancing multilingual dimensional ABSA.

Open-source software (OSS) dependencies are a dominant component of modern software code bases. Using proven and well-tested OSS components lets developers reduce development time and cost while improving quality. However, heavy reliance on open-source software also introduces significant security risks, including the incorporation of known vulnerabilities into the codebase. To mitigate these risks, metadata-based dependency scanners, which are lightweight and fast, and code-centric scanners, which enable the detection of modified dependencies hidden from metadata-based approaches, have been developed. In this paper, we present Unshade, a hybrid approach towards dependency scanning in Java that combines the efficiency of metadata-based scanning with the ability to detect modified dependencies of code-centric approaches. Unshade first augments a Java project's software bill of materials (SBOM) by identifying modified and hidden dependencies via a bytecode-based fingerprinting mechanism. This augmented SBOM is then passed to a metadata-based vulnerability scanner to identify known vulnerabilities in both declared and newly revealed dependencies. Leveraging Unshade's high scalability, we conducted a large-scale study of the 1,808 most popular open-source Java Maven projects on GitHub. The results show that nearly 50% of these projects contain at least one modified, hidden dependency associated with a known vulnerability. On average, each affected project includes more than eight such hidden vulnerable dependencies, all missed by traditional metadata-based scanners. Overall, Unshade identified 7,712 unique CVEs in hidden dependencies that would remain undetected when relying on metadata-based scanning alone.

Memory agents, which depart from predefined memory-processing pipelines by endogenously managing the processing, storage, and retrieval of memories, have garnered increasing attention for their autonomy and adaptability. However, existing training paradigms remain constrained: agents often traverse long-horizon sequences of memory operations before receiving sparse and delayed rewards, which hinders truly end-to-end optimization of memory management policies. To address this limitation, we introduce Mem-T, an autonomous memory agent that interfaces with a lightweight hierarchical memory database to perform dynamic updates and multi-turn retrieval over streaming inputs. To effectively train long-horizon memory management capabilities, we further propose MoT-GRPO, a tree-guided reinforcement learning framework that transforms sparse terminal feedback into dense, step-wise supervision via memory operation tree backpropagation and hindsight credit assignment, thereby enabling the joint optimization of memory construction and retrieval. Extensive experiments demonstrate that Mem-T is (1) high-performing, surpassing frameworks such as A-Mem and Mem0 by up to $14.92\%$, and (2) economical, operating on a favorable accuracy-efficiency Pareto frontier and reducing inference tokens per query by $\sim24.45\%$ relative to GAM without sacrificing performance.

Reliable control of myoelectric prostheses is often hindered by high inter-subject variability and the clinical impracticality of high-density sensor arrays. This study proposes a deep learning framework for accurate gesture recognition using only two surface electromyography (sEMG) channels. The method employs a Convolutional Sparse Autoencoder (CSAE) to extract temporal feature representations directly from raw signals, eliminating the need for heuristic feature engineering. On a 6-class gesture set, our model achieved a multi-subject F1-score of 94.3% $\pm$ 0.3%. To address subject-specific differences, we present a few-shot transfer learning protocol that improved performance on unseen subjects from a baseline of 35.1% $\pm$ 3.1% to 92.3% $\pm$ 0.9% with minimal calibration data. Furthermore, the system supports functional extensibility through an incremental learning strategy, allowing for expansion to a 10-class set with a 90.0% $\pm$ 0.2% F1-score without full model retraining. By combining high precision with minimal computational and sensor overhead, this framework provides a scalable and efficient approach for the next generation of affordable and adaptive prosthetic systems.

Offline Reinforcement Learning (RL) relies on policy constraints to mitigate extrapolation error, where both the constraint form and constraint strength critically shape performance. However, most existing methods commit to a single constraint family: weighted behavior cloning, density regularization, or support constraints, without a unified principle that explains their connections or trade-offs. In this work, we propose Continuous Constraint Interpolation (CCI), a unified optimization framework in which these three constraint families arise as special cases along a common constraint spectrum. The CCI framework introduces a single interpolation parameter that enables smooth transitions and principled combinations across constraint types. Building on CCI, we develop Automatic Constraint Policy Optimization (ACPO), a practical primal--dual algorithm that adapts the interpolation parameter via a Lagrangian dual update. Moreover, we establish a maximum-entropy performance difference lemma and derive performance lower bounds for both the closed-form optimal policy and its parametric projection. Experiments on D4RL and NeoRL2 demonstrate robust gains across diverse domains, achieving state-of-the-art performance overall.

Smart contracts are the backbone of the decentralized web, yet ensuring their functional correctness and security remains a critical challenge. While Large Language Models (LLMs) have shown promise in code generation, they often struggle with the rigorous requirements of smart contracts, frequently producing code that is buggy or vulnerable. To address this, we propose SolAgent, a novel tool-augmented multi-agent framework that mimics the workflow of human experts. SolAgent integrates a \textbf{dual-loop refinement mechanism}: an inner loop using the \textit{Forge} compiler to ensure functional correctness, and an outer loop leveraging the \textit{Slither} static analyzer to eliminate security vulnerabilities. Additionally, the agent is equipped with file system capabilities to resolve complex project dependencies. Experiments on the SolEval+ Benchmark, a rigorous suite derived from high-quality real-world projects, demonstrate that SolAgent achieves a Pass@1 rate of up to \textbf{64.39\%}, significantly outperforming state-of-the-art LLMs ($\sim$25\%), AI IDEs (e.g., GitHub Copilot), and existing agent frameworks. Moreover, it reduces security vulnerabilities by up to \textbf{39.77\%} compared to human-written baselines. Finally, we demonstrate that the high-quality trajectories generated by SolAgent can be used to distill smaller, open-source models, democratizing access to secure smart contract generation. We release our data and code at https://github.com/openpaperz/SolAgent.

Supervised fine-tuning (SFT) is fundamental to adapting large language models, yet training on complete datasets incurs prohibitive costs with diminishing returns. Existing data selection methods suffer from severe domain specificity: techniques optimized for general instruction-following fail on reasoning tasks, and vice versa. We observe that measuring entropy differences between base models and minimally instruction-tuned calibrated models reveals a pattern -- samples with the lowest differential entropy consistently yield optimal performance across domains, yet this principle manifests domain-adaptively: reasoning tasks favor entropy increase (cognitive expansion), while general tasks favor entropy decrease (cognitive compression). We introduce InstructDiff, a unified framework that operationalizes differential entropy as a domain-adaptive selection criterion through warmup calibration, bi-directional NLL filtering, and entropy-based ranking. Extensive experiments show that InstructDiff achieves 17\% relative improvement over full data training on mathematical reasoning and 52\% for general instruction-following, outperforming prior baselines while using only 10\% of the data.

Large Language Models (LLMs) increasingly shape global discourse, making fairness and ideological neutrality essential for responsible AI deployment. Despite growing attention to political bias in LLMs, prior work largely focuses on high-resource, Western languages or narrow multilingual settings, leaving cross-lingual consistency and safe post-hoc mitigation underexplored. To address this gap, we present a large-scale multilingual evaluation of political bias spanning 50 countries and 33 languages. We introduce a complementary post-hoc mitigation framework, Cross-Lingual Alignment Steering (CLAS), designed to augment existing steering methods by aligning ideological representations across languages and dynamically regulating intervention strength. This method aligns latent ideological representations induced by political prompts into a shared ideological subspace, ensuring cross lingual consistency, with the adaptive mechanism prevents over correction and preserves coherence. Experiments demonstrate substantial bias reduction along both economic and social axes with minimal degradation in response quality. The proposed framework establishes a scalable and interpretable paradigm for fairness-aware multilingual LLM governance, balancing ideological neutrality with linguistic and cultural diversity.

While large language models (LLMs) have emerged as a significant advancement in artificial intelligence, the hardware and computational costs for training LLMs are also significantly burdensome. Among the state-of-the-art optimizers, AdamW relies on diagonal curvature estimates and ignores structural properties, while Muon applies global spectral normalization at the expense of losing curvature information. In this study, we restriked manifold optimization methods for training LLMs, which may address both optimizers' limitations, while conventional manifold optimization methods have been largely overlooked due to the poor performance in large-scale model optimization. By innovatively projecting the momentum onto the tangent space of model parameters and constraining it on a rotational Oblique manifold, we propose a novel, powerful, and efficient optimizer **Mano** that is the first to bridge the performance gap between manifold optimization and modern optimizers. Extensive experiments on the LLaMA and Qwen3 models demonstrate that Mano consistently and significantly outperforms AdamW and Muon even with less memory consumption and computational complexity, respectively, suggesting an expanded Pareto frontier in terms of space and time efficiency.

Agentic AI systems act through tools and evolve their behavior over long, stochastic interaction traces. This setting complicates assurance, because behavior depends on nondeterministic environments and probabilistic model outputs. Prior work introduced runtime verification for agentic AI via Dynamic Probabilistic Assurance (DPA), learning an MDP online and model checking quantitative properties. A key limitation is that developers must manually define the state abstraction, which couples verification to application-specific heuristics and increases adoption friction. This paper proposes TriCEGAR, a trace-driven abstraction mechanism that automates state construction from execution logs and supports online construction of an agent behavioral MDP. TriCEGAR represents abstractions as predicate trees learned from traces and refined using counterexamples. We describe a framework-native implementation that (i) captures typed agent lifecycle events, (ii) builds abstractions from traces, (iii) constructs an MDP, and (iv) performs probabilistic model checking to compute bounds such as Pmax(success) and Pmin(failure). We also show how run likelihoods enable anomaly detection as a guardrailing signal.

We introduce the Value-at-Risk Constrained Policy Optimization algorithm (VaR-CPO), a sample efficient and conservative method designed to optimize Value-at-Risk (VaR) constraints directly. Empirically, we demonstrate that VaR-CPO is capable of safe exploration, achieving zero constraint violations during training in feasible environments, a critical property that baseline methods fail to uphold. To overcome the inherent non-differentiability of the VaR constraint, we employ the one-sided Chebyshev inequality to obtain a tractable surrogate based on the first two moments of the cost return. Additionally, by extending the trust-region framework of the Constrained Policy Optimization (CPO) method, we provide rigorous worst-case bounds for both policy improvement and constraint violation during the training process.

Reconstructing 3D fetal MR volumes from motion-corrupted stacks of 2D slices is a crucial and challenging task. Conventional slice-to-volume reconstruction (SVR) methods are time-consuming and require multiple orthogonal stacks for reconstruction. While learning-based SVR approaches have significantly reduced the time required at the inference stage, they heavily rely on ground truth information for training, which is inaccessible in practice. To address these challenges, we propose GaussianSVR, a self-supervised framework for slice-to-volume reconstruction. GaussianSVR represents the target volume using 3D Gaussian representations to achieve high-fidelity reconstruction. It leverages a simulated forward slice acquisition model to enable self-supervised training, alleviating the need for ground-truth volumes. Furthermore, to enhance both accuracy and efficiency, we introduce a multi-resolution training strategy that jointly optimizes Gaussian parameters and spatial transformations across different resolution levels. Experiments show that GaussianSVR outperforms the baseline methods on fetal MR volumetric reconstruction. Code will be available upon acceptance.

We present ArabicDialectHub, a cross-dialectal Arabic learning resource comprising 552 phrases across six varieties (Moroccan Darija, Lebanese, Syrian, Emirati, Saudi, and MSA) and an interactive web platform. Phrases were generated using LLMs and validated by five native speakers, stratified by difficulty, and organized thematically. The open-source platform provides translation exploration, adaptive quizzing with algorithmic distractor generation, cloud-synchronized progress tracking, and cultural context. Both the dataset and complete platform source code are released under MIT license. Platform: https://arabic-dialect-hub.netlify.app.

We propose dgMARK, a decoding-guided watermarking method for discrete diffusion language models (dLLMs). Unlike autoregressive models, dLLMs can generate tokens in arbitrary order. While an ideal conditional predictor would be invariant to this order, practical dLLMs exhibit strong sensitivity to the unmasking order, creating a new channel for watermarking. dgMARK steers the unmasking order toward positions whose high-reward candidate tokens satisfy a simple parity constraint induced by a binary hash, without explicitly reweighting the model's learned probabilities. The method is plug-and-play with common decoding strategies (e.g., confidence, entropy, and margin-based ordering) and can be strengthened with a one-step lookahead variant. Watermarks are detected via elevated parity-matching statistics, and a sliding-window detector ensures robustness under post-editing operations including insertion, deletion, substitution, and paraphrasing.

Diagnosing the failure mechanisms of Deep Research Agents (DRAs) remains a critical challenge. Existing benchmarks predominantly rely on end-to-end evaluation, obscuring critical intermediate hallucinations, such as flawed planning, that accumulate throughout the research trajectory. To bridge this gap, we propose a shift from outcome-based to process-aware evaluation by auditing the full research trajectory. We introduce the PIES Taxonomy to categorize hallucinations along functional components (Planning vs. Summarization) and error properties (Explicit vs. Implicit). We instantiate this taxonomy into a fine-grained evaluation framework that decomposes the trajectory to rigorously quantify these hallucinations. Leveraging this framework to isolate 100 distinctively hallucination-prone tasks including adversarial scenarios, we curate DeepHalluBench. Experiments on six state-of-theart DRAs reveal that no system achieves robust reliability. Furthermore, our diagnostic analysis traces the etiology of these failures to systemic deficits, specifically hallucination propagation and cognitive biases, providing foundational insights to guide future architectural optimization. Data and code are available at https://github.com/yuhao-zhan/DeepHalluBench.

Recent provenance-based intrusion detection systems (PIDSs) have demonstrated strong potential for detecting advanced persistent threats (APTs) by applying machine learning to system provenance graphs. However, evaluating and comparing PIDSs remains difficult: prior work uses inconsistent preprocessing pipelines, non-standard dataset splits, and incompatible ground-truth labeling and metrics. These discrepancies undermine reproducibility, impede fair comparison, and impose substantial re-implementation overhead on researchers. We present PIDSMaker, an open-source framework for developing and evaluating PIDSs under consistent protocols. PIDSMaker consolidates eight state-of-the-art systems into a modular, extensible architecture with standardized preprocessing and ground-truth labels, enabling consistent experiments and apples-to-apples comparisons. A YAML-based configuration interface supports rapid prototyping by composing components across systems without code changes. PIDSMaker also includes utilities for ablation studies, hyperparameter tuning, multi-run instability measurement, and visualization, addressing methodological gaps identified in prior work. We demonstrate PIDSMaker through concrete use cases and release it with preprocessed datasets and labels to support shared evaluation for the PIDS community.

Factory automation has become increasingly important due to labor shortages, leading to the introduction of autonomous mobile robots for tasks such as material transportation. Markers are commonly used for robot self-localization and object identification. In the RoboCup Logistics League (RCLL), ArUco markers are employed both for robot localization and for identifying processing modules. Conventional recognition relies on OpenCV-based image processing, which detects black-and-white marker patterns. However, these methods often fail under noise, motion blur, defocus, or varying illumination conditions. Deep-learning-based recognition offers improved robustness under such conditions, but requires large amounts of annotated data. Annotation must typically be done manually, as the type and position of objects cannot be detected automatically, making dataset preparation a major bottleneck. In contrast, ArUco markers include built-in recognition modules that provide both ID and positional information, enabling automatic annotation. This paper proposes an automated annotation method for training deep-learning models on ArUco marker images. By leveraging marker detection results obtained from the ArUco module, the proposed approach eliminates the need for manual labeling. A YOLO-based model is trained using the automatically annotated dataset, and its performance is evaluated under various conditions. Experimental results demonstrate that the proposed method improves recognition performance compared with conventional image-processing techniques, particularly for images affected by blur or defocus. Automatic annotation also reduces human effort and ensures consistent labeling quality. Future work will investigate the relationship between confidence thresholds and recognition performance.

Structured sparsity has emerged as a popular model pruning technique, widely adopted in various architectures, including CNNs, Transformer models, and especially large language models (LLMs) in recent years. A promising direction to further improve post-pruning performance is weight permutation, which reorders model weights into patterns more amenable to pruning. However, the exponential growth of the permutation search space with the scale of Transformer architectures forces most methods to rely on greedy or heuristic algorithms, limiting the effectiveness of reordering. In this work, we propose a novel end-to-end learnable permutation framework. Our method introduces a learnable permutation cost matrix to quantify the cost of swapping any two input channels of a given weight matrix, a differentiable bipartite matching solver to obtain the optimal binary permutation matrix given a cost matrix, and a sparsity optimization loss function to directly optimize the permutation operator. We extensively validate our approach on vision and language Transformers, demonstrating that our method achieves state-of-the-art permutation results for structured sparsity.

As AI systems evolve from isolated predictors into complex, heterogeneous ecosystems of foundation models and specialized adapters, distinguishing genuine behavioral novelty from functional redundancy becomes a critical governance challenge. Here, we introduce a statistical framework for auditing model uniqueness based on In-Silico Quasi-Experimental Design (ISQED). By enforcing matched interventions across models, we isolate intrinsic model identity and quantify uniqueness as the Peer-Inexpressible Residual (PIER), i.e. the component of a target's behavior strictly irreducible to any stochastic convex combination of its peers, with vanishing PIER characterizing when such a routing-based substitution becomes possible. We establish the theoretical foundations of ecosystem auditing through three key contributions. First, we prove a fundamental limitation of observational logs: uniqueness is mathematically non-identifiable without intervention control. Second, we derive a scaling law for active auditing, showing that our adaptive query protocol achieves minimax-optimal sample efficiency ($dσ^2γ^{-2}\log(Nd/δ)$). Third, we demonstrate that cooperative game-theoretic methods, such as Shapley values, fundamentally fail to detect redundancy. We implement this framework via the DISCO (Design-Integrated Synthetic Control) estimator and deploy it across diverse ecosystems, including computer vision models (ResNet/ConvNeXt/ViT), large language models (BERT/RoBERTa), and city-scale traffic forecasters. These results move trustworthy AI beyond explaining single models: they establish a principled, intervention-based science of auditing and governing heterogeneous model ecosystems.

Reinforcement Learning with Verifiable Rewards (RLVR) has become a cornerstone for unlocking complex reasoning in Large Language Models (LLMs). Yet, scaling up RL is bottlenecked by limited existing verifiable data, where improvements increasingly saturate over prolonged training. To overcome this, we propose Golden Goose, a simple trick to synthesize unlimited RLVR tasks from unverifiable internet text by constructing a multiple-choice question-answering version of the fill-in-the-middle task. Given a source text, we prompt an LLM to identify and mask key reasoning steps, then generate a set of diverse, plausible distractors. This enables us to leverage reasoning-rich unverifiable corpora typically excluded from prior RLVR data construction (e.g., science textbooks) to synthesize GooseReason-0.7M, a large-scale RLVR dataset with over 0.7 million tasks spanning mathematics, programming, and general scientific domains. Empirically, GooseReason effectively revives models saturated on existing RLVR data, yielding robust, sustained gains under continuous RL and achieving new state-of-the-art results for 1.5B and 4B-Instruct models across 15 diverse benchmarks. Finally, we deploy Golden Goose in a real-world setting, synthesizing RLVR tasks from raw FineWeb scrapes for the cybersecurity domain, where no prior RLVR data exists. Training Qwen3-4B-Instruct on the resulting data GooseReason-Cyber sets a new state-of-the-art in cybersecurity, surpassing a 7B domain-specialized model with extensive domain-specific pre-training and post-training. This highlights the potential of automatically scaling up RLVR data by exploiting abundant, reasoning-rich, unverifiable internet text.

Recent advances in large language models (LLMs) have substantially accelerated the development of embodied agents. LLM-based multi-agent systems mitigate the inefficiency of single agents in complex tasks. However, they still suffer from issues such as memory inconsistency and agent behavioral conflicts. To address these challenges, we propose MiTa, a hierarchical memory-integrated task allocative framework to enhance collaborative efficiency. MiTa organizes agents into a manager-member hierarchy, where the manager incorporates additional allocation and summary modules that enable (1) global task allocation and (2) episodic memory integration. The allocation module enables the manager to allocate tasks from a global perspective, thereby avoiding potential inter-agent conflicts. The summary module, triggered by task progress updates, performs episodic memory integration by condensing recent collaboration history into a concise summary that preserves long-horizon context. By combining task allocation with episodic memory, MiTa attains a clearer understanding of the task and facilitates globally consistent task distribution. Experimental results confirm that MiTa achieves superior efficiency and adaptability in complex multi-agent cooperation over strong baseline methods.

Policies learned via continuous actor-critic methods often exhibit erratic, high-frequency oscillations, making them unsuitable for physical deployment. Current approaches attempt to enforce smoothness by directly regularizing the policy's output. We argue that this approach treats the symptom rather than the cause. In this work, we theoretically establish that policy non-smoothness is fundamentally governed by the differential geometry of the critic. By applying implicit differentiation to the actor-critic objective, we prove that the sensitivity of the optimal policy is bounded by the ratio of the Q-function's mixed-partial derivative (noise sensitivity) to its action-space curvature (signal distinctness). To empirically validate this theoretical insight, we introduce PAVE (Policy-Aware Value-field Equalization), a critic-centric regularization framework that treats the critic as a scalar field and stabilizes its induced action-gradient field. PAVE rectifies the learning signal by minimizing the Q-gradient volatility while preserving local curvature. Experimental results demonstrate that PAVE achieves smoothness and robustness comparable to policy-side smoothness regularization methods, while maintaining competitive task performance, without modifying the actor.

Nash regret has recently emerged as a principled fairness-aware performance metric for stochastic multi-armed bandits, motivated by the Nash Social Welfare objective. Although this notion has been extended to linear bandits, existing results suffer from suboptimality in ambient dimension $d$, stemming from proof techniques that rely on restrictive concentration inequalities. In this work, we resolve this open problem by introducing new analytical tools that yield an order-optimal Nash regret bound in linear bandits. Beyond Nash regret, we initiate the study of $p$-means regret in linear bandits, a unifying framework that interpolates between fairness and utility objectives and strictly generalizes Nash regret. We propose a generic algorithmic framework, FairLinBandit, that works as a meta-algorithm on top of any linear bandit strategy. We instantiate this framework using two bandit algorithms: Phased Elimination and Upper Confidence Bound, and prove that both achieve sublinear $p$-means regret for the entire range of $p$. Extensive experiments on linear bandit instances generated from real-world datasets demonstrate that our methods consistently outperform the existing state-of-the-art baseline.

We investigate the functional role of emergent outliers in large language models, specifically attention sinks (a few tokens that consistently receive large attention logits) and residual sinks (a few fixed dimensions with persistently large activations across most tokens). We hypothesize that these outliers, in conjunction with the corresponding normalizations (\textit{e.g.}, softmax attention and RMSNorm), effectively rescale other non-outlier components. We term this phenomenon \textit{outlier-driven rescaling} and validate this hypothesis across different model architectures and training token counts. This view unifies the origin and mitigation of both sink types. Our main conclusions and observations include: (1) Outliers function jointly with normalization: removing normalization eliminates the corresponding outliers but degrades training stability and performance; directly clipping outliers while retaining normalization leads to degradation, indicating that outlier-driven rescaling contributes to training stability. (2) Outliers serve more as rescale factors rather than contributors, as the final contributions of attention and residual sinks are significantly smaller than those of non-outliers. (3) Outliers can be absorbed into learnable parameters or mitigated via explicit gated rescaling, leading to improved training performance (average gain of 2 points) and enhanced quantization robustness (1.2 points degradation under W4A4 quantization).

Prevailing medical AI operates on an unrealistic ''one-shot'' model, diagnosing from a complete patient file. However, real-world diagnosis is an iterative inquiry where Clinicians sequentially ask questions and order tests to strategically gather information while managing cost and time. To address this, we first propose Med-Inquire, a new benchmark designed to evaluate an agent's ability to perform multi-turn diagnosis. Built upon a dataset of real-world clinical cases, Med-Inquire simulates the diagnostic process by hiding a complete patient file behind specialized Patient and Examination agents. They force the agent to proactively ask questions and order tests to gather information piece by piece. To tackle the challenges posed by Med-Inquire, we then introduce EvoClinician, a self-evolving agent that learns efficient diagnostic strategies at test time. Its core is a ''Diagnose-Grade-Evolve'' loop: an Actor agent attempts a diagnosis; a Process Grader agent performs credit assignment by evaluating each action for both clinical yield and resource efficiency; finally, an Evolver agent uses this feedback to update the Actor's strategy by evolving its prompt and memory. Our experiments show EvoClinician outperforms continual learning baselines and other self-evolving agents like memory agents. The code is available at https://github.com/yf-he/EvoClinician

Conflict-Free Replicated Data Types (CRDTs) are used in a range of fields for their coordination-free replication with strong eventual consistency. By prioritising availability over consistency under partition, nodes accumulate events in different orders, and rely on an associative, commutative and idempotent merge function to present a materialised view of the CRDT. Under some circumstances, the state of the materialised view over time can appear to ''roll back'' previously applied events. When the materialised view is used to manage group permissions such as ones found in instant messaging applications, this can lead to surprising behaviour. This can occur when there are multiple concurrent events, such as in the Duelling Admins problem where two equally permissioned admins concurrently revoke each other's permissions. Who wins? This article argues that a Byzantine admin can exploit concurrency to win the duel. As a result, an external arbiter is required to arbitrate an immutable happens-before relation between concurrent events. Arbitration occurs asynchronously in batches via optional ''epoch events'', preserving availability. This introduces a bounded total order within epochs, and the resulting ''finality'' improves on the level of consistency CRDTs can provide.

Large language model (LLM) research in software engineering has largely focused on tasks such as code generation and bug repair. In practice, teams often draft multiple candidate proposals for fixing an issue and then deliberate on one golden proposal for implementation. This selection requires not only assessing the issue's scope, impact, and urgency, but also a clear understanding of each proposal's strengths and weaknesses. A good selection could make issue resolution more reliable while reducing regression and operational risk, whereas a poor choice can increase risk and even cause unpredictable failures. We first conduct a manual study of real-world issues to characterize the rationales maintainers use when selecting among competing proposals. Motivated by these findings, we introduce SWE-Manager, a joint selection and synthesis approach that selects the best proposal and synthesizes a golden proposal. SWE-Manager is an 8B model trained via reinforcement learning (RL) to compare proposals, justify its choice, and synthesize a golden proposal for implementation. We view proposal selection as a reasoning task, mirroring how technical managers review competing proposals by weighing issue context and each proposal's solution without executing code or running tests. On the SWE-Lancer Manager benchmark, SWE-Manager achieves 53.21 selection accuracy and 57.75 earn rate, earning 152,750 dollars and outperforming strong baselines including GPT-5. To further evaluate the effectiveness of SWE-Manager in real-world issue resolution, we design the P2A framework, which simulates a real-world workflow where multiple proposals are drafted, reviewed, and a golden proposal is selected for implementation ...

Diffusion Large Language Models (dLLMs) have emerged as a promising alternative to purely autoregressive language models because they can decode multiple tokens in parallel. However, state-of-the-art block-wise dLLMs rely on a "remasking" mechanism that decodes only the most confident tokens and discards the rest, effectively wasting computation. We demonstrate that recycling computation from the discarded tokens is beneficial, as these tokens retain contextual information useful for subsequent decoding iterations. In light of this, we propose Residual Context Diffusion (RCD), a module that converts these discarded token representations into contextual residuals and injects them back for the next denoising step. RCD uses a decoupled two-stage training pipeline to bypass the memory bottlenecks associated with backpropagation. We validate our method on both long CoT reasoning (SDAR) and short CoT instruction following (LLaDA) models. We demonstrate that a standard dLLM can be efficiently converted to the RCD paradigm with merely ~1 billion tokens. RCD consistently improves frontier dLLMs by 5-10 points in accuracy with minimal extra computation overhead across a wide range of benchmarks. Notably, on the most challenging AIME tasks, RCD nearly doubles baseline accuracy and attains up to 4-5x fewer denoising steps at equivalent accuracy levels.

Static Application Security Testing (SAST) tools are essential for identifying software vulnerabilities, but they often produce a high volume of false positives (FPs), imposing a substantial manual triage burden on developers. Recent advances in Large Language Model (LLM) agents offer a promising direction by enabling iterative reasoning, tool use, and environment interaction to refine SAST alerts. However, the comparative effectiveness of different LLM-based agent architectures for FP filtering remains poorly understood. In this paper, we present a comparative study of three state-of-the-art LLM-based agent frameworks, i.e., Aider, OpenHands, and SWE-agent, for vulnerability FP filtering. We evaluate these frameworks using the vulnerabilities from the OWASP Benchmark and real-world open-source Java projects. The experimental results show that LLM-based agents can remove the majority of SAST noise, reducing an initial FP detection rate of over 92% on the OWASP Benchmark to as low as 6.3% in the best configuration. On real-world dataset, the best configuration of LLM-based agents can achieve an FP identification rate of up to 93.3% involving CodeQL alerts. However, the benefits of agents are strongly backbone- and CWE-dependent: agentic frameworks significantly outperform vanilla prompting for stronger models such as Claude Sonnet 4 and GPT-5, but yield limited or inconsistent gains for weaker backbones. Moreover, aggressive FP reduction can come at the cost of suppressing true vulnerabilities, highlighting important trade-offs. Finally, we observe large disparities in computational cost across agent frameworks. Overall, our study demonstrates that LLM-based agents are a powerful but non-uniform solution for SAST FP filtering, and that their practical deployment requires careful consideration of agent design, backbone model choice, vulnerability category, and operational cost.

We introduce \textit{OneFlowSBI}, a unified framework for simulation-based inference that learns a single flow-matching generative model over the joint distribution of parameters and observations. Leveraging a query-aware masking distribution during training, the same model supports multiple inference tasks, including posterior sampling, likelihood estimation, and arbitrary conditional distributions, without task-specific retraining. We evaluate \textit{OneFlowSBI} on ten benchmark inference problems and two high-dimensional real-world inverse problems across multiple simulation budgets. \textit{OneFlowSBI} is shown to deliver competitive performance against state-of-the-art generalized inference solvers and specialized posterior estimators, while enabling efficient sampling with few ODE integration steps and remaining robust under noisy and partially observed data.

Perplexity -- a function measuring a model's overall level of "surprise" when encountering a particular output -- has gained significant traction in recent years, both as a loss function and as a simple-to-compute metric of model quality. Prior studies have pointed out several limitations of perplexity, often from an empirical manner. Here we leverage recent results on Transformer continuity to show in a rigorous manner how perplexity may be an unsuitable metric for model selection. Specifically, we prove that, if there is any sequence that a compact decoder-only Transformer model predicts accurately and confidently -- a necessary pre-requisite for strong generalisation -- it must imply existence of another sequence with very low perplexity, but not predicted correctly by that same model. Further, by analytically studying iso-perplexity plots, we find that perplexity will not always select for the more accurate model -- rather, any increase in model confidence must be accompanied by a commensurate rise in accuracy for the new model to be selected.

Graph-based fraud detection on text-attributed graphs (TAGs) requires jointly modeling rich textual semantics and relational dependencies. However, existing LLM-enhanced GNN approaches are constrained by predefined prompting and decoupled training pipelines, limiting reasoning autonomy and weakening semantic-structural alignment. We propose FraudCoT, a unified framework that advances TAG-based fraud detection through autonomous, graph-aware chain-of-thought (CoT) reasoning and scalable LLM-GNN co-training. To address the limitations of predefined prompts, we introduce a fraud-aware selective CoT distillation mechanism that generates diverse reasoning paths and enhances semantic-structural understanding. These distilled CoTs are integrated into node texts, providing GNNs with enriched, multi-hop semantic and structural cues for fraud detection. Furthermore, we develop an efficient asymmetric co-training strategy that enables end-to-end optimization while significantly reducing the computational cost of naive joint training. Extensive experiments on public and industrial benchmarks demonstrate that FraudCoT achieves up to 8.8% AUPRC improvement over state-of-the-art methods and delivers up to 1,066x speedup in training throughput, substantially advancing both detection performance and efficiency.

A fundamental question in cognitive science and AI concerns whether different learning modalities: language, vision, and action, give rise to distinct or shared internal representations. Traditional views assume that models trained on different data types develop specialized, non-transferable representations. However, recent evidence suggests unexpected convergence: models optimized for distinct tasks may develop similar representational geometries. We investigate whether this convergence extends to embodied action learning by training a transformer-based agent to execute goal-directed behaviors in response to natural language instructions. Using behavioral cloning on the BabyAI platform, we generated action-grounded language embeddings shaped exclusively by sensorimotor control requirements. We then compared these representations with those extracted from state-of-the-art large language models (LLaMA, Qwen, DeepSeek, BERT) and vision-language models (CLIP, BLIP). Despite substantial differences in training data, modality, and objectives, we observed robust cross-modal alignment. Action representations aligned strongly with decoder-only language models and BLIP (precision@15: 0.70-0.73), approaching the alignment observed among language models themselves. Alignment with CLIP and BERT was significantly weaker. These findings indicate that linguistic, visual, and action representations converge toward partially shared semantic structures, supporting modality-independent semantic organization and highlighting potential for cross-domain transfer in embodied AI systems.

Masked diffusion language models (MDLMs) have emerged as a promising alternative to dominant autoregressive approaches. Although they achieve competitive performance on several tasks, a substantial gap remains in open-ended text generation. We hypothesize that one cause of this gap is that strict positional prediction makes MDLM decoding highly sensitive to token misalignment, and we show through controlled interventions that a one-position shift can severely disrupt semantics. This observation suggests that enforcing strict positional supervision during training is misaligned with the irreversible denoising dynamics of MDLM decoding. Motivated by this mismatch, we adopt an alignment-flexible supervision strategy during fine-tuning. Specifically, we introduce a special token <slack> via the connectionist temporal classification objective. We apply this approach to the widely used MDLM model and conduct experiments on five open-ended text generation benchmarks. Our method consistently outperforms the original model and improves robustness to positional shifts, indicating that relaxing strict positional supervision is an important factor in improving generation quality in MDLMs.

Machine learning models are increasingly used for software security tasks. These models are commonly trained and evaluated on large Internet-derived datasets, which often contain duplicated or highly similar samples. When such samples are split across training and test sets, data leakage may occur, allowing models to memorize patterns instead of learning to generalize. We investigate duplication in a widely used benchmark dataset of hard coded secrets and show how data leakage can substantially inflate the reported performance of AI-based secret detectors, resulting in a misleading picture of their real-world effectiveness.

Out-of-distribution (OOD) generalization remains challenging when models simultaneously encounter correlation shifts across environments and diversity shifts driven by rare or hard samples. Existing invariant risk minimization (IRM) methods primarily address spurious correlations at the environment level, but often overlook sample-level heterogeneity within environments, which can critically impact OOD performance. In this work, we propose \emph{Environment-Conditioned Tail Reweighting for Total Variation Invariant Risk Minimization} (ECTR), a unified framework that augments TV-based invariant learning with environment-conditioned tail reweighting to jointly address both types of distribution shift. By integrating environment-level invariance with within-environment robustness, the proposed approach makes these two mechanisms complementary under mixed distribution shifts. We further extend the framework to scenarios without explicit environment annotations by inferring latent environments through a minimax formulation. Experiments across regression, tabular, time-series, and image classification benchmarks under mixed distribution shifts demonstrate consistent improvements in both worst-environment and average OOD performance.

Graph coarsening reduces the size of a graph while preserving certain properties. Most existing methods preserve either spectral or spatial characteristics. Recent research has shown that preserving topological features helps maintain the predictive performance of graph neural networks (GNNs) trained on the coarsened graph but suffers from exponential time complexity. To address these problems, we propose Scalable Topology-Preserving Graph Coarsening (STPGC) by introducing the concepts of graph strong collapse and graph edge collapse extended from algebraic topology. STPGC comprises three new algorithms, GStrongCollapse, GEdgeCollapse, and NeighborhoodConing based on these two concepts, which eliminate dominated nodes and edges while rigorously preserving topological features. We further prove that STPGC preserves the GNN receptive field and develop approximate algorithms to accelerate GNN training. Experiments on node classification with GNNs demonstrate the efficiency and effectiveness of STPGC.

As intelligent sensing expands into high-privacy environments such as restrooms and changing rooms, the field faces a critical privacy-security paradox. Traditional RGB surveillance raises significant concerns regarding visual recording and storage, while existing privacy-preserving methods-ranging from physical desensitization to traditional cryptographic or obfuscation techniques-often compromise semantic understanding capabilities or fail to guarantee mathematical irreversibility against reconstruction attacks. To address these challenges, this study presents a novel privacy-preserving perception technology based on the AI Flow theoretical framework and an edge-cloud collaborative architecture. The proposed methodology integrates source desensitization with irreversible feature mapping. Leveraging Information Bottleneck theory, the edge device performs millisecond-level processing to transform raw imagery into abstract feature vectors via non-linear mapping and stochastic noise injection. This process constructs a unidirectional information flow that strips identity-sensitive attributes, rendering the reconstruction of original images impossible. Subsequently, the cloud platform utilizes multimodal family models to perform joint inference solely on these abstract vectors to detect abnormal behaviors. This approach fundamentally severs the path to privacy leakage at the architectural level, achieving a breakthrough from video surveillance to de-identified behavior perception and offering a robust solution for risk management in high-sensitivity public spaces.

Modern Integrated Development Environments (IDEs) increasingly leverage Large Language Models (LLMs) to provide advanced features like code autocomplete. While powerful, training these models on user-written code introduces significant privacy risks, making the models themselves a new type of data vulnerability. Malicious actors can exploit this by launching attacks to reconstruct sensitive training data or infer whether a specific code snippet was used for training. This paper investigates the use of Differential Privacy (DP) as a robust defense mechanism for training an LLM for Kotlin code completion. We fine-tune a \texttt{Mellum} model using DP and conduct a comprehensive evaluation of its privacy and utility. Our results demonstrate that DP provides a strong defense against Membership Inference Attacks (MIAs), reducing the attack's success rate close to a random guess (AUC from 0.901 to 0.606). Furthermore, we show that this privacy guarantee comes at a minimal cost to model performance, with the DP-trained model achieving utility scores comparable to its non-private counterpart, even when trained on 100x less data. Our findings suggest that DP is a practical and effective solution for building private and trustworthy AI-powered IDE features.

We study a sampling problem whose target distribution is $π\propto \exp(-f-r)$ where the data fidelity term $f$ is Lipschitz smooth while the regularizer term $r=r_1-r_2$ is a non-smooth difference-of-convex (DC) function, i.e., $r_1,r_2$ are convex. By leveraging the DC structure of $r$, we can smooth out $r$ by applying Moreau envelopes to $r_1$ and $r_2$ separately. In line of DC programming, we then redistribute the concave part of the regularizer to the data fidelity and study its corresponding proximal Langevin algorithm (termed DC-LA). We establish convergence of DC-LA to the target distribution $π$, up to discretization and smoothing errors, in the $q$-Wasserstein distance for all $q \in \mathbb{N}^*$, under the assumption that $V$ is distant dissipative. Our results improve previous work on non-log-concave sampling in terms of a more general framework and assumptions. Numerical experiments show that DC-LA produces accurate distributions in synthetic settings and reliably provides uncertainty quantification in a real-world Computed Tomography application.

The prevalence of rapidly evolving slang, neologisms, and highly stylized expressions in informal user-generated text, particularly on Chinese social media, poses significant challenges for Machine Translation (MT) benchmarking. Specifically, we identify two primary obstacles: (1) data scarcity, as high-quality parallel data requires bilingual annotators familiar with platform-specific slang, and stylistic cues in both languages; and (2) metric limitations, where traditional evaluators like COMET often fail to capture stylistic fidelity and nonstandard expressions. To bridge these gaps, we introduce CSM-MTBench, a benchmark covering five Chinese-foreign language directions and consisting of two expert-curated subsets: Fun Posts, featuring context-rich, slang- and neologism-heavy content, and Social Snippets, emphasizing concise, emotion- and style- driven expressions. Furthermore, we propose tailored evaluation approaches for each subset: measuring the translation success rate of slang and neologisms in Fun Posts, while assessing tone and style preservation in Social Snippets via a hybrid of embedding-based metrics and LLM-as-a-judge. Experiments on over 20 models reveal substantial variation in how current MT systems handle semantic fidelity and informal, social-media-specific stylistic cues. CSM-MTBench thus serves as a rigorous testbed for advancing MT systems capable of mastering real-world Chinese social media texts.

Trajectory planning is a core task in autonomous driving, requiring the prediction of safe and comfortable paths across diverse scenarios. Integrating Multi-modal Large Language Models (MLLMs) with Reinforcement Learning (RL) has shown promise in addressing "long-tail" scenarios. However, existing methods are constrained to single-turn reasoning, limiting their ability to handle complex tasks requiring iterative refinement. To overcome this limitation, we present MTDrive, a multi-turn framework that enables MLLMs to iteratively refine trajectories based on environmental feedback. MTDrive introduces Multi-Turn Group Relative Policy Optimization (mtGRPO), which mitigates reward sparsity by computing relative advantages across turns. We further construct an interactive trajectory understanding dataset from closed-loop simulation to support multi-turn training. Experiments on the NAVSIM benchmark demonstrate superior performance compared to existing methods, validating the effectiveness of our multi-turn reasoning paradigm. Additionally, we implement system-level optimizations to reduce data transfer overhead caused by high-resolution images and multi-turn sequences, achieving 2.5x training throughput. Our data, models, and code will be made available soon.

Image embeddings are generally assumed to pose limited privacy risk. We challenge this assumption by formalizing semantic leakage as the ability to recover semantic structures from compressed image embeddings. Surprisingly, we show that semantic leakage does not require exact reconstruction of the original image. Preserving local semantic neighborhoods under embedding alignment is sufficient to expose the intrinsic vulnerability of image embeddings. Crucially, this preserved neighborhood structure allows semantic information to propagate through a sequence of lossy mappings. Based on this conjecture, we propose Semantic Leakage from Image Embeddings (SLImE), a lightweight inference framework that reveals semantic information from standalone compressed image embeddings, incorporating a locally trained semantic retriever with off-the-shelf models, without training task-specific decoders. We thoroughly validate each step of the framework empirically, from aligned embeddings to retrieved tags, symbolic representations, and grammatical and coherent descriptions. We evaluate SLImE across a range of open and closed embedding models, including GEMINI, COHERE, NOMIC, and CLIP, and demonstrate consistent recovery of semantic information across diverse inference tasks. Our results reveal a fundamental vulnerability in image embeddings, whereby the preservation of semantic neighborhoods under alignment enables semantic leakage, highlighting challenges for privacy preservation.1

Large Language Models (LLMs) are becoming increasingly popular in pervasive computing due to their versatility and strong performance. However, despite their ubiquitous use, the exact mechanisms underlying their outstanding performance remain unclear. Different methods for LLM explainability exist, and many are, as a method, not fully understood themselves. We started with the question of how linguistic abstraction emerges in LLMs, aiming to detect it across different LLM modules (attention heads and input embeddings). For this, we used methods well-established in the literature: (1) probing for token-level relational structures, and (2) feature-mapping using embeddings as carriers of human-interpretable properties. Both attempts failed for different methodological reasons: Attention-based explanations collapsed once we tested the core assumption that later-layer representations still correspond to tokens. Property-inference methods applied to embeddings also failed because their high predictive scores were driven by methodological artifacts and dataset structure rather than meaningful semantic knowledge. These failures matter because both techniques are widely treated as evidence for what LLMs supposedly understand, yet our results show such conclusions are unwarranted. These limitations are particularly relevant in pervasive and distributed computing settings where LLMs are deployed as system components and interpretability methods are relied upon for debugging, compression, and explaining models.

Recent years have witnessed a rapid surge in research leveraging Large Language Models (LLMs) for recommendation. These methods typically employ supervised fine-tuning (SFT) to adapt LLMs to recommendation scenarios, and utilize beam search during inference to efficiently retrieve $B$ top-ranked recommended items. However, we identify a critical training-inference inconsistency: while SFT optimizes the overall probability of positive items, it does not guarantee that such items will be retrieved by beam search even if they possess high overall probabilities. Due to the greedy pruning mechanism, beam search can prematurely discard a positive item once its prefix probability is insufficient. To address this inconsistency, we propose BEAR (Beam-SEarch-Aware Regularization), a novel fine-tuning objective that explicitly accounts for beam search behavior during training. Rather than directly simulating beam search for each instance during training, which is computationally prohibitive, BEAR enforces a relaxed necessary condition: each token in a positive item must rank within the top-$B$ candidate tokens at each decoding step. This objective effectively mitigates the risk of incorrect pruning while incurring negligible computational overhead compared to standard SFT. Extensive experiments across four real-world datasets demonstrate that BEAR significantly outperforms strong baselines. Code will be released upon acceptance.

Early detection of security bug reports (SBRs) is critical for timely vulnerability mitigation. We present an evaluation of prompt-based engineering and fine-tuning approaches for predicting SBRs using Large Language Models (LLMs). Our findings reveal a distinct trade-off between the two approaches. Prompted proprietary models demonstrate the highest sensitivity to SBRs, achieving a G-measure of 77% and a recall of 74% on average across all the datasets, albeit at the cost of a higher false-positive rate, resulting in an average precision of only 22%. Fine-tuned models, by contrast, exhibit the opposite behavior, attaining a lower overall G-measure of 51% but substantially higher precision of 75% at the cost of reduced recall of 36%. Though a one-time investment in building fine-tuned models is necessary, the inference on the largest dataset is up to 50 times faster than that of proprietary models. These findings suggest that further investigations to harness the power of LLMs for SBR prediction are necessary.

Data is both the key enabler and a major bottleneck for machine learning in autonomous driving. Effective model training requires not only large quantities of sensor data but also balanced coverage that includes rare yet safety-critical scenarios. Capturing such events demands extensive driving time and efficient selection. This paper introduces the Lambda framework, an edge-native platform that enables on-vehicle data filtering and processing through user-defined functions. The framework provides a serverless-inspired abstraction layer that separates application logic from low-level execution concerns such as scheduling, deployment, and isolation. By adapting Function-as-a-Service (FaaS) principles to resource-constrained automotive environments, it allows developers to implement modular, event-driven filtering algorithms while maintaining compatibility with ROS 2 and existing data recording pipelines. We evaluate the framework on an NVIDIA Jetson Orin Nano and compare it against native ROS 2 deployments. Results show competitive performance, reduced latency and jitter, and confirm that lambda-based abstractions can support real-time data processing in embedded autonomous driving systems. The source code is available at https://github.com/LASFAS/jblambda.

The missing data problem is one of the important issues to address for achieving data quality. While imputation-based methods are designed to achieve data completeness, their efficacy is observed to be diminishing as and when there is increasing in the missingness percentage. Further, extant approaches often struggle to handle mixed-type datasets, typically supporting either numerical and/or categorical data. In this work, we propose LLMDR, automatic data recovery framework which operates in two stage approach, wherein the Stage-I: DBSCAN clustering algorithm is employed to select the most representative samples and in the Stage-II: Multi-LLMs are employed for data recovery considering the local and global representative samples; Later, this framework invokes the consensus algorithm for recommending a more accurate value based on other LLMs of local and global effective samples. Experimental results demonstrate that proposed framework works effectively on various mixed datasets in terms of Accuracy, KS-Statistic, SMAPE, and MSE. Further, we have also shown the advantage of the consensus mechanism for final recommendation in mixed-type data.

Large pre-trained models achieve remarkable success across diverse domains, yet fully fine-tuning incurs prohibitive computational and memory costs. Parameter-efficient fine-tuning (PEFT) has thus become a mainstream paradigm. Among them, Low-Rank Adaptation (LoRA) introduces trainable low-rank matrices and shows strong performance, nevertheless, its fixed-rank design limits flexibility. Dynamic rank allocation methods mitigate this issue by pruning redundant directions; however, they often rely on heuristic, element-level metrics that globally sort rank directions without matrix-wise distinction, and they lack mechanisms to expand capacity in layers requiring additional adaptation. To overcome these limitations, we propose FlexLoRA, an entropy-guided flexible low-rank adaptation framework that (i) evaluates matrix importance via spectral energy entropy, (ii) supports rank pruning and expansion under a global budget, and (iii) employs zero-impact initialization for newly added singular directions to ensure stability. By addressing granularity, flexibility, and stability limitations, FlexLoRA provides a more principled solution for PEFT. Extensive experiments show that FlexLoRA consistently outperforms state-of-the-art baselines across benchmarks. Codes are available at https://github.com/Chongjie-Si/Subspace-Tuning.

Recent studies have explored using pretrained Vision Foundation Models (VFMs) such as DINO for generative autoencoders, showing strong generative performance. Unfortunately, existing approaches often suffer from limited reconstruction fidelity due to the loss of high-frequency details. In this work, we present the DINO Spherical Autoencoder (DINO-SAE), a framework that bridges semantic representation and pixel-level reconstruction. Our key insight is that semantic information in contrastive representations is primarily encoded in the direction of feature vectors, while forcing strict magnitude matching can hinder the encoder from preserving fine-grained details. To address this, we introduce Hierarchical Convolutional Patch Embedding module that enhances local structure and texture preservation, and Cosine Similarity Alignment objective that enforces semantic consistency while allowing flexible feature magnitudes for detail retention. Furthermore, leveraging the observation that SSL-based foundation model representations intrinsically lie on a hypersphere, we employ Riemannian Flow Matching to train a Diffusion Transformer (DiT) directly on this spherical latent manifold. Experiments on ImageNet-1K demonstrate that our approach achieves state-of-the-art reconstruction quality, reaching 0.37 rFID and 26.2 dB PSNR, while maintaining strong semantic alignment to the pretrained VFM. Notably, our Riemannian Flow Matching-based DiT exhibits efficient convergence, achieving a gFID of 3.47 at 80 epochs.

Reinforcement Learning with Verifiable Rewards (RLVR) is widely used to improve reasoning in multiple domains, yet outcome-only scalar rewards are often sparse and uninformative, especially on failed samples, where they merely indicate failure and provide no insight into why the reasoning fails. In this paper, we investigate how to leverage richer verbal feedback to guide RLVR training on failed samples, and how to convert such feedback into a trainable learning signal. Specifically, we propose a multi-turn feedback-guided reinforcement learning framework. It builds on three mechanisms: (1) dynamic multi-turn regeneration guided by feedback, triggered only on failed samples, (2) two complementary learning signals for within-turn and cross-turn optimization, and (3) structured feedback injection into the model's reasoning process. Trained on sampled OpenR1-Math, the approach outperforms supervised fine-tuning and RLVR baselines in-domain and generalizes well out-of-domain.

Decision trees are widely used due to their interpretability and efficiency, but they struggle in regression tasks that require reliable extrapolation and well-calibrated uncertainty. Piecewise-constant leaf predictions are bounded by the training targets and often become overconfident under distribution shift. We propose a single-tree Bayesian model that extends VSPYCT by equipping each leaf with a GP predictor. Bayesian oblique splits provide uncertainty-aware partitioning of the input space, while GP leaves model local functional behaviour and enable principled extrapolation beyond the observed target range. We present an efficient inference and prediction scheme that combines posterior sampling of split parameters with \gls{gp} posterior predictions, and a gating mechanism that activates GP-based extrapolation when inputs fall outside the training support of a leaf. Experiments on benchmark regression tasks show improvements in the predictive performance compared to standard variational oblique trees, and substantial performance gains in extrapolation scenarios.

Large language models (LLMs) have enabled rapid progress in automatic heuristic discovery (AHD), yet most existing methods are predominantly limited by static evaluation against fixed instance distributions, leading to potential overfitting and poor generalization under distributional shifts. We propose Algorithm Space Response Oracles (ASRO), a game-theoretic framework that reframes heuristic discovery as a program level co-evolution between solver and instance generator. ASRO models their interaction as a two-player zero-sum game, maintains growing strategy pools on both sides, and iteratively expands them via LLM-based best-response oracles against mixed opponent meta-strategies, thereby replacing static evaluation with an adaptive, self-generated curriculum. Across multiple combinatorial optimization domains, ASRO consistently outperforms static-training AHD baselines built on the same program search mechanisms, achieving substantially improved generalization and robustness on diverse and out-of-distribution instances.

The goal of probabilistic prediction is to issue predictive distributions that are as informative as possible, subject to being calibrated. Despite substantial progress in the univariate setting, achieving multivariate calibration remains challenging. Recent work has introduced pre-rank functions, scalar projections of multivariate forecasts and observations, as flexible diagnostics for assessing specific aspects of multivariate calibration, but their use has largely been limited to post-hoc evaluation. We propose a regularization-based calibration method that enforces multivariate calibration during training of multivariate distributional regression models using pre-rank functions. We further introduce a novel PCA-based pre-rank that projects predictions onto principal directions of the predictive distribution. Through simulation studies and experiments on 18 real-world multi-output regression datasets, we show that the proposed approach substantially improves multivariate pre-rank calibration without compromising predictive accuracy, and that the PCA pre-rank reveals dependence-structure misspecifications that are not detected by existing pre-ranks.

A central challenge in multi-task reinforcement learning (RL) is to train generalist policies capable of performing tasks not seen during training. To facilitate such generalization, linear temporal logic (LTL) has recently emerged as a powerful formalism for specifying structured, temporally extended tasks to RL agents. While existing approaches to LTL-guided multi-task RL demonstrate successful generalization across LTL specifications, they are unable to generalize to unseen vocabularies of propositions (or "symbols"), which describe high-level events in LTL. We present PlatoLTL, a novel approach that enables policies to zero-shot generalize not only compositionally across LTL formula structures, but also parametrically across propositions. We achieve this by treating propositions as instances of parameterized predicates rather than discrete symbols, allowing policies to learn shared structure across related propositions. We propose a novel architecture that embeds and composes predicates to represent LTL specifications, and demonstrate successful zero-shot generalization to novel propositions and tasks across challenging environments.

Current speech language models generate responses directly without explicit reasoning, leading to errors that cannot be corrected once audio is produced. We introduce \textbf{``Silent Thought, Spoken Answer''} -- a paradigm where speech LLMs generate internal text reasoning alongside spoken responses, with thinking traces informing speech quality. To realize this, we present \method{}, the first diffusion-based speech-text language model supporting both understanding and generation, unifying discrete text and tokenized speech under a single masked diffusion framework. Unlike autoregressive approaches, \method{} jointly generates reasoning traces and speech tokens through iterative denoising, with modality-specific masking schedules. We also construct \dataset{}, the first speech QA dataset with paired text reasoning traces, containing 26K samples totaling 319 hours. Experiments show \method{} achieves state-of-the-art speech-to-speech QA accuracy, outperforming the best baseline by up to 9 points, while attaining the best TTS quality among generative models (6.2\% WER) and preserving language understanding (66.2\% MMLU). Ablations confirm that both the diffusion architecture and thinking traces contribute to these gains.

More than 80% of the 1.6 billion English speakers do not use Standard American English (SAE) and experience higher failure rates and stereotyped responses when interacting with LLMs as a result. Yet multi-dialectal performance remains underexplored. We introduce $\textbf{MDial}$, the first large-scale framework for generating multi-dialectal conversational data encompassing the three pillars of written dialect -- lexical (vocabulary), orthographic (spelling), and morphosyntactic (grammar) features -- for nine English dialects. Partnering with native linguists, we design an annotated and scalable rule-based LLM transformation to ensure precision. Our approach challenges the assumption that models should mirror users' morphosyntactic features, showing that up to 90% of the grammatical features of a dialect should not be reproduced by models. Independent evaluations confirm data quality, with annotators preferring MDial outputs over prior methods in 98% of pairwise comparisons for dialect naturalness. Using this pipeline, we construct the dialect-parallel $\textbf{MDialBench}$mark with 50k+ dialogs, resulting in 97k+ QA pairs, and evaluate 17 LLMs on dialect identification and response generation tasks. Even frontier models achieve under 70% accuracy, fail to reach 50% for Canadian English, and systematically misclassify non-SAE dialects as American or British. As dialect identification underpins natural language understanding, these errors risk cascading failures into downstream tasks.

Autoregressive sequence modeling stands as the cornerstone of modern Generative AI, powering results across diverse modalities ranging from text generation to image generation. However, a fundamental limitation of this paradigm is the rigid structural coupling of model capacity to computational cost: expanding a model's parametric memory -- its repository of factual knowledge or visual patterns -- traditionally requires deepening or widening the network, which incurs a proportional rise in active FLOPs. In this work, we introduce $\textbf{MoVE (Mixture of Value Embeddings)}$, a mechanism that breaks this coupling and establishes a new axis for scaling capacity. MoVE decouples memory from compute by introducing a global bank of learnable value embeddings shared across all attention layers. For every step in the sequence, the model employs a differentiable soft gating mechanism to dynamically mix retrieved concepts from this bank into the standard value projection. This architecture allows parametric memory to be scaled independently of network depth by simply increasing the number of embedding slots. We validate MoVE through strictly controlled experiments on two representative applications of autoregressive modeling: Text Generation and Image Generation. In both domains, MoVE yields consistent performance improvements over standard and layer-wise memory baselines, enabling the construction of "memory-dense" models that achieve lower perplexity and higher fidelity than their dense counterparts at comparable compute budgets.

Skill extraction is a critical component of modern recruitment systems, enabling efficient job matching, personalized recommendations, and labor market analysis. Despite Türkiye's significant role in the global workforce, Turkish, a morphologically complex language, lacks both a skill taxonomy and a dedicated skill extraction dataset, resulting in underexplored research in skill extraction for Turkish. This article seeks the answers to three research questions: 1) How can skill extraction be effectively performed for this language, in light of its low resource nature? 2)~What is the most promising model? 3) What is the impact of different Large Language Models (LLMs) and prompting strategies on skill extraction (i.e., dynamic vs. static few-shot samples, varying context information, and encouraging causal reasoning)? The article introduces the first Turkish skill extraction dataset and performance evaluations of automated skill extraction using LLMs. The manually annotated dataset contains 4,819 labeled skill spans from 327 job postings across different occupation areas. The use of LLM outperforms supervised sequence labeling when used in an end-to-end pipeline, aligning extracted spans with standardized skills in the ESCO taxonomy more effectively. The best-performing configuration, utilizing Claude Sonnet 3.7 with dynamic few-shot prompting for skill identification, embedding-based retrieval, and LLM-based reranking for skill linking, achieves an end-to-end performance of 0.56, positioning Turkish alongside similar studies in other languages, which are few in the literature. Our findings suggest that LLMs can improve skill extraction performance in low-resource settings, and we hope that our work will accelerate similar research on skill extraction for underrepresented languages.

Microservice systems (MSS) have become a predominant architectural style for cloud services. Yet the community still lacks high-quality, publicly available datasets for anomaly detection (AD) and root cause analysis (RCA) in MSS. Most benchmarks emphasize performance-related faults and provide only one or two monitoring modalities, limiting research on broader failure modes and cross-modal methods. To address these gaps, we introduce a new multimodal anomaly dataset built on two open-source microservice systems: SocialNetwork and TrainTicket. We design and inject four categories of anomalies (Ano): performance-level, service-level, database-level, and code-level, to emulate realistic anomaly modes. For each scenario, we collect five modalities (Mod): logs, metrics, distributed traces, API responses, and code coverage reports, offering a richer, end-to-end view of system state and inter-service interactions. We name our dataset, reflecting its unique properties, as AnoMod. This dataset enables (1) evaluation of cross-modal anomaly detection and fusion/ablation strategies, and (2) fine-grained RCA studies across service and code regions, supporting end-to-end troubleshooting pipelines that jointly consider detection and localization.

We study the joint operation and sizing of cooling infrastructure for commercial HVAC systems using reinforcement learning, with the objective of minimizing life-cycle cost over a 30-year horizon. The cooling system consists of a fixed-capacity electric chiller and a thermal energy storage (TES) unit, jointly operated to meet stochastic hourly cooling demands under time-varying electricity prices. The life-cycle cost accounts for both capital expenditure and discounted operating cost, including electricity consumption and maintenance. A key challenge arises from the strong asymmetry in capital costs: increasing chiller capacity by one unit is far more expensive than an equivalent increase in TES capacity. As a result, identifying the right combination of chiller and TES sizes, while ensuring zero loss-of-cooling-load under optimal operation, is a non-trivial co-design problem. To address this, we formulate the chiller operation problem for a fixed infrastructure configuration as a finite-horizon Markov Decision Process (MDP), in which the control action is the chiller part-load ratio (PLR). The MDP is solved using a Deep Q Network (DQN) with a constrained action space. The learned DQN RL policy minimizes electricity cost over historical traces of cooling demand and electricity prices. For each candidate chiller-TES sizing configuration, the trained policy is evaluated. We then restrict attention to configurations that fully satisfy the cooling demand and perform a life-cycle cost minimization over this feasible set to identify the cost-optimal infrastructure design. Using this approach, we determine the optimal chiller and thermal energy storage capacities to be 700 and 1500, respectively.

Time series data are essential for a wide range of applications, particularly in developing robust machine learning models. However, access to high-quality datasets is often limited due to privacy concerns, acquisition costs, and labeling challenges. Synthetic time series generation has emerged as a promising solution to address these constraints. In this work, we present a framework for generating synthetic time series by leveraging complex networks mappings. Specifically, we investigate whether time series transformed into Quantile Graphs (QG) -- and then reconstructed via inverse mapping -- can produce synthetic data that preserve the statistical and structural properties of the original. We evaluate the fidelity and utility of the generated data using both simulated and real-world datasets, and compare our approach against state-of-the-art Generative Adversarial Network (GAN) methods. Results indicate that our quantile graph-based methodology offers a competitive and interpretable alternative for synthetic time series generation.

Spiking neural networks (SNNs) have emerged as a promising candidate for energy-efficient LLM inference. However, current energy evaluations for SNNs primarily focus on counting accumulate operations, and fail to account for real-world hardware costs such as data movement, which can consume nearly 80% of the total energy. In this paper, we propose Matterhorn, a spiking transformer that integrates a novel masked time-to-first-spike (M-TTFS) encoding method to reduce spike movement and a memristive synapse unit (MSU) to eliminate weight access overhead. M-TTFS employs a masking strategy that reassigns the zero-energy silent state (a spike train of all 0s) to the most frequent membrane potential rather than the lowest. This aligns the coding scheme with the data distribution, minimizing spike movement energy without information loss. We further propose a `dead zone' strategy that maximizes sparsity by mapping all values within a given range to the silent state. At the hardware level, the MSU utilizes compute-in-memory (CIM) technology to perform analog integration directly within memory, effectively removing weight access costs. On the GLUE benchmark, Matterhorn establishes a new state-of-the-art, surpassing existing SNNs by 1.42% in average accuracy while delivering a 2.31 times improvement in energy efficiency.

In terms of annotation structure, most learner corpora rely on holistic flat label inventories which, even when extensive, do not explicitly separate multiple linguistic dimensions. This makes linguistically deep annotation difficult and complicates fine-grained analyses aimed at understanding why and how learners produce specific errors. To address these limitations, this paper presents a semi-automated annotation methodology for learner corpora, built upon a recently proposed faceted taxonomy, and implemented through a novel annotation extension framework. The taxonomy provides a theoretically grounded, multi-dimensional categorization that captures the linguistic properties underlying each error instance, thereby enabling standardized, fine-grained, and interpretable enrichment beyond flat annotations. The annotation extension tool, implemented based on the proposed extension framework for Turkish, automatically extends existing flat annotations by inferring additional linguistic and metadata information as facets within the taxonomy to provide richer learner-specific context. It was systematically evaluated and yielded promising performance results, achieving a facet-level accuracy of 95.86%. The resulting taxonomically enriched corpus offers enhanced querying capabilities and supports detailed exploratory analyses across learner corpora, enabling researchers to investigate error patterns through complex linguistic and pedagogical dimensions. This work introduces the first collaboratively annotated and taxonomically enriched Turkish Learner Corpus, a manual annotation guideline with a refined tagset, and an annotation extender. As the first corpus designed in accordance with the recently introduced taxonomy, we expect our study to pave the way for subsequent enrichment efforts of existing error-annotated learner corpora.

Achieving precise and controllable emotional expression is crucial for producing natural and context-appropriate speech in text-to-speech (TTS) synthesis. However, many emotion-aware TTS systems, including large language model (LLM)-based designs, rely on scaling fixed emotion embeddings or external guidance, limiting their ability to model emotion-specific latent characteristics. To address this gap, we present EmoShift, a lightweight activation-steering framework incorporating a EmoSteer layer, which learns a steering vector for each target emotion in the output embedding space to capture its latent offset and maintain stable, appropriate expression across utterances and categories. With only 10M trainable parameters,less than 1/30 of full fine-tuning, EmoShift outperforms zero-shot and fully fine-tuned baselines in objective and subjective evaluations, enhancing emotional expressiveness while preserving naturalness and speaker similarity. Further analysis confirms the proposed EmoSteer layer's effectiveness and reveals its potential for controllable emotional intensity in speech synthesis.

As foundation models (FMs) approach human-level fluency, distinguishing synthetic from organic content has become a key challenge for Trustworthy Web Intelligence. This paper presents JudgeGPT and RogueGPT, a dual-axis framework that decouples "authenticity" from "attribution" to investigate the mechanisms of human susceptibility. Analyzing 918 evaluations across five FMs (including GPT-4 and Llama-2), we employ Structural Causal Models (SCMs) as a principal framework for formulating testable causal hypotheses about detection accuracy. Contrary to partisan narratives, we find that political orientation shows a negligible association with detection performance ($r=-0.10$). Instead, "fake news familiarity" emerges as a candidate mediator ($r=0.35$), suggesting that exposure may function as adversarial training for human discriminators. We identify a "fluency trap" where GPT-4 outputs (HumanMachineScore: 0.20) bypass Source Monitoring mechanisms, rendering them indistinguishable from human text. These findings suggest that "pre-bunking" interventions should target cognitive source monitoring rather than demographic segmentation to ensure trustworthy information ecosystems.

Anomaly detection is often formulated under the assumption that abnormality is an intrinsic property of an observation, independent of context. This assumption breaks down in many real-world settings, where the same object or action may be normal or anomalous depending on latent contextual factors (e.g., running on a track versus on a highway). We revisit \emph{contextual anomaly detection}, classically defined as context-dependent abnormality, and operationalize it in the visual domain, where anomaly labels depend on subject--context compatibility rather than intrinsic appearance. To enable systematic study of this setting, we introduce CAAD-3K, a benchmark that isolates contextual anomalies by controlling subject identity while varying context. We further propose a conditional compatibility learning framework that leverages vision--language representations to model subject--context relationships under limited supervision. Our method substantially outperforms existing approaches on CAAD-3K and achieves state-of-the-art performance on MVTec-AD and VisA, demonstrating that modeling context dependence complements traditional structural anomaly detection. Our code and dataset will be publicly released.

Battery energy storage systems are increasingly deployed as fast-responding resources for grid balancing services such as frequency regulation and for mitigating renewable generation uncertainty. However, repeated charging and discharging induces cycling degradation and reduces battery lifetime. This paper studies the real-time scheduling of a heterogeneous battery fleet that collectively tracks a stochastic balancing signal subject to per-battery ramp-rate and capacity constraints, while minimizing long-term cycling degradation. Cycling degradation is fundamentally path-dependent: it is determined by charge-discharge cycles formed by the state-of-charge (SoC) trajectory and is commonly quantified via rainflow cycle counting. This non-Markovian structure makes it difficult to express degradation as an additive per-time-step cost, complicating classical dynamic programming approaches. We address this challenge by formulating the fleet scheduling problem as a Markov decision process (MDP) with constrained action space and designing a dense proxy reward that provides informative feedback at each time step while remaining aligned with long-term cycle-depth reduction. To scale learning to large state-action spaces induced by fine-grained SoC discretization and asymmetric per-battery constraints, we develop a function-approximation reinforcement learning method using an Extreme Learning Machine (ELM) as a random nonlinear feature map combined with linear temporal-difference learning. We evaluate the proposed approach on a toy Markovian signal model and on a Markovian model trained from real-world regulation signal traces obtained from the University of Delaware, and demonstrate consistent reductions in cycle-depth occurrence and degradation metrics compared to baseline scheduling policies.

Neural networks and machine learning models for uncertainty quantification suffer from limited scalability and poor reliability compared to their deterministic counterparts. In industry-scale active learning settings, where generating a single high-fidelity simulation may require days or weeks of computation and produce data volumes on the order of gigabytes, they quickly become impractical. This paper proposes a scalable and reliable Bayesian surrogate model, termed the Bayesian Interpolating Neural Network (B-INN). The B-INN combines high-order interpolation theory with tensor decomposition and alternating direction algorithm to enable effective dimensionality reduction without compromising predictive accuracy. We theoretically show that the function space of a B-INN is a subset of that of Gaussian processes, while its Bayesian inference exhibits linear complexity, $\mathcal{O}(N)$, with respect to the number of training samples. Numerical experiments demonstrate that B-INNs can be from 20 times to 10,000 times faster with a robust uncertainty estimation compared to Bayesian neural networks and Gaussian processes. These capabilities make B-INN a practical foundation for uncertainty-driven active learning in large-scale industrial simulations, where computational efficiency and robust uncertainty calibration are paramount.

The evolution of Large Language Model (LLM) agents for software engineering (SWE) is constrained by the scarcity of verifiable datasets, a bottleneck stemming from the complexity of constructing executable environments across diverse languages. To address this, we introduce MEnvAgent, a Multi-language framework for automated Environment construction that facilitates scalable generation of verifiable task instances. MEnvAgent employs a multi-agent Planning-Execution-Verification architecture to autonomously resolve construction failures and integrates a novel Environment Reuse Mechanism that reduces computational overhead by incrementally patching historical environments. Evaluations on MEnvBench, a new benchmark comprising 1,000 tasks across 10 languages, demonstrate that MEnvAgent outperforms baselines, improving Fail-to-Pass (F2P) rates by 8.6% while reducing time costs by 43%. Additionally, we demonstrate the utility of MEnvAgent by constructing MEnvData-SWE, the largest open-source polyglot dataset of realistic verifiable Docker environments to date, alongside solution trajectories that enable consistent performance gains on SWE tasks across a wide range of models. Our code, benchmark, and dataset are available at https://github.com/ernie-research/MEnvAgent.

We introduce Text Encoded Extrusion (TEE), a text-based representation that expresses mesh construction as sequences of face extrusions rather than polygon lists, and a method for generating 3D meshes from TEE using a large language model (LLM). By learning extrusion sequences that assemble a mesh, similar to the way artists create meshes, our approach naturally supports arbitrary output face counts and produces manifold meshes by design, in contrast to recent transformer-based models. The learnt extrusion sequences can also be applied to existing meshes - enabling editing in addition to generation. To train our model, we decompose a library of quadrilateral meshes with non-self-intersecting face loops into constituent loops, which can be viewed as their building blocks, and finetune an LLM on the steps for reassembling the meshes by performing a sequence of extrusions. We demonstrate that our representation enables reconstruction, novel shape synthesis, and the addition of new features to existing meshes.

Multimodal Attributed Graphs (MAGs) have been widely adopted for modeling complex systems by integrating multi-modal information, such as text and images, on nodes. However, we identify a discrepancy between the implicit semantic structure induced by different modality embeddings and the explicit graph structure. For instance, neighbors in the explicit graph structure may be close in one modality but distant in another. Since existing methods typically perform message passing over the fixed explicit graph structure, they inadvertently aggregate dissimilar features, introducing modality-specific noise and impeding effective node representation learning. To address this, we propose OptiMAG, an Unbalanced Optimal Transport-based regularization framework. OptiMAG employs the Fused Gromov-Wasserstein distance to explicitly guide cross-modal structural consistency within local neighborhoods, effectively mitigating structural-semantic conflicts. Moreover, a KL divergence penalty enables adaptive handling of cross-modal inconsistencies. This framework can be seamlessly integrated into existing multimodal graph models, acting as an effective drop-in regularizer. Experiments demonstrate that OptiMAG consistently outperforms baselines across multiple tasks, ranging from graph-centric tasks (e.g., node classification, link prediction) to multimodal-centric generation tasks (e.g., graph2text, graph2image). The source code will be available upon acceptance.

Since the introduction of the Transformer architecture for large language models, the softmax-based attention layer has faced increasing scrutinity due to its quadratic-time computational complexity. Attempts have been made to replace it with less complex methods, at the cost of reduced performance in most cases. We introduce Hierarchical Shift Mixing (HSM), a general framework for token mixing that distributes pairwise token interactions across Transformer layers rather than computing them densely within each layer. HSM enables linear-time complexity while remaining agnostic to the specific mixing function. We show that even simple HSM variants achieve performance close to softmax attention, and that hybrid architectures combining HSM with softmax attention can outperform a GPT-style Transformer baseline while reducing computational cost during both training and inference.

Training Large Language Models (LLMs) with high multilingual coverage is becoming increasingly important -- especially when monolingual resources are scarce. Recent studies have found that LLMs process multilingual inputs in shared concept spaces, thought to support generalization and cross-lingual transfer. However, these prior studies often do not use causal methods, lack deeper error analysis or focus on the final model only, leaving open how these spaces emerge during training. We investigate the development of language-agnostic concept spaces during pretraining of EuroLLM through the causal interpretability method of activation patching. We isolate cross-lingual concept representations, then inject them into a translation prompt to investigate how consistently translations can be altered, independently of the language. We find that shared concept spaces emerge early} and continue to refine, but that alignment with them is language-dependent}. Furthermore, in contrast to prior work, our fine-grained manual analysis reveals that some apparent gains in translation quality reflect shifts in behavior -- like selecting senses for polysemous words or translating instead of copying cross-lingual homographs -- rather than improved translation ability. Our findings offer new insight into the training dynamics of cross-lingual alignment and the conditions under which causal interpretability methods offer meaningful insights in multilingual contexts.

Flow-based models have proven successful for time-series generation, particularly when defined in lower-dimensional latent spaces that enable efficient sampling. However, how to design latent representations with desirable equivariance properties for time-series generative modelling remains underexplored. In this work, we propose a latent flow-matching framework in which equivariance is explicitly encouraged through a simple regularisation of a pre-trained autoencoder. Specifically, we introduce an equivariance loss that enforces consistency between transformed signals and their reconstructions, and use it to fine-tune latent spaces with respect to basic time-series transformations such as translation and amplitude scaling. We show that these equivariance-regularised latent spaces improve generation quality while preserving the computational advantages of latent flow models. Experiments on multiple real-world datasets demonstrate that our approach consistently outperforms existing diffusion-based baselines in standard time-series generation metrics, while achieving orders-of-magnitude faster sampling. These results highlight the practical benefits of incorporating geometric inductive biases into latent generative models for time series.

We report on Just-in-Time catching test generation at Meta, designed to prevent bugs in large scale backend systems of hundreds of millions of line of code. Unlike traditional hardening tests, which pass at generation time, catching tests are meant to fail, surfacing bugs before code lands. The primary challenge is to reduce development drag from false positive test failures. Analyzing 22,126 generated tests, we show code-change-aware methods improve candidate catch generation 4x over hardening tests and 20x over coincidentally failing tests. To address false positives, we use rule-based and LLM-based assessors. These assessors reduce human review load by 70%. Inferential statistical analysis showed that human-accepted code changes are assessed to have significantly more false positives, while human-rejected changes have significantly more true positives. We reported 41 candidate catches to engineers; 8 were confirmed to be true positives, 4 of which would have led to serious failures had they remained uncaught. Overall, our results show that Just-in-Time catching is scalable, industrially applicable, and that it prevents serious failures from reaching production.

Continual learning (CL) in vision-language models (VLMs) faces significant challenges in improving task adaptation and avoiding catastrophic forgetting. Existing methods usually have heavy inference burden or rely on external knowledge, while Low-Rank Adaptation (LoRA) has shown potential in reducing these issues by enabling parameter-efficient tuning. However, considering directly using LoRA to alleviate the catastrophic forgetting problem is non-trivial, we introduce a novel framework that restructures a single LoRA module as a decomposable Rank-1 Expert Pool. Our method learns to dynamically compose a sparse, task-specific update by selecting from this expert pool, guided by the semantics of the [CLS] token. In addition, we propose an Activation-Guided Orthogonal (AGO) loss that orthogonalizes critical parts of LoRA weights across tasks. This sparse composition and orthogonalization enable fewer parameter updates, resulting in domain-aware learning while minimizing inter-task interference and maintaining downstream task performance. Extensive experiments across multiple settings demonstrate state-of-the-art results in all metrics, surpassing zero-shot upper bounds in generalization. Notably, it reduces trainable parameters by 96.7% compared to the baseline method, eliminating reliance on external datasets or task-ID discriminators. The merged LoRAs retain less weights and incur no inference latency, making our method computationally lightweight.

We study offline reinforcement learning of style-conditioned policies using explicit style supervision via subtrajectory labeling functions. In this setting, aligning style with high task performance is particularly challenging due to distribution shift and inherent conflicts between style and reward. Existing methods, despite introducing numerous definitions of style, often fail to reconcile these objectives effectively. To address these challenges, we propose a unified definition of behavior style and instantiate it into a practical framework. Building on this, we introduce Style-Conditioned Implicit Q-Learning (SCIQL), which leverages offline goal-conditioned RL techniques, such as hindsight relabeling and value learning, and combine it with a new Gated Advantage Weighted Regression mechanism to efficiently optimize task performance while preserving style alignment. Experiments demonstrate that SCIQL achieves superior performance on both objectives compared to prior offline methods. Code, datasets and visuals are available in: https://sciql-iclr-2026.github.io/.

Large-scale Electronic Health Record (EHR) databases have become indispensable in supporting clinical decision-making through data-driven treatment recommendations. However, existing medication recommender methods often struggle with a user (i.e., patient) cold-start problem, where recommendations for new patients are usually unreliable due to the lack of sufficient prescription history for patient profiling. While prior studies have utilized medical knowledge graphs to connect medication concepts through pharmacological or chemical relationships, these methods primarily focus on mitigating the item cold-start issue and fall short in providing personalized recommendations that adapt to individual patient characteristics. Meta-learning has shown promise in handling new users with sparse interactions in recommender systems. However, its application to EHRs remains underexplored due to the unique sequential structure of EHR data. To tackle these challenges, we propose MetaDrug, a multi-level, uncertainty-aware meta-learning framework designed to address the patient cold-start problem in medication recommendation. MetaDrug proposes a novel two-level meta-adaptation mechanism, including self-adaptation, which adapts the model to new patients using their own medical events as support sets to capture temporal dependencies; and peer-adaptation, which adapts the model using similar visits from peer patients to enrich new patient representations. Meanwhile, to further improve meta-adaptation outcomes, we introduce an uncertainty quantification module that ranks the support visits and filters out the unrelated information for adaptation consistency. We evaluate our approach on the MIMIC-III and Acute Kidney Injury (AKI) datasets. Experimental results on both datasets demonstrate that MetaDrug consistently outperforms state-of-the-art medication recommendation methods on cold-start patients.

Fine-tuned LLMs can covertly encode prompt secrets into outputs via steganographic channels. Prior work demonstrated this threat but relied on trivially recoverable encodings. We formalize payload recoverability via classifier accuracy and show previous schemes achieve 100\% recoverability. In response, we introduce low-recoverability steganography, replacing arbitrary mappings with embedding-space-derived ones. For Llama-8B (LoRA) and Ministral-8B (LoRA) trained on TrojanStego prompts, exact secret recovery rises from 17$\rightarrow$30\% (+78\%) and 24$\rightarrow$43\% (+80\%) respectively, while on Llama-70B (LoRA) trained on Wiki prompts, it climbs from 9$\rightarrow$19\% (+123\%), all while reducing payload recoverability. We then discuss detection. We argue that detecting fine-tuning-based steganographic attacks requires approaches beyond traditional steganalysis. Standard approaches measure distributional shift, which is an expected side-effect of fine-tuning. Instead, we propose a mechanistic interpretability approach: linear probes trained on later-layer activations detect the secret with up to 33\% higher accuracy in fine-tuned models compared to base models, even for low-recoverability schemes. This suggests that malicious fine-tuning leaves actionable internal signatures amenable to interpretability-based defenses.

Advances in generative modeling have recently been adapted to tabular data containing discrete and continuous features. However, generating mixed-type features that combine discrete states with an otherwise continuous distribution in a single feature remains challenging. We advance the state-of-the-art in diffusion models for tabular data with a cascaded approach. We first generate a low-resolution version of a tabular data row, that is, the collection of the purely categorical features and a coarse categorical representation of numerical features. Next, this information is leveraged in the high-resolution flow matching model via a novel guided conditional probability path and data-dependent coupling. The low-resolution representation of numerical features explicitly accounts for discrete outcomes, such as missing or inflated values, and therewith enables a more faithful generation of mixed-type features. We formally prove that this cascade tightens the transport cost bound. The results indicate that our model generates significantly more realistic samples and captures distributional details more accurately, for example, the detection score increases by 40%.

The NVFP4 lower-precision format, supported in hardware by NVIDIA Blackwell GPUs, promises to allow, for the first time, end-to-end fully-quantized pre-training of massive models such as LLMs. Yet, existing quantized training methods still sacrifice some of the representation capacity of this format in favor of more accurate unbiased quantized gradient estimation by stochastic rounding (SR), losing noticeable accuracy relative to standard FP16 and FP8 training. In this paper, improve the state of the art for quantized training in NVFP4 via a novel unbiased quantization routine for micro-scaled formats, called MS-EDEN, that has more than 2x lower quantization error than SR. We integrate it into a novel fully-NVFP4 quantization scheme for linear layers, called Quartet II. We show analytically that Quartet II achieves consistently better gradient estimation across all major matrix multiplications, both on the forward and on the backward passes. In addition, our proposal synergizes well with recent training improvements aimed specifically at NVFP4. We further validate Quartet II on end-to-end LLM training with up to 1.9B parameters on 38B tokens. We provide kernels for execution on NVIDIA Blackwell GPUs with up to 4.2x speedup over BF16. Our code is available at https://github.com/IST-DASLab/Quartet-II .

The standard approach to representation learning on attributed graphs -- i.e., simultaneously reconstructing node attributes and graph structure -- is geometrically flawed, as it merges two potentially incompatible metric spaces. This forces a destructive alignment that erodes information about the graph's underlying generative process. To recover this lost signal, we introduce a custom variational autoencoder that separates manifold learning from structural alignment. By quantifying the metric distortion needed to map the attribute manifold onto the graph's Heat Kernel, we transform geometric conflict into an interpretable structural descriptor. Experiments show our method uncovers connectivity patterns and anomalies undetectable by conventional approaches, proving both their theoretical inadequacy and practical limitations.

Hierarchical sequence models replace fixed tokenization with learned segmentations that compress long byte sequences for efficient autoregressive modeling. While recent end-to-end methods can learn meaningful boundaries from the language-modeling objective alone, it remains difficult to quantitatively assess and systematically steer where compute is spent. We introduce a router-agnostic metric of boundary quality, boundary enrichment B, which measures how strongly chunk starts concentrate on positions with high next-byte surprisal. Guided by this metric, we propose Sombrero, which steers boundary placement toward predictive difficulty via a confidence-alignment boundary loss and stabilizes boundary learning by applying confidence-weighted smoothing at the input level rather than on realized chunks. On 1B scale, across UTF-8 corpora covering English and German text as well as code and mathematical content, Sombrero improves the accuracy-efficiency trade-off and yields boundaries that more consistently align compute with hard-to-predict positions.

Code verifiers play a critical role in post-verification for LLM-based code generation, yet existing supervised fine-tuning methods suffer from data scarcity, high failure rates, and poor inference efficiency. While reinforcement learning (RL) offers a promising alternative by optimizing models through execution-driven rewards without labeled supervision, our preliminary results show that naive RL with only functionality rewards fails to generate effective unit tests for difficult branches and samples. We first theoretically analyze showing that branch coverage, sample difficulty, syntactic and functional correctness can be jointly modeled as RL rewards, where optimizing these signals can improve the reliability of unit-test-based verification. Guided by this analysis, we design syntax- and functionality-aware rewards and further propose branch- and sample-difficulty--aware RL using exponential reward shaping and static analysis metrics. With this formulation, CVeDRL achieves state-of-the-art performance with only 0.6B parameters, yielding up to 28.97% higher pass rate and 15.08% higher branch coverage than GPT-3.5, while delivering over $20\times$ faster inference than competitive baselines. Code is available at https://github.com/LIGHTCHASER1/CVeDRL.git

Reinforcement learning has become central to post-training large language models, yet dominant algorithms rely on clipping mechanisms that introduce optimization issues at scale, including zero-gradient regions, reward hacking, and training instability. We propose Clipping-Free Policy Optimization (CFPO), which replaces heuristic clipping with a convex quadratic penalty derived from Total Variation divergence constraints, yielding an everywhere-differentiable objective that enforces stable policy updates without hard boundaries. We evaluate CFPO across both reasoning and alignment settings. In reasoning, CFPO matches clipping-based methods on downstream benchmarks while extending the stable training regime. In alignment, CFPO mitigates verbosity exploitation and reduces capability degradation, while achieving competitive instruction-following performance. CFPO requires only a one-line code change and no additional hyperparameters. Our results suggest that CFPO is a promising drop-in alternative to clipping-based methods for LLM post-training.

Understanding user behavior is essential for improving digital experiences, optimizing business conversions, and mitigating threats like account takeovers, fraud, and bot attacks. Most platforms separate product analytics and security, creating fragmented visibility and delayed threat detection. Trackly, a scalable SaaS platform, unifies comprehensive user behavior analytics with real time, rule based anomaly detection. It tracks sessions, IP based geo location, device browser fingerprints, and granular events such as page views, add to cart, and checkouts. Suspicious activities logins from new devices or locations, impossible travel (Haversine formula), rapid bot like actions, VPN proxy usage, or multiple accounts per IP are flagged via configurable rules with weighted risk scoring, enabling transparent, explainable decisions. A real time dashboard provides global session maps, DAU MAU, bounce rates, and session durations. Integration is simplified with a lightweight JavaScript SDK and secure REST APIs. Implemented on a multi tenant microservices stack (ASP.NET Core, MongoDB, RabbitMQ, Next.js), Trackly achieved 98.1% accuracy, 97.7% precision, and 2.25% false positives on synthetic datasets, proving its efficiency for SMEs and ecommerce.

Transformer-based large language models (LLMs) are comprised of billions of parameters arranged in deep and wide computational graphs. Several studies on LLM efficiency optimization argue that it is possible to prune a significant portion of the parameters, while only marginally impacting performance. This suggests that the computation is not uniformly distributed across the parameters. We introduce here a technique to systematically quantify computation density in LLMs. In particular, we design a density estimator drawing on mechanistic interpretability. We experimentally test our estimator and find that: (1) contrary to what has been often assumed, LLM processing generally involves dense computation; (2) computation density is dynamic, in the sense that models shift between sparse and dense processing regimes depending on the input; (3) per-input density is significantly correlated across LLMs, suggesting that the same inputs trigger either low or high density. Investigating the factors influencing density, we observe that predicting rarer tokens requires higher density, and increasing context length often decreases the density. We believe that our computation density estimator will contribute to a better understanding of the processing at work in LLMs, challenging their symbolic interpretation.

We present CALM, a joint Contextual Acoustic-Linguistic Modeling framework for multi-speaker automatic speech recognition (ASR). In personalized AI scenarios, the joint availability of acoustic and linguistic cues naturally motivates the integration of target-speaker conditioning with contextual biasing in overlapping conversations. CALM implements this integration in an end-to-end framework through speaker embedding-driven target-speaker extraction and dynamic vocabulary-based contextual biasing. We evaluate CALM on simulated English (LibriSpeechMix) and Japanese (Corpus of Spontaneous Japanese mixtures, CSJMix). On two-speaker mixtures, CALM reduces biased word error rate (B-WER) from 12.7 to 4.7 on LibriSpeech2Mix and biased character error rate (B-CER) from 16.6 to 8.4 on CSJMix2 (eval3), demonstrating the effectiveness of joint acoustic-linguistic modeling across languages. We additionally report results on the AMI corpus (IHM-mix condition) to validate performance on standardized speech mixtures.

Boundary value analysis and testing (BVT) is fundamental in software quality assurance because faults tend to cluster at input extremes, yet testers often struggle to understand and justify why certain input-output pairs represent meaningful behavioral boundaries. Large Language Models (LLMs) could help by producing natural-language rationales, but their value for BVT has not been empirically assessed. We therefore conducted an exploratory study on LLM-generated boundary explanations: in a survey, twenty-seven software professionals rated GPT-4.1 explanations for twenty boundary pairs on clarity, correctness, completeness and perceived usefulness, and six of them elaborated in follow-up interviews. Overall, 63.5% of all ratings were positive (4-5 on a five-point Likert scale) compared to 17% negative (1-2), indicating general agreement but also variability in perceptions. Participants favored explanations that followed a clear structure, cited authoritative sources, and adapted their depth to the reader's expertise; they also stressed the need for actionable examples to support debugging and documentation. From these insights, we distilled a seven-item requirement checklist that defines concrete design criteria for future LLM-based boundary explanation tools. The results suggest that, with further refinement, LLM-based tools can support testing workflows by making boundary explanations more actionable and trustworthy.

Large reasoning models have shown strong performance through extended chain-of-thought reasoning, yet their computational cost remains significant. Probably approximately correct (PAC) reasoning provides statistical guarantees for efficient reasoning by adaptively switching between thinking and non-thinking models, but the guarantee holds only in the marginal case and does not provide exact conditional coverage. We propose G-PAC reasoning, a practical framework that provides PAC-style guarantees at the group level by partitioning the input space. We develop two instantiations: Group PAC (G-PAC) reasoning for known group structures and Clustered PAC (C-PAC) reasoning for unknown groupings. We prove that both G-PAC and C-PAC achieve group-conditional risk control, and that grouping can strictly improve efficiency over marginal PAC reasoning in heterogeneous settings. Our experiments on diverse reasoning benchmarks demonstrate that G-PAC and C-PAC successfully achieve group-conditional risk control while maintaining substantial computational savings.

Post-training compression is currently divided into two contrasting regimes. On the one hand, fast, data-free, and model-agnostic methods (e.g., NF4 or HQQ) offer maximum accessibility but suffer from functional collapse at extreme bit-rates below 4 bits. On the other hand, techniques leveraging calibration data or extensive recovery training achieve superior fidelity but impose high computational constraints and face uncertain robustness under data distribution shifts. We introduce EntQuant, the first framework to unite the advantages of these distinct paradigms. By matching the performance of data-dependent methods with the speed and universality of data-free techniques, EntQuant enables practical utility in the extreme compression regime. Our method decouples numerical precision from storage cost via entropy coding, compressing a 70B parameter model in less than 30 minutes. We demonstrate that EntQuant does not only achieve state-of-the-art results on standard evaluation sets and models, but also retains functional performance on more complex benchmarks with instruction-tuned models, all at modest inference overhead.

The pursuit of Artificial General Intelligence (AGI) is a central goal in language model development, in which consciousness-like processing could serve as a key facilitator. While current language models are not conscious, they exhibit behaviors analogous to certain aspects of consciousness. This paper investigates the implementation of a leading theory of consciousness, Integrated Information Theory (IIT), within language models via a reward-based learning paradigm. IIT provides a formal, axiom-based mathematical framework for quantifying consciousness. Drawing inspiration from its core principles, we formulate a novel reward function that quantifies a text's causality, coherence and integration, characteristics associated with conscious processing. Empirically, it is found that optimizing for this IIT-inspired reward leads to more concise text generation. On out of domain tasks, careful tuning achieves up to a 31% reduction in output length while preserving accuracy levels comparable to the base model. In addition to primary task performance, the broader effects of this training methodology on the model's confidence calibration and test-time computational scaling is analyzed. The proposed framework offers significant practical advantages: it is conceptually simple, computationally efficient, requires no external data or auxiliary models, and leverages a general, capability-driven signal rather than task-specific heuristics. Code available at https://github.com/MH-Sameti/LLM_PostTraining.git

We introduce a rank-statistic approximation of $f$-divergences that avoids explicit density-ratio estimation by working directly with the distribution of ranks. For a resolution parameter $K$, we map the mismatch between two univariate distributions $μ$ and $ν$ to a rank histogram on $\{ 0, \ldots, K\}$ and measure its deviation from uniformity via a discrete $f$-divergence, yielding a rank-statistic divergence estimator. We prove that the resulting estimator of the divergence is monotone in $K$, is always a lower bound of the true $f$-divergence, and we establish quantitative convergence rates for $K\to\infty$ under mild regularity of the quantile-domain density ratio. To handle high-dimensional data, we define the sliced rank-statistic $f$-divergence by averaging the univariate construction over random projections, and we provide convergence results for the sliced limit as well. We also derive finite-sample deviation bounds along with asymptotic normality results for the estimator. Finally, we empirically validate the approach by benchmarking against neural baselines and illustrating its use as a learning objective in generative modelling experiments.

Large-scale biodiversity monitoring platforms increasingly rely on multimodal wildlife observations. While recent foundation models enable rich semantic representations across vision, audio, and language, retrieving relevant observations from massive archives remains challenging due to the computational cost of high-dimensional similarity search. In this work, we introduce compact hypercube embeddings for fast text-based wildlife observation retrieval, a framework that enables efficient text-based search over large-scale wildlife image and audio databases using compact binary representations. Building on the cross-view code alignment hashing framework, we extend lightweight hashing beyond a single-modality setup to align natural language descriptions with visual or acoustic observations in a shared Hamming space. Our approach leverages pretrained wildlife foundation models, including BioCLIP and BioLingual, and adapts them efficiently for hashing using parameter-efficient fine-tuning. We evaluate our method on large-scale benchmarks, including iNaturalist2024 for text-to-image retrieval and iNatSounds2024 for text-to-audio retrieval, as well as multiple soundscape datasets to assess robustness under domain shift. Results show that retrieval using discrete hypercube embeddings achieves competitive, and in several cases superior, performance compared to continuous embeddings, while drastically reducing memory and search cost. Moreover, we observe that the hashing objective consistently improves the underlying encoder representations, leading to stronger retrieval and zero-shot generalization. These results demonstrate that binary, language-based retrieval enables scalable and efficient search over large wildlife archives for biodiversity monitoring systems.

Large-scale, high-quality interaction trajectories are essential for advancing mobile Graphical User Interface (GUI) agents. While existing methods typically rely on labor-intensive human demonstrations or automated model exploration to generate GUI trajectories, they lack fine-grained control over task difficulty. This fundamentally restricts learning effectiveness due to the mismatch between the training difficulty and the agent's capabilities. Inspired by how humans acquire skills through progressively challenging tasks, we propose MobileGen, a novel data generation framework that adaptively aligns training difficulty with the GUI agent's capability frontier. Specifically, MobileGen explicitly decouples task difficulty into structural (e.g., trajectory length) and semantic (e.g., task goal) dimensions. It then iteratively evaluates the agent on a curated prior dataset to construct a systematic profile of its capability frontier across these two dimensions. With this profile, the probability distribution of task difficulty is adaptively computed, from which the target difficulty for the next round of training can be sampled. Guided by the sampled difficulty, a multi-agent controllable generator is finally used to synthesize high-quality interaction trajectories along with corresponding task instructions. Extensive experiments show that MobileGen consistently outperforms existing data generation methods by improving the average performance of GUI agents by 1.57 times across multiple challenging benchmarks. This highlights the importance of capability-aligned data generation for effective mobile GUI agent training.

Simultaneous speech translation (SST) produces target text incrementally from partial speech input. Recent speech large language models (Speech LLMs) have substantially improved SST quality, yet they still struggle to correctly translate rare and domain-specific terminology. While retrieval augmentation has been effective for terminology translation in machine translation, bringing retrieval to SST is non-trivial: it requires fast and accurate cross-modal (speech-to-text) retrieval under partial, continually arriving input, and the model must decide whether and when to apply retrieved terms during incremental generation. We propose Retrieval-Augmented Simultaneous Speech Translation (RASST), which tightly integrates cross-modal retrieval into the SST pipeline. RASST trains a lightweight speech-text retriever and performs efficient sliding-window retrieval, providing chunkwise terminology hints to the Speech LLM. We further synthesize training data that teaches the Speech LLM to leverage retrieved terms precisely. Experiments on three language directions of the ACL 60/60 dev set show that RASST improves terminology translation accuracy by up to 16% and increases overall translation quality by up to 3 BLEU points, with ablations confirming the contribution of each component.

Multi-turn tool-integrated reasoning enables Large Language Models (LLMs) to solve complex tasks through iterative information retrieval. However, current reinforcement learning (RL) frameworks for search-augmented reasoning predominantly rely on sparse outcome-level rewards, leading to a "Double Homogenization Dilemma." This manifests as (1) Process homogenization, where the thinking, reasoning, and tooling involved in generation are ignored. (2) Intra-group homogenization, coarse-grained outcome rewards often lead to inefficiencies in intra-group advantage estimation with methods like Group Relative Policy Optimization (GRPO) during sampling. To address this, we propose Turn-level Stage-aware Policy Optimization (TSPO). TSPO introduces the First-Occurrence Latent Reward (FOLR) mechanism, allocating partial rewards to the step where the ground-truth answer first appears, thereby preserving process-level signals and increasing reward variance within groups without requiring external reward models or any annotations. Extensive experiments demonstrate that TSPO significantly outperforms state-of-the-art baselines, achieving average performance gains of 24% and 13.6% on Qwen2.5-3B and 7B models, respectively.

Safety cases, structured arguments that a system is acceptably safe, are becoming central to the governance of AI systems. Yet, traditional safety-case practices from aviation or nuclear engineering rely on well-specified system boundaries, stable architectures, and known failure modes. Modern AI systems such as generative and agentic AI are the opposite. Their capabilities emerge unpredictably from low-level training objectives, their behaviour varies with prompts, and their risk profiles shift through fine-tuning, scaffolding, or deployment context. This study examines how safety cases are currently constructed for AI systems and why classical approaches fail to capture these dynamics. It then proposes a framework of reusable safety-case templates, each following a predefined structure of claims, arguments, and evidence tailored for AI systems. The framework introduces comprehensive taxonomies for AI-specific claim types (assertion-based, constrained-based, capability-based), argument types (demonstrative, comparative, causal/explanatory, risk-based, and normative), and evidence families (empirical, mechanistic, comparative, expert-driven, formal methods, operational/field data, and model-based). Each template is illustrated through end-to-end patterns addressing distinctive challenges such as evaluation without ground truth, dynamic model updates, and threshold-based risk decisions. The result is a systematic, composable, and reusable approach to constructing and maintaining safety cases that are credible, auditable, and adaptive to the evolving behaviour of generative and frontier AI systems.

Feature-based explanation methods aim to quantify how features influence the model's behavior, either locally or globally, but different methods often disagree, producing conflicting explanations. This disagreement arises primarily from two sources: how feature interactions are handled and how feature dependencies are incorporated. We propose GRANITE, a generalized regional explanation framework that partitions the feature space into regions where interaction and distribution influences are minimized. This approach aligns different explanation methods, yielding more consistent and interpretable explanations. GRANITE unifies existing regional approaches, extends them to feature groups, and introduces a recursive partitioning algorithm to estimate such regions. We demonstrate its effectiveness on real-world datasets, providing a practical tool for consistent and interpretable feature explanations.

Dominant approaches, e.g. the EU's "Trustworthy AI framework", treat trust as a property that can be designed for, evaluated, and governed according to normative and technical criteria. They do not address how trust is subjectively cultivated and experienced, culturally embedded, and inherently relational. This paper proposes some expanded principles for trust in AI that can be incorporated into common development methods and frame trust as a dynamic, temporal relationship, which involves transparency and mutual respect. We draw on relational ethics and, in particular, African communitarian philosophies, to foreground the nuances of inclusive, participatory processes and long-term relationships with communities. Involving communities throughout the AI lifecycle can foster meaningful relationships with AI design and development teams that incrementally build trust and promote more equitable and context-sensitive AI systems. We illustrate how trust-enabling principles based on African relational ethics can be operationalised, using two use-cases for AI: healthcare and education.

Recent work has revealed a link between self-attention mechanisms in transformers and test-time kernel regression via the Nadaraya-Watson estimator, with standard softmax attention corresponding to a Gaussian kernel. However, a kernel-theoretic understanding of sparse attention mechanisms is currently missing. In this paper, we establish a formal correspondence between sparse attention and compact (bounded support) kernels. We show that normalized ReLU and sparsemax attention arise from Epanechnikov kernel regression under fixed and adaptive normalizations, respectively. More generally, we demonstrate that widely used kernels in nonparametric density estimation -- including Epanechnikov, biweight, and triweight -- correspond to $α$-entmax attention with $α= 1 + \frac{1}{n}$ for $n \in \mathbb{N}$, while the softmax/Gaussian relationship emerges in the limit $n \to \infty$. This unified perspective explains how sparsity naturally emerges from kernel design and provides principled alternatives to heuristic top-$k$ attention and other associative memory mechanisms. Experiments with a kernel-regression-based variant of transformers -- Memory Mosaics -- show that kernel-based sparse attention achieves competitive performance on language modeling, in-context learning, and length generalization tasks, offering a principled framework for designing attention mechanisms.

The completion of a Euclidean distance matrix (EDM) from sparse and noisy observations is a fundamental challenge in signal processing, with applications in sensor network localization, acoustic room reconstruction, molecular conformation, and manifold learning. Traditional approaches, such as rank-constrained optimization and semidefinite programming, enforce geometric constraints but often struggle under sparse or noisy conditions. This paper introduces a hierarchical Bayesian framework that places structured priors directly on the latent point set generating the EDM, naturally embedding geometric constraints. By incorporating a hierarchical prior on latent point set, the model enables automatic regularization and robust noise handling. Posterior inference is performed using a Metropolis-Hastings within Gibbs sampler to handle coupled latent point posterior. Experiments on synthetic data demonstrate improved reconstruction accuracy compared to deterministic baselines in sparse regimes.

Music often shares notable parallels with language, motivating the use of pretrained large language models (LLMs) for symbolic music understanding and generation. Despite growing interest, the practical effectiveness of adapting instruction-tuned LLMs to symbolic music remains insufficiently characterized. We present a controlled comparative study of finetuning strategies for ABC-based generation and understanding, comparing an off-the-shelf instruction-tuned backbone to domain-adapted variants and a music-specialized LLM baseline. Across multiple symbolic music corpora and evaluation signals, we provide some insights into adaptation choices for symbolic music applications. We highlight the domain adaptation vs.~preserving prior information tradeoff as well as the distinct behaviour of metrics used to measure the domain adaptation for symbolic music.

The performance of deep learning models critically depends on efficient kernel implementations, yet developing high-performance kernels for specialized accelerators remains time-consuming and expertise-intensive. While recent work demonstrates that large language models (LLMs) can generate correct and performant GPU kernels, kernel generation for neural processing units (NPUs) remains largely underexplored due to domain-specific programming models, limited public examples, and sparse documentation. Consequently, directly generating AscendC kernels with LLMs yields extremely low correctness, highlighting a substantial gap between GPU and NPU kernel generation. We present AscendCraft, a DSL-guided approach for automatic AscendC kernel generation. AscendCraft introduces a lightweight DSL that abstracts non-essential complexity while explicitly modeling Ascend-specific execution semantics. Kernels are first generated in the DSL using category-specific expert examples and then transcompiled into AscendC through structured, constraint-driven LLM lowering passes. Evaluated on MultiKernelBench across seven operator categories, AscendCraft achieves 98.1% compilation success and 90.4% functional correctness. Moreover, 46.2% of generated kernels match or exceed PyTorch eager execution performance, demonstrating that DSL-guided transcompilation can enable LLMs to generate both correct and competitive NPU kernels. Beyond benchmarks, AscendCraft further demonstrates its generality by successfully generating two correct kernels for newly proposed mHC architecture, achieving performance that substantially surpasses PyTorch eager execution.

As generative artificial intelligence advances, Large Language Models (LLMs) are being explored for automated graphical user interface (GUI) design. This study investigates the usability and adaptability of LLM-generated interfaces by analysing their ability to meet diverse user needs. The experiments included utilization of three state-of-the-art models from January 2025 (OpenAI GPT o3-mini-high, DeepSeek R1, and Anthropic Claude 3.5 Sonnet) generating mockups for three interface types: a chat system, a technical team panel, and a manager dashboard. Expert evaluations revealed that while LLMs are effective at creating structured layouts, they face challenges in meeting accessibility standards and providing interactive functionality. Further testing showed that LLMs could partially tailor interfaces for different user personas but lacked deeper contextual understanding. The results suggest that while LLMs are promising tools for early-stage UI prototyping, human intervention remains critical to ensure usability, accessibility, and user satisfaction.

Large language model agents often fail to accumulate knowledge from experience, treating each task as an independent challenge. Recent methods extract experience as flattened textual knowledge, which cannot capture procedural logic of complex subtasks. They also lack maintenance mechanisms, causing repository degradation as experience accumulates. We introduce AutoRefine, a framework that extracts and maintains dual-form Experience Patterns from agent execution histories. For procedural subtasks, we extract specialized subagents with independent reasoning and memory. For static knowledge, we extract skill patterns as guidelines or code snippets. A continuous maintenance mechanism scores, prunes, and merges patterns to prevent repository degradation. Evaluated on ALFWorld, ScienceWorld, and TravelPlanner, AutoRefine achieves 98.4%, 70.4%, and 27.1% respectively, with 20-73% step reductions. On TravelPlanner, automatic extraction exceeds manually designed systems (27.1% vs 12.1%), demonstrating its ability to capture procedural coordination.

Molecular generative models, often employing GPT-style language modeling on molecular string representations, have shown promising capabilities when scaled to large datasets and model sizes. However, it remains unclear and subject to debate whether these models adhere to predictable scaling laws under fixed computational budgets, which is a crucial understanding for optimally allocating resources between model size, data volume, and molecular representation. In this study, we systematically investigate the scaling behavior of molecular language models across both pretraining and downstream tasks. We train 300 models and conduct over 10,000 experiments, rigorously controlling compute budgets while independently varying model size, number of training tokens, and molecular representation. Our results demonstrate clear scaling laws in molecular models for both pretraining and downstream transfer, reveal the substantial impact of molecular representation on performance, and explain previously observed inconsistencies in scaling behavior for molecular generation. Additionally, we publicly release the largest library of molecular language models to date to facilitate future research and development. Code and models are available at https://github.com/SZU-ADDG/MLM-Scaling.

Deep neural networks often generalize well despite heavy over-parameterization, challenging classical parameter-based analyses. We study generalization from a representation-centric perspective and analyze how the geometry of learned embeddings controls predictive performance for a fixed trained model. We show that population risk can be bounded by two factors: (i) the intrinsic dimension of the embedding distribution, which determines the convergence rate of empirical embedding distribution to the population distribution in Wasserstein distance, and (ii) the sensitivity of the downstream mapping from embeddings to predictions, characterized by Lipschitz constants. Together, these yield an embedding-dependent error bound that does not rely on parameter counts or hypothesis class complexity. At the final embedding layer, architectural sensitivity vanishes and the bound is dominated by embedding dimension, explaining its strong empirical correlation with generalization performance. Experiments across architectures and datasets validate the theory and demonstrate the utility of embedding-based diagnostics.

Industrial troubleshooting guides encode diagnostic procedures in flowchart-like diagrams where spatial layout and technical language jointly convey meaning. To integrate this knowledge into operator support systems, which assist shop-floor personnel in diagnosing and resolving equipment issues, the information must first be extracted and structured for machine interpretation. However, when performed manually, this extraction is labor-intensive and error-prone. Vision Language Models offer potential to automate this process by jointly interpreting visual and textual meaning, yet their performance on such guides remains underexplored. This paper evaluates two VLMs on extracting structured knowledge, comparing two prompting strategies: standard instruction-guided versus an augmented approach that cues troubleshooting layout patterns. Results reveal model-specific trade-offs between layout sensitivity and semantic robustness, informing practical deployment decisions.

We propose Obfuscated Semantic Null space Injection for Privacy (OSNIP), a lightweight client-side encryption framework for privacy-preserving LLM inference. Generalizing the geometric intuition of linear kernels to the high-dimensional latent space of LLMs, we formally define the ``Obfuscated Semantic Null Space'', a high-dimensional regime that preserves semantic fidelity while enforcing near-orthogonality to the original embedding. By injecting perturbations that project the original embedding into this space, OSNIP ensures privacy without any post-processing. Furthermore, OSNIP employs a key-dependent stochastic mapping that synthesizes individualized perturbation trajectories unique to each user. Evaluations on 12 generative and classification benchmarks show that OSNIP achieves state-of-the-art performance, sharply reducing attack success rates while maintaining strong model utility under strict security constraints.

Physical systems near singularities, interfaces, and critical points exhibit power-law scaling, yet standard neural networks leave the governing exponents implicit. We introduce physics-informed M"untz-Sz'asz Networks (MSN-PINN), a power-law basis network that treats scaling exponents as trainable parameters. The model outputs both the solution and its scaling structure. We prove identifiability, or unique recovery, and show that, under these conditions, the squared error between learned and true exponents scales as $O(|μ- α|^2)$. Across experiments, MSN-PINN achieves single-exponent recovery with 1--5% error under noise and sparse sampling. It recovers corner singularity exponents for the two-dimensional Laplace equation with 0.009% error, matches the classical result of Kondrat'ev (1967), and recovers forcing-induced exponents in singular Poisson problems with 0.03% and 0.05% errors. On a 40-configuration wedge benchmark, it reaches a 100% success rate with 0.022% mean error. Constraint-aware training encodes physical requirements such as boundary condition compatibility and improves accuracy by three orders of magnitude over naive training. By combining the expressiveness of neural networks with the interpretability of asymptotic analysis, MSN-PINN produces learned parameters with direct physical meaning.

Software reuse has long been recognized as a critical and widely studied topic in software engineering, offering substantial benefits in reducing development costs, improving software quality, and enhancing operational efficiency. This paradigm extends into deep learning through model reuse. Recently, model merging has emerged in the domain of large language models (LLMs) as a training-free approach that takes multiple task-specific models with the same architecture as source models and merges them without retraining, enhancing model reuse within LLMs. However, no prior work has systematically investigated whether such an approach can be effectively applied to other deep learning models with different architectures across domains. To bridge this gap, we present the first systematic study that evaluates five model merging techniques on three distinct model architectures across three domains: LLMs, image classification, and autonomous driving. Our findings reveal that directly applying existing model merging techniques leads to highly inconsistent results and falls notably short of their success within LLMs. Moreover, a single model merging technique often fails to handle the heterogeneous structural properties within a model, limiting its applicability to different model architectures across domains. Furthermore, the effectiveness of model merging techniques is highly sensitive to hyperparameter configurations, thereby constraining their potential for broader adoption. Inspired by these insights, we propose AutoMerge, a novel search-based model merging framework that first segments complex models into multiple heterogeneous blocks and then systematically explores the merging space to identify the merging technique and its hyperparameter configuration.

Urban region profiling, the task of characterizing geographical areas, is crucial for urban planning and resource allocation. However, existing research in this domain faces two significant limitations. First, most methods are confined to single-task prediction, failing to capture the interconnected, multi-faceted nature of urban environments where numerous indicators are deeply correlated. Second, the field lacks a standardized experimental benchmark, which severely impedes fair comparison and reproducible progress. To address these challenges, we first establish a comprehensive benchmark for multi-task urban region profiling, featuring multi-modal features and a diverse set of strong baselines to ensure a fair and rigorous evaluation environment. Concurrently, we propose UrbanMoE, the first sparse multi-modal, multi-expert framework specifically architected to solve the multi-task challenge. Leveraging a sparse Mixture-of-Experts architecture, it dynamically routes multi-modal features to specialized sub-networks, enabling the simultaneous prediction of diverse urban indicators. We conduct extensive experiments on three real-world datasets within our benchmark, where UrbanMoE consistently demonstrates superior performance over all baselines. Further in-depth analysis validates the efficacy and efficiency of our approach, setting a new state-of-the-art and providing the community with a valuable tool for future research in urban analytics

The Softmax loss is one of the most widely employed surrogate objectives for classification and ranking tasks. To elucidate its theoretical properties, the Fenchel-Young framework situates it as a canonical instance within a broad family of surrogates. Concurrently, another line of research has addressed scalability when the number of classes is exceedingly large, in which numerous approximations have been proposed to retain the benefits of the exact objective while improving efficiency. Building on these two perspectives, we present a principled investigation of the Softmax-family losses. We examine whether different surrogates achieve consistency with classification and ranking metrics, and analyze their gradient dynamics to reveal distinct convergence behaviors. We also introduce a systematic bias-variance decomposition for approximate methods that provides convergence guarantees, and further derive a per-epoch complexity analysis, showing explicit trade-offs between effectiveness and efficiency. Extensive experiments on a representative task demonstrate a strong alignment between consistency, convergence, and empirical performance. Together, these results establish a principled foundation and offer practical guidance for loss selections in large-class machine learning applications.

Diffusion-based face swapping achieves state-of-the-art performance, yet it also exacerbates the potential harm of malicious face swapping to violate portraiture right or undermine personal reputation. This has spurred the development of proactive defense methods. However, existing approaches face a core trade-off: large perturbations distort facial structures, while small ones weaken protection effectiveness. To address these issues, we propose FaceDefense, an enhanced proactive defense framework against diffusion-based face swapping. Our method introduces a new diffusion loss to strengthen the defensive efficacy of adversarial examples, and employs a directional facial attribute editing to restore perturbation-induced distortions, thereby enhancing visual imperceptibility. A two-phase alternating optimization strategy is designed to generate final perturbed face images. Extensive experiments show that FaceDefense significantly outperforms existing methods in both imperceptibility and defense effectiveness, achieving a superior trade-off.

Legal judgments may contain errors due to the complexity of case circumstances and the abstract nature of legal concepts, while existing appellate review mechanisms face efficiency pressures from a surge in case volumes. Although current legal AI research focuses on tasks like judgment prediction and legal document generation, the task of judgment review differs fundamentally in its objectives and paradigm: it centers on detecting, classifying, and correcting errors after a judgment is issued, constituting anomaly detection rather than prediction or generation. To address this research gap, we introduce a novel task APPELLATE REVIEW, aiming to assess models' diagnostic reasoning and reliability in legal practice. We also construct a novel dataset benchmark AR-BENCH, which comprises 8,700 finely annotated decisions and 34,617 supplementary corpora. By evaluating 14 large language models, we reveal critical limitations in existing models' ability to identify legal application errors, providing empirical evidence for future improvements.

Causal inference from observational data provides strong evidence for the best action in decision-making without performing expensive randomized trials. The effect of an action is usually not identifiable under unobserved confounding, even with an infinite amount of data. Recent work uses neural networks to obtain practical bounds to such causal effects, which is often an intractable problem. However, these approaches may overfit to the dataset and be overconfident in their causal effect estimates. Moreover, there is currently no systematic approach to disentangle how much of the width of causal effect bounds is due to fundamental non-identifiability versus how much is due to finite-sample limitations. We propose a novel framework to address this problem by considering a confidence set around the empirical observational distribution and obtaining the intersection of causal effect bounds for all distributions in this confidence set. This allows us to distinguish the part of the interval that can be reduced by collecting more samples, which we call sample uncertainty, from the part that can only be reduced by observing more variables, such as latent confounders or instrumental variables, but not with more data, which we call non-ID uncertainty. The upper and lower bounds to this intersection are obtained by solving min-max and max-min problems with neural causal models by searching over all distributions that the dataset might have been sampled from, and all SCMs that entail the corresponding distribution. We demonstrate via extensive experiments on synthetic and real-world datasets that our algorithm can determine when collecting more samples will not help determine the best action. This can guide practitioners to collect more variables or lean towards a randomized study for best action identification.

Recent advances in multimodal large language models (MLLMs) mark a shift from non-thinking models to post-trained reasoning models capable of solving complex problems through thinking. However, whether such thinking mitigates hallucinations in multimodal perception and reasoning remains unclear. Self-reflective reasoning enhances robustness but introduces additional hallucinations, and subtle perceptual errors still result in incorrect or coincidentally correct answers. Existing benchmarks primarily focus on models before the emergence of reasoning MLLMs, neglecting the internal thinking process and failing to measure the hallucinations that occur during thinking. To address these challenges, we introduce MM-THEBench, a comprehensive benchmark for assessing hallucinations of intermediate CoTs in reasoning MLLMs. MM-THEBench features a fine-grained taxonomy grounded in cognitive dimensions, diverse data with verified reasoning annotations, and a multi-level automated evaluation framework. Extensive experiments on mainstream reasoning MLLMs reveal insights into how thinking affects hallucination and reasoning capability in various multimodal tasks.

Compressing long chains of thought (CoT) into compact latent tokens is crucial for efficient reasoning with large language models (LLMs). Recent studies employ autoencoders to achieve this by reconstructing textual CoT from latent tokens, thus encoding CoT semantics. However, treating textual CoT as the reconstruction target forces latent tokens to preserve surface-level linguistic features (e.g., word choice and syntax), introducing a strong linguistic inductive bias that prioritizes linguistic form over reasoning structure and limits logical abstraction. Thus, we propose ImgCoT that replaces the reconstruction target from textual CoT to the visual CoT obtained by rendering CoT into images. This substitutes linguistic bias with spatial inductive bias, i.e., a tendency to model spatial layouts of the reasoning steps in visual CoT, enabling latent tokens to better capture global reasoning structure. Moreover, although visual latent tokens encode abstract reasoning structure, they may blur reasoning details. We thus propose a loose ImgCoT, a hybrid reasoning that augments visual latent tokens with a few key textual reasoning steps, selected based on low token log-likelihood. This design allows LLMs to retain both global reasoning structure and fine-grained reasoning details with fewer tokens than the complete CoT. Extensive experiments across multiple datasets and LLMs demonstrate the effectiveness of the two versions of ImgCoT.

Recent advances in diffusion models have significantly elevated the visual fidelity of Virtual Try-On (VTON) systems, yet reliable evaluation remains a persistent bottleneck. Traditional metrics struggle to quantify fine-grained texture details and semantic consistency, while existing datasets fail to meet commercial standards in scale and diversity. We present OpenVTON-Bench, a large-scale benchmark comprising approximately 100K high-resolution image pairs (up to $1536 \times 1536$). The dataset is constructed using DINOv3-based hierarchical clustering for semantically balanced sampling and Gemini-powered dense captioning, ensuring a uniform distribution across 20 fine-grained garment categories. To support reliable evaluation, we propose a multi-modal protocol that measures VTON quality along five interpretable dimensions: background consistency, identity fidelity, texture fidelity, shape plausibility, and overall realism. The protocol integrates VLM-based semantic reasoning with a novel Multi-Scale Representation Metric based on SAM3 segmentation and morphological erosion, enabling the separation of boundary alignment errors from internal texture artifacts. Experimental results show strong agreement with human judgments (Kendall's $τ$ of 0.833 vs. 0.611 for SSIM), establishing a robust benchmark for VTON evaluation.

Accurate molecular geometries are a prerequisite for reliable quantum-chemical predictions, yet density functional theory (DFT) optimization remains a major bottleneck for high-throughput molecular screening. Here we present GeoOpt-Net, a multi-branch SE(3)-equivariant geometry refinement network that predicts DFT-quality structures at the B3LYP/TZVP level of theory in a single forward pass starting from inexpensive initial conformers generated at a low-cost force-field level. GeoOpt-Net is trained using a two-stage strategy in which a broadly pretrained geometric representation is subsequently fine-tuned to approach B3LYP/TZVP-level accuracy, with theory- and basis-set-aware calibration enabled by a fidelity-aware feature modulation (FAFM) mechanism. Benchmarking against representative approaches spanning classical conformer generation (RDKit), semiempirical quantum methods (xTB), data-driven geometry refinement pipelines (Auto3D), and machine-learning interatomic potentials (UMA) on external drug-like molecules demonstrates that GeoOpt-Net achieves sub-milli-Å all-atom RMSD with near-zero B3LYP/TZVP single-point energy deviations, indicating DFT-ready geometries that closely reproduce both structural and energetic references. Beyond geometric metrics, GeoOpt-Net generates initial guesses intrinsically compatible with DFT convergence criteria, yielding nonzero ``All-YES'' convergence rates (65.0\% under loose and 33.4\% under default thresholds), and substantially reducing re-optimization steps and wall-clock time. GeoOpt-Net further exhibits smooth and predictable energy scaling with molecular complexity while preserving key electronic observables such as dipole moments. Collectively, these results establish GeoOpt-Net as a scalable, physically consistent geometry refinement framework that enables efficient acceleration of DFT-based quantum-chemical workflows.

Recent work has found that neural networks with stronger generalization tend to exhibit higher representational alignment with one another across architectures and training paradigms. In this work, we show that models with stronger generalization also align more strongly with human neural activity. Moreover, generalization performance, model--model alignment, and model--brain alignment are all significantly correlated with each other. We further show that these relationships can be explained by a single geometric property of learned representations: the local intrinsic dimension of embeddings. Lower local dimension is consistently associated with stronger model--model alignment, stronger model--brain alignment, and better generalization, whereas global dimension measures fail to capture these effects. Finally, we find that increasing model capacity and training data scale systematically reduces local intrinsic dimension, providing a geometric account of the benefits of scaling. Together, our results identify local intrinsic dimension as a unifying descriptor of representational convergence in artificial and biological systems.

Creating attack paths for cyber defence exercises requires substantial expert effort. Existing automation requires vulnerability graphs or exploit sets curated in advance, limiting where it can be applied. We present AEGIS, a system that generates attack paths using LLMs, white-box access, and Monte Carlo Tree Search over real exploit execution. LLM-based search discovers exploits dynamically without pre-existing vulnerability graphs, while white-box access enables validating exploits in isolation before committing to attack paths. Evaluation at CIDeX 2025, a large-scale exercise spanning 46 IT hosts, showed that AEGIS-generated paths are comparable to human-authored scenarios across four dimensions of training experience (perceived learning, engagement, believability, challenge). Results were measured with a validated questionnaire extensible to general simulation-based training. By automating exploit chain discovery and validation, AEGIS reduces scenario development from months to days, shifting expert effort from technical validation to scenario design.

Reinforcement learning (RL) post-training has increasingly demonstrated strong ability to elicit reasoning behaviors in large language models (LLMs). For training efficiency, rollouts are typically generated in an off-policy manner using an older sampling policy and then used to update the current target policy. To correct the resulting discrepancy between the sampling and target policies, most existing RL objectives rely on a token-level importance sampling ratio, primarily due to its computational simplicity and numerical stability. However, we observe that token-level correction often leads to unstable training dynamics when the degree of off-policyness is large. In this paper, we revisit LLM policy optimization under off-policy conditions and show that the theoretically rigorous correction term is the prefix importance ratio, and that relaxing it to a token-level approximation can induce instability in RL post-training. To stabilize LLM optimization under large off-policy drift, we propose a simple yet effective objective, Minimum Prefix Ratio (MinPRO). MinPRO replaces the unstable cumulative prefix ratio with a non-cumulative surrogate based on the minimum token-level ratio observed in the preceding prefix. Extensive experiments on both dense and mixture-of-experts LLMs, across multiple mathematical reasoning benchmarks, demonstrate that MinPRO substantially improves training stability and peak performance in off-policy regimes.

Current quantization methods for LLMs predominantly rely on block-wise structures to maintain efficiency, often at the cost of representational flexibility. In this work, we demonstrate that element-wise quantization can be made as efficient as block-wise scaling while providing strictly superior expressive power by modeling the scaling manifold as continuous low-rank matrices ($S = BA$). We propose Low-Rank Decomposed Scaling (LoRDS), a unified framework that rethinks quantization granularity through this low-rank decomposition. By "breaking the blocks" of spatial constraints, LoRDS establishes a seamless efficiency lifecycle: it provides high-fidelity PTQ initialization refined via iterative optimization, enables joint QAT of weights and scaling factors, and facilitates high-rank multiplicative PEFT adaptation. Unlike additive PEFT approaches such as QLoRA, LoRDS enables high-rank weight updates within a low-rank budget while incurring no additional inference overhead. Supported by highly optimized Triton kernels, LoRDS consistently outperforms state-of-the-art baselines across various model families in both quantization and downstream fine-tuning tasks. Notably, on Llama3-8B, our method achieves up to a 27.0% accuracy improvement at 3 bits over NormalFloat quantization and delivers a 1.5x inference speedup on NVIDIA RTX 4090 while enhancing PEFT performance by 9.6% on downstream tasks over 4bit QLoRA, offering a robust and integrated solution for unified compression and adaptation of LLMs.

Latent Action Models (LAMs) have rapidly gained traction as an important component in the pre-training pipelines of leading Vision-Language-Action models. However, they fail when observations contain action-correlated distractors, often encoding noise instead of meaningful latent actions. Humans, on the other hand, can effortlessly distinguish task-relevant motions from irrelevant details in any video given only a brief task description. In this work, we propose to utilize the common-sense reasoning abilities of Vision-Language Models (VLMs) to provide promptable representations, effectively separating controllable changes from the noise in unsupervised way. We use these representations as targets during LAM training and benchmark a wide variety of popular VLMs, revealing substantial variation in the quality of promptable representations as well as their robustness to different prompts and hyperparameters. Interestingly, we find that more recent VLMs may perform worse than older ones. Finally, we show that simply asking VLMs to ignore distractors can substantially improve latent action quality, yielding up to a six-fold increase in downstream success rates on Distracting MetaWorld.

Early-exit neural networks have become popular for reducing inference latency by allowing intermediate predictions when sufficient confidence is achieved. However, standard approaches typically rely on single-model confidence thresholds, which are frequently unreliable due to inherent calibration issues. To address this, we introduce SQUAD (Scalable Quorum Adaptive Decisions), the first inference scheme that integrates early-exit mechanisms with distributed ensemble learning, improving uncertainty estimation while reducing the inference time. Unlike traditional methods that depend on individual confidence scores, SQUAD employs a quorum-based stopping criterion on early-exit learners by collecting intermediate predictions incrementally in order of computational complexity until a consensus is reached and halting the computation at that exit if the consensus is statistically significant. To maximize the efficacy of this voting mechanism, we also introduce QUEST (Quorum Search Technique), a Neural Architecture Search method to select early-exit learners with optimized hierarchical diversity, ensuring learners are complementary at every intermediate layer. This consensus-driven approach yields statistically robust early exits, improving the test accuracy up to 5.95% compared to state-of-the-art dynamic solutions with a comparable computational cost and reducing the inference latency up to 70.60% compared to static ensembles while maintaining a good accuracy.

Modern LLMs are increasingly accessed via black-box APIs, requiring users to transmit sensitive prompts, outputs, and fine-tuning data to external providers, creating a critical privacy risk at the API boundary. We introduce AlienLM, a deployable API-only privacy layer that protects text by translating it into an Alien Language via a vocabulary-scale bijection, enabling lossless recovery on the client side. Using only standard fine-tuning APIs, Alien Adaptation Training (AAT) adapts target models to operate directly on alienized inputs. Across four LLM backbones and seven benchmarks, AlienLM retains over 81\% of plaintext-oracle performance on average, substantially outperforming random-bijection and character-level baselines. Under adversaries with access to model weights, corpus statistics, and learning-based inverse translation, recovery attacks reconstruct fewer than 0.22\% of alienized tokens. Our results demonstrate a practical pathway for privacy-preserving LLM deployment under API-only access, substantially reducing plaintext exposure while maintaining task performance.

Vision-Language Models (VLMs) achieve strong multimodal performance but are costly to deploy, and post-training quantization often causes significant accuracy loss. Despite its potential, quantization-aware training for VLMs remains underexplored. We propose GRACE, a framework unifying knowledge distillation and QAT under the Information Bottleneck principle: quantization constrains information capacity while distillation guides what to preserve within this budget. Treating the teacher as a proxy for task-relevant information, we introduce confidence-gated decoupled distillation to filter unreliable supervision, relational centered kernel alignment to transfer visual token structures, and an adaptive controller via Lagrangian relaxation to balance fidelity against capacity constraints. Across extensive benchmarks on LLaVA and Qwen families, our INT4 models consistently outperform FP16 baselines (e.g., LLaVA-1.5-7B: 70.1 vs. 66.8 on SQA; Qwen2-VL-2B: 76.9 vs. 72.6 on MMBench), nearly matching teacher performance. Using real INT4 kernel, we achieve 3$\times$ throughput with 54% memory reduction. This principled framework significantly outperforms existing quantization methods, making GRACE a compelling solution for resource-constrained deployment.

Low-Rank Adaptation (LoRA) is a fundamental parameter-efficient fine-tuning method that balances efficiency and performance in large-scale neural networks. However, the proliferation of LoRA variants has led to fragmentation in methodology, theory, code, and evaluation. To this end, this work presents the first unified study of LoRA variants, offering a systematic taxonomy, unified theoretical review, structured codebase, and standardized empirical assessment. First, we categorize LoRA variants along four principal axes: rank, optimization dynamics, initialization, and integration with Mixture-of-Experts. Then, we review their relationships and evolution within a common theoretical framework focused on low-rank update dynamics. Further, we introduce LoRAFactory, a modular codebase that implements variants through a unified interface, supporting plug-and-play experimentation and fine-grained analysis. Last, using this codebase, we conduct a large-scale evaluation across natural language generation, natural language understanding, and image classification tasks, systematically exploring key hyperparameters. Our results uncover several findings, notably: LoRA and its variants exhibit pronounced sensitivity to the choices of learning rate compared to other hyperparameters; moreover, with proper hyperparameter configurations, LoRA consistently matches or surpasses the performance of most of its variants.

IR-drop is a critical power integrity challenge in modern VLSI designs that can cause timing degradation, reliability issues, and functional failures if not detected early in the design flow. Conventional IR-drop analysis relies on physics-based signoff tools, which provide high accuracy but incur significant computational cost and require near-final layout information, making them unsuitable for rapid early-stage design exploration. In this work, we propose a deep learning-based surrogate modeling approach for early-stage IR-drop estimation using a CNN. The task is formulated as a dense pixel-wise regression problem, where spatial physical layout features are mapped directly to IR-drop heatmaps. A U-Net-based encoder-decoder architecture with skip connections is employed to effectively capture both local and global spatial dependencies within the layout. The model is trained on a physics-inspired synthetic dataset generated by us, which incorporates key physical factors including power grid structure, cell density distribution, and switching activity. Model performance is evaluated using standard regression metrics such as Mean Squared Error (MSE) and Peak Signal-to-Noise Ratio (PSNR). Experimental results demonstrate that the proposed approach can accurately predict IR-drop distributions with millisecond-level inference time, enabling fast pre-signoff screening and iterative design optimization. The proposed framework is intended as a complementary early-stage analysis tool, providing designers with rapid IR-drop insight prior to expensive signoff analysis. The implementation, dataset generation scripts, and the interactive inference application are publicly available at: https://github.com/riteshbhadana/IR-Drop-Predictor. The live application can be accessed at: https://ir-drop-predictor.streamlit.app/.

Large Language Models (LLMs) have demonstrated remarkable capabilities in code generation, but their proficiency in producing secure code remains a critical, under-explored area. Existing benchmarks often fall short by relying on synthetic vulnerabilities or evaluating functional correctness in isolation, failing to capture the complex interplay between functionality and security found in real-world software. To address this gap, we introduce RealSec-bench, a new benchmark for secure code generation meticulously constructed from real-world, high-risk Java repositories. Our methodology employs a multi-stage pipeline that combines systematic SAST scanning with CodeQL, LLM-based false positive elimination, and rigorous human expert validation. The resulting benchmark contains 105 instances grounded in real-word repository contexts, spanning 19 Common Weakness Enumeration (CWE) types and exhibiting a wide diversity of data flow complexities, including vulnerabilities with up to 34-hop inter-procedural dependencies. Using RealSec-bench, we conduct an extensive empirical study on 5 popular LLMs. We introduce a novel composite metric, SecurePass@K, to assess both functional correctness and security simultaneously. We find that while Retrieval-Augmented Generation (RAG) techniques can improve functional correctness, they provide negligible benefits to security. Furthermore, explicitly prompting models with general security guidelines often leads to compilation failures, harming functional correctness without reliably preventing vulnerabilities. Our work highlights the gap between functional and secure code generation in current LLMs.

Batch inference for agentic workloads stresses the GPU key-value (KV) cache in a sustained and cumulative manner, often causing severe throughput degradation well before memory capacity is exhausted. We identify this phenomenon as middle-phase thrashing, a previously under-characterized pathology in which cache efficiency collapses as long-lived agents accumulate state over time. We argue that mitigating this pathology requires moving beyond reactive, request-level cache management to proactive, agent-level admission control. Drawing inspiration from congestion control in distributed systems, we view the KV cache as a shared resource whose efficient utilization depends on feedback-driven regulation. Based on this insight, we present CONCUR, a lightweight control layer that regulates agent admission to bound aggregate cache pressure while preserving execution continuity. CONCUR adapts a cache-aware control algorithm to dynamically adjust the number of active agents using runtime cache signals. Across large models and real-world agent workloads, CONCUR prevents middle-phase thrashing and improves batch inference throughput by up to 4.09x on Qwen3-32B and 1.9x on DeepSeek-V3, while remaining compatible with existing LLM serving systems.

Machine learning (ML) in medicine has transitioned from research to concrete applications aimed at supporting several medical purposes like therapy selection, monitoring and treatment. Acceptance and effective adoption by clinicians and patients, as well as regulatory approval, require evidence of trustworthiness. A major factor for the development of trustworthy AI is the quantification of data quality for AI model training and testing. We have recently proposed the METRIC-framework for systematically evaluating the suitability (fit-for-purpose) of data for medical ML for a given task. Here, we operationalize this theoretical framework by introducing a collection of data quality metrics - the metric library - for practically measuring data quality dimensions. For each metric, we provide a metric card with the most important information, including definition, applicability, examples, pitfalls and recommendations, to support the understanding and implementation of these metrics. Furthermore, we discuss strategies and provide decision trees for choosing an appropriate set of data quality metrics from the metric library given specific use cases. We demonstrate the impact of our approach exemplarily on the PTB-XL ECG-dataset. This is a first step to enable fit-for-purpose evaluation of training and test data in practice as the base for establishing trustworthy AI in medicine.

Vision-Language Models (VLMs) have become powerful backbones for agents to autonomously operate in digital environments like the web and operating systems. However, these models suffer from inadaptability to fast-changing environments like the web, which can be alleviated by fine-tuning requiring expansive model training and data collection. In this work, we introduce a novel paradigm for enhancing agentic VLM policies at inference without policy retraining. Fundamentally, our approach decouples the VLM's role as a high-capacity action proposer from the final action selection mechanism. We keep the VLM policy frozen and use it to generate a set of candidate actions for a given state. Then, a lightweight, offline-trained Q-function reranks these candidates, and the agent executes the action with the highest estimated value. The main contribution is to apply the Q-function directly during inference for immediate policy improvement, and not offline to relabel data for policy retraining. We demonstrate on the academic WebVoyager benchmark that our method significantly boosts agent success rates, improving a Qwen2.5-VL-7B agent from 38.8% to 55.7% and a proprietary GPT-4.1 agent from 82.4% to 88.8%.

Performance of Large Language Models (LLMs) on multiple-choice tasks differs markedly between symbol-based and cloze-style evaluation formats. The observed discrepancies are systematically attributable to task characteristics: natural language continuation benefits from likelihood scoring, whereas explicit comparison is better suited to symbol-based selection. These trends are consistent across various decoder-based LLMs, indicating model-agnostic effects. To address these inconsistencies, a dynamic format-alignment strategy is introduced that employs a lightweight classifier trained on latent model-preference signals. In contrast to human-designed heuristics, which often degrade performance, this approach uses model-generated signals to determine the optimal format for each problem instance. The proposed method achieves substantial and consistent improvements in zero-shot accuracy across reasoning and knowledge benchmarks, better revealing the models' latent capabilities.

Recent vision-language models (VLMs) achieve strong zero-shot performance via large-scale image-text pretraining and have been widely adopted in medical image analysis. However, existing VLMs remain notably weak at understanding negated clinical statements, largely due to contrastive alignment objectives that treat negation as a minor linguistic variation rather than a meaning-inverting operator. In multi-label settings, prompt-based InfoNCE fine-tuning further reinforces easy-positive image-prompt alignments, limiting effective learning of disease absence. To overcome these limitations, we reformulate vision-language alignment as a conditional semantic comparison problem, which is instantiated through a bi-directional multiple-choice learning framework(Bi-MCQ). By jointly training Image-to-Text and Text-to-Image MCQ tasks with affirmative, negative, and mixed prompts, our method implements fine-tuning as conditional semantic comparison instead of global similarity maximization. We further introduce direction-specific Cross-Attention fusion modules to address asymmetric cues required by bi-directional reasoning and reduce alignment interference. Experiments on ChestXray14, Open-I, CheXpert, and PadChest show that Bi-MCQ improves negation understanding by up to 0.47 AUC over the zero-shot performance of the state-of-the-art CARZero model, while achieving up to a 0.08 absolute gain on positive-negative combined (PNC) evaluation. Additionally, Bi-MCQ reduces the affirmative-negative AUC gap by an average of 0.12 compared to InfoNCE-based fine-tuning, demonstrating that objective reformulation can substantially enhance negation understanding in medical VLMs.

Reconstructing detailed 3D human meshes from a single in-the-wild image remains a fundamental challenge in computer vision. Existing SMPLX-based methods often suffer from slow inference, produce only coarse body poses, and exhibit misalignments or unnatural artifacts in fine-grained regions such as the face and hands. These issues make current approaches difficult to apply to downstream tasks. To address these challenges, we propose PEAR-a fast and robust framework for pixel-aligned expressive human mesh recovery. PEAR explicitly tackles three major limitations of existing methods: slow inference, inaccurate localization of fine-grained human pose details, and insufficient facial expression capture. Specifically, to enable real-time SMPLX parameter inference, we depart from prior designs that rely on high resolution inputs or multi-branch architectures. Instead, we adopt a clean and unified ViT-based model capable of recovering coarse 3D human geometry. To compensate for the loss of fine-grained details caused by this simplified architecture, we introduce pixel-level supervision to optimize the geometry, significantly improving the reconstruction accuracy of fine-grained human details. To make this approach practical, we further propose a modular data annotation strategy that enriches the training data and enhances the robustness of the model. Overall, PEAR is a preprocessing-free framework that can simultaneously infer EHM-s (SMPLX and scaled-FLAME) parameters at over 100 FPS. Extensive experiments on multiple benchmark datasets demonstrate that our method achieves substantial improvements in pose estimation accuracy compared to previous SMPLX-based approaches. Project page: https://wujh2001.github.io/PEAR

The development of large language models (LLMs) is costly and has significant commercial value. Consequently, preventing unauthorized appropriation of open-source LLMs and protecting developers' intellectual property rights have become critical challenges. In this work, we propose the Functional Network Fingerprint (FNF), a training-free, sample-efficient method for detecting whether a suspect LLM is derived from a victim model, based on the consistency between their functional network activity. We demonstrate that models that share a common origin, even with differences in scale or architecture, exhibit highly consistent patterns of neuronal activity within their functional networks across diverse input samples. In contrast, models trained independently on distinct data or with different objectives fail to preserve such activity alignment. Unlike conventional approaches, our method requires only a few samples for verification, preserves model utility, and remains robust to common model modifications (such as fine-tuning, pruning, and parameter permutation), as well as to comparisons across diverse architectures and dimensionalities. FNF thus provides model owners and third parties with a simple, non-invasive, and effective tool for protecting LLM intellectual property. The code is available at https://github.com/WhatAboutMyStar/LLM_ACTIVATION.

Large language models (LLMs) based on the Transformer have demonstrated strong performance across diverse tasks. However, current models still exhibit substantial limitations in out-of-distribution (OOD) generalization compared with humans. We investigate this gap through periodicity, one of the basic OOD scenarios. Periodicity captures invariance amid variation. Periodicity generalization represents a model's ability to extract periodic patterns from training data and generalize to OOD scenarios. We introduce a unified interpretation of periodicity from the perspective of abstract algebra and reasoning, including both single and composite periodicity, to explain why Transformers struggle to generalize periodicity. Then we construct Coper about composite periodicity, a controllable generative benchmark with two OOD settings, Hollow and Extrapolation. Experiments reveal that periodicity generalization in Transformers is limited, where models can memorize periodic data during training, but cannot generalize to unseen composite periodicity. We release the source code to support future research.

Language models generate reasoning sequentially, preventing them from decoupling irrelevant exploration paths during search. We introduce Tree-Structured Language Modeling (TSLM), which uses special tokens to encode branching structure, enabling models to generate and selectively expand multiple search paths within a single generation process. By training on complete search trees including both successful and failed attempts, TSLM learns to internalize systematic exploration without redundant recomputation of shared prefixes. TSLM achieves robust performance and superior inference efficiency by avoiding the multiple independent forward passes required by external search methods. These results suggest a new paradigm of inference-time scaling for robust reasoning, demonstrating that supervised learning on complete tree-structured traces provides an efficient alternative for developing systematic exploration capabilities in language models.

Consistency models have been proposed for fast generative modeling, achieving results competitive with diffusion and flow models. However, these methods exhibit inherent instability and limited reproducibility when training from scratch, motivating subsequent work to explain and stabilize these issues. While these efforts have provided valuable insights, the explanations remain fragmented, and the theoretical relationships remain unclear. In this work, we provide a theoretical examination of consistency models by analyzing them from a flow map-based perspective. This joint analysis clarifies how training stability and convergence behavior can give rise to degenerate solutions. Building on these insights, we revisit self-distillation as a practical remedy for certain forms of suboptimal convergence and reformulate it to avoid excessive gradient norms for stable optimization. We further demonstrate that our strategy extends beyond image generation to diffusion-based policy learning, without reliance on a pretrained diffusion model for initialization, thereby illustrating its broader applicability.

Full-graph and mini-batch Graph Neural Network (GNN) training approaches have distinct system design demands, making it crucial to choose the appropriate approach to develop. A core challenge in comparing these two GNN training approaches lies in characterizing their model performance (i.e., convergence and generalization) and computational efficiency. While a batch size has been an effective lens in analyzing such behaviors in deep neural networks (DNNs), GNNs extend this lens by introducing a fan-out size, as full-graph training can be viewed as mini-batch training with the largest possible batch size and fan-out size. However, the impact of the batch and fan-out size for GNNs remains insufficiently explored. To this end, this paper systematically compares full-graph vs. mini-batch training of GNNs through empirical and theoretical analyses from the view points of the batch size and fan-out size. Our key contributions include: 1) We provide a novel generalization analysis using the Wasserstein distance to study the impact of the graph structure, especially the fan-out size. 2) We uncover the non-isotropic effects of the batch size and the fan-out size in GNN convergence and generalization, providing practical guidance for tuning these hyperparameters under resource constraints. Finally, full-graph training does not always yield better model performance or computational efficiency than well-tuned smaller mini-batch settings. The implementation can be found in the github link: https://github.com/LIUMENGFAN-gif/GNN_fullgraph_minibatch_training.

Logs serve as a primary source of information for engineers to diagnose failures in large-scale online service systems. Log parsing, which extracts structured events from massive unstructured log data, is a critical first step for downstream tasks like anomaly detection and failure diagnosis. With advances in large language models (LLMs), leveraging their strong text understanding capabilities has proven effective for accurate log parsing. However, existing LLM-based log parsers all focus on the constant part of logs, ignoring the potential contribution of the variable part to log parsing. This constant-centric strategy brings four key problems. First, inefficient log grouping and sampling with only constant information. Second, a relatively large number of LLM invocations due to constant-based cache, leading to low log parsing accuracy and efficiency. Third, a relatively large number of consumed constant tokens in prompts leads to high LLM invocation costs. At last, these methods only retain placeholders in the results, losing the system visibility brought by variable information in logs. Facing these problems, we propose a variable-centric log parsing strategy named VarParser. Through variable contribution sampling, variable-centric parsing cache, and adaptive variable-aware in-context learning, our approach can efficiently capture the variable parts of logs and leverage their contributions to parsing. By introducing variable units, we preserve rich variable information, enhancing the integrity of log parsing results. Extensive evaluations on large-scale datasets demonstrate that VarParser achieves higher accuracy compared to existing methods, significantly improving parsing efficiency while reducing the LLM invocation costs.

Spiking neural networks (SNNs) have gained traction in vision due to their energy efficiency, bio-plausibility, and inherent temporal processing. Yet, despite this temporal capacity, most progress concentrates on static image benchmarks, and SNNs still underperform on dynamic video tasks compared to artificial neural networks (ANNs). In this work, we diagnose a fundamental pass-band mismatch: Standard spiking dynamics behave as a temporal low pass that emphasizes static content while attenuating motion bearing bands, where task relevant information concentrates in dynamic tasks. This phenomenon explains why SNNs can approach ANNs on static tasks yet fall behind on tasks that demand richer temporal understanding.To remedy this, we propose the Pass-Bands Optimizer (PBO), a plug-and-play module that optimizes the temporal pass-band toward task-relevant motion bands. PBO introduces only two learnable parameters, and a lightweight consistency constraint that preserves semantics and boundaries, incurring negligible computational overhead and requires no architectural changes. PBO deliberately suppresses static components that contribute little to discrimination, effectively high passing the stream so that spiking activity concentrates on motion bearing content. On UCF101, PBO yields over ten percentage points improvement. On more complex multi-modal action recognition and weakly supervised video anomaly detection, PBO delivers consistent and significant gains, offering a new perspective for SNN based video processing and understanding.

Federated Learning (FL) facilitates decentralized collaborative learning without transmitting raw data. However, reliance on fixed global rounds or validation data for hyperparameter tuning hinders practical deployment by incurring high computational costs and privacy risks. To address this, we propose a data-free early stopping framework that determines the optimal stopping point by monitoring the task vector's growth rate using solely server-side parameters. The numerical results on skin lesion/blood cell classification demonstrate that our approach is comparable to validation-based early stopping across various state-of-the-art FL methods. In particular, the proposed framework spends an average of 47/20 (skin lesion/blood cell) rounds to achieve over 12.5%/10.3% higher performance than early stopping based on validation data. To the best of our knowledge, this is the first work to propose an early stopping framework for FL methods without using any validation data.

This paper examines the organizational implications of Generative AI adoption in software engineering through a multiple-case comparative study. We contrast two development environments: a traditional enterprise (brownfield) and an AI-native startup (greenfield). Our analysis reveals that transitioning from Horizontal Layering (functional specialization) to Vertical Integration (end-to-end ownership) yields 8-fold to 33-fold reductions in resource consumption. We attribute these gains to the emergence of Super Employees, AI-augmented engineers who span traditional role boundaries, and the elimination of inter-functional coordination overhead. Theoretically, we propose Human-AI Collaboration Efficacy as the primary optimization target for engineering organizations, supplanting individual productivity metrics. Our Total Factor Productivity analysis identifies an AI Distortion Effect that diminishes returns to labor scale while amplifying technological leverage. We conclude with managerial strategies for organizational redesign, including the reactivation of idle cognitive bandwidth in senior engineers and the suppression of blind scale expansion.

Reinforcement Learning from Human Feedback (RLHF) is a pivotal technique for aligning large language models (LLMs) with human preferences, yet it is susceptible to reward overoptimization, in which policy models overfit to the reward model, exploit spurious reward patterns instead of faithfully capturing human intent. Prior mitigations primarily relies on surface semantic information and fails to efficiently address the misalignment between the reward model (RM) and the policy model caused by continuous policy distribution shifts. This inevitably leads to an increasing reward discrepancy, exacerbating reward overoptimization. To address these limitations, we introduce R2M (Real-Time Aligned Reward Model), a novel lightweight RLHF framework. R2M goes beyond vanilla reward models that solely depend on the semantic representations of a pretrained LLM. Instead, it leverages the evolving hidden states of the policy (namely policy feedback) to align with the real-time distribution shift of the policy during the RL process. This work points to a promising new direction for improving the performance of reward models through real-time utilization of feedback from policy models.

As the quality of synthetic images improves, identifying the underlying concepts of model-generated images is becoming increasingly crucial for copyright protection and ensuring model transparency. Existing methods achieve this attribution goal by training models using annotated pairs of synthetic images and their original training sources. However, obtaining such paired supervision is challenging, as it requires either well-designed synthetic concepts or precise annotations from millions of training sources. To eliminate the need for costly paired annotations, in this paper, we explore the possibility of unsupervised synthetic image attribution. We propose a simple yet effective unsupervised method called Alignment and Disentanglement. Specifically, we begin by performing basic concept alignment using contrastive self-supervised learning. Next, we enhance the model's attribution ability by promoting representation disentanglement with the Infomax loss. This approach is motivated by an interesting observation: contrastive self-supervised models, such as MoCo and DINO, inherently exhibit the ability to perform simple cross-domain alignment. By formulating this observation as a theoretical assumption on cross-covariance, we provide a theoretical explanation of how alignment and disentanglement can approximate the concept-matching process through a decomposition of the canonical correlation analysis objective. On the real-world benchmarks, AbC, we show that our unsupervised method surprisingly outperforms the supervised methods. As a starting point, we expect our intuitive insights and experimental findings to provide a fresh perspective on this challenging task.

Large language models (LLMs) have shown strong capabilities across diverse decision-making tasks. However, existing approaches often overlook the specialization differences among available models, treating all LLMs as uniformly applicable regardless of task characteristics. This limits their ability to adapt to varying reasoning demands and task complexities. In this work, we propose Task-Aware LLM Council (TALC), a task-adaptive decision framework that integrates a council of LLMs with Monte Carlo Tree Search (MCTS) to enable dynamic expert selection and efficient multi-step planning. Each LLM is equipped with a structured success memory profile derived from prior task trajectories, enabling semantic matching between current reasoning context and past successes. At each decision point, TALC routes control to the most contextually appropriate model and estimates node value using a dual-signal mechanism that fuses model-based evaluations with historical utility scores. These signals are adaptively weighted based on intra-node variance and used to guide MCTS selection, allowing the system to balance exploration depth with planning confidence. Experiments on WebShop, HumanEval, and the Game of 24 demonstrate that TALC achieves superior task success rates and improved search efficiency compared to strong baselines, validating the benefits of specialization-aware routing and adaptive planning.

We investigate progressive freezing as an alternative to straight-through estimators (STE) for training binary networks from scratch. Under controlled training conditions, we find that while global progressive freezing works for binary-weight networks, it fails for full binary neural networks due to activation-induced gradient blockades. We introduce StoMPP (Stochastic Masked Partial Progressive Binarization), which uses layerwise stochastic masking to progressively replace differentiable clipped weights/activations with hard binary step functions, while only backpropagating through the unfrozen (clipped) subset (i.e., no straight-through estimator). Under a matched minimal training recipe, StoMPP improves accuracy over a BinaryConnect-style STE baseline, with gains that increase with depth (e.g., for ResNet-50 BNN: +18.0 on CIFAR-10, +13.5 on CIFAR-100, and +3.8 on ImageNet; for ResNet-18: +3.1, +4.7, and +1.3). For binary-weight networks, StoMPP achieves 91.2\% accuracy on CIFAR-10 and 69.5\% on CIFAR-100 with ResNet-50. We analyze training dynamics under progressive freezing, revealing non-monotonic convergence and improved depth scaling under binarization constraints.

Prevailing methods for integrating graphs into Language Models (LMs) typically rely on a segregated architecture: external Graph Neural Networks (GNNs) encode structural topology, while LMs process textual semantics. We argue this approach is suboptimal for text-graphs: it creates a conceptually disjointed interaction paradigm. By segregating structural encoding from semantic processing, these systems must perform a complex implicit alignment between abstract graph tokens and concrete textual elements. Challenging the necessity of external encoders, we propose NAG (Native Architecture for Graphs), a unified framework that internalizes graph processing within the LM's native manifold. Instead of bridging disparate embedding spaces, NAG repurposes the self-attention mechanism to enforce topological dependencies and recalibrates positional IDs to ensure structural equivalence. This allows the model to harness its intrinsic linguistic capability to simultaneously comprehend node and edge content alongside structural topology. We introduce two efficient implementations: NAG-Zero for absolute preservation of the base model's linguistic capabilities, and NAG-LoRA for enhanced structural adaptation. Experiments across diverse graph tasks validate that NAG achieves robust graph comprehension without the overhead of external encoders, offering a simpler, more coherent paradigm for text-graph modeling.

LLMs demonstrate promising performance in software vulnerability detection after fine-tuning. However, it remains unclear whether these gains reflect a genuine understanding of vulnerability root causes or merely an exploitation of functional patterns. In this paper, we identify a critical failure mode termed the "semantic trap," where fine-tuned LLMs achieve high detection scores by associating certain functional domains with vulnerability likelihood rather than reasoning about the underlying security semantics.To systematically evaluate this phenomenon, we propose TrapEval, a comprehensive evaluation framework designed to disentangle vulnerability root cause from functional pattern. TrapEval introduces two complementary datasets derived from real-world open-source projects: V2N, which pairs vulnerable code with unrelated benign code, and V2P, which pairs vulnerable code with its corresponding patched version, forcing models to distinguish near-identical code that differs only in subtle security-critical logic. Using TrapEval, we fine-tune five representative state-of-the-art LLMs across three model families and evaluate them under cross-dataset testing, semantic-preserving perturbations, and varying degrees of semantic gap measured by CodeBLEU.Our empirical results reveal that, despite improvements in metrics, fine-tuned LLMs consistently struggle to distinguish vulnerable code from its patched counterpart, exhibit severe robustness degradation under minor semantic-preserving transformations, and rely heavily on functional-context shortcuts when the semantic gap is small. These findings provide strong evidence that current fine-tuning practices often fail to impart true vulnerability reasoning. Our findings serve as a wake-up call: high benchmark scores on traditional datasets may be illusory, masking the model's inability to understand the true causal logic of vulnerabilities.

Artificial intelligence (AI) copilots are increasingly integrated into enterprise cybersecurity platforms to assist analysts in threat detection, triage, and remediation. However, the effectiveness of these systems depends not only on the accuracy of underlying models but also on the degree to which users can understand and trust their outputs. Existing research on algorithmic explainability has largely focused on model internals, while little attention has been given to how explanations should be surfaced in user interfaces for high-stakes decision-making contexts [8], [5], [6]. We present a mixed-methods study of explanation design strategies in AI-driven security dashboards. Through a taxonomy of explanation styles and a controlled user study with security practitioners, we compare natural language rationales, confidence visualizations, counterfactual explanations, and hybrid approaches. Our findings show that explanation style significantly affects user trust calibration, decision accuracy, and cognitive load. We contribute (1) empirical evidence on the usability of explanation interfaces for security copilots, (2) design guidelines for integrating explainability into enterprise UIs, and (3) a framework for aligning explanation strategies with analyst needs in security operations centers (SOCs). This work advances the design of human-centered AI tools in cybersecurity and provides broader implications for explainability in other high-stakes domains.

Spectral gradient methods, such as the Muon optimizer, modify gradient updates by preserving directional information while discarding scale, and have shown strong empirical performance in deep learning. We investigate the mechanisms underlying these gains through a dynamical analysis of a nonlinear phase retrieval model with anisotropic Gaussian inputs, equivalent to training a two-layer neural network with the quadratic activation and fixed second-layer weights. Focusing on a spiked covariance setting where the dominant variance direction is orthogonal to the signal, we show that gradient descent (GD) suffers from a variance-induced misalignment: during the early escaping stage, the high-variance but uninformative spike direction is multiplicatively amplified, degrading alignment with the true signal under strong anisotropy. In contrast, spectral gradient descent (SpecGD) removes this spike amplification effect, leading to stable alignment and accelerated noise contraction. Numerical experiments confirm the theory and show that these phenomena persist under broader anisotropic covariances.

Training-data attribution for vision generative models aims to identify which training data influenced a given output. While most methods score individual examples, practitioners often need group-level answers (e.g., artistic styles or object classes). Group-wise attribution is counterfactual: how would a model's behavior on a generated sample change if a group were absent from training? A natural realization of this counterfactual is Leave-One-Group-Out (LOGO) retraining, which retrains the model with each group removed; however, it becomes computationally prohibitive as the number of groups grows. We propose GUDA (Group Unlearning-based Data Attribution) for diffusion models, which approximates each counterfactual model by applying machine unlearning to a shared full-data model instead of training from scratch. GUDA quantifies group influence using differences in a likelihood-based scoring rule (ELBO) between the full model and each unlearned counterfactual. Experiments on CIFAR-10 and artistic style attribution with Stable Diffusion show that GUDA identifies primary contributing groups more reliably than semantic similarity, gradient-based attribution, and instance-level unlearning approaches, while achieving x100 speedup on CIFAR-10 over LOGO retraining.

The inference of conditional distributions is a fundamental problem in statistics, essential for prediction, uncertainty quantification, and probabilistic modeling. A wide range of methodologies have been developed for this task. This article reviews and compares several representative approaches spanning classical nonparametric methods and modern generative models. We begin with the single-index method of Hall and Yao (2005), which estimates the conditional distribution through a dimension-reducing index and nonparametric smoothing of the resulting one-dimensional cumulative conditional distribution function. We then examine the basis-expansion approaches, including FlexCode (Izbicki and Lee, 2017) and DeepCDE (Dalmasso et al., 2020), which convert conditional density estimation into a set of nonparametric regression problems. In addition, we discuss two recent generative simulation-based methods that leverage modern deep generative architectures: the generative conditional distribution sampler (Zhou et al., 2023) and the conditional denoising diffusion probabilistic model (Fu et al., 2024; Yang et al., 2025). A systematic numerical comparison of these approaches is provided using a unified evaluation framework that ensures fairness and reproducibility. The performance metrics used for the estimated conditional distribution include the mean-squared errors of conditional mean and standard deviation, as well as the Wasserstein distance. We also discuss their flexibility and computational costs, highlighting the distinct advantages and limitations of each approach.

The key to building trustworthy Large Language Models (LLMs) lies in endowing them with inherent uncertainty expression capabilities to mitigate the hallucinations that restrict their high-stakes applications. However, existing RL paradigms such as GRPO often suffer from Advantage Bias due to binary decision spaces and static uncertainty rewards, inducing either excessive conservatism or overconfidence. To tackle this challenge, this paper unveils the root causes of reward hacking and overconfidence in current RL paradigms incorporating uncertainty-based rewards, based on which we propose the UnCertainty-Aware Policy Optimization (UCPO) framework. UCPO employs Ternary Advantage Decoupling to separate and independently normalize deterministic and uncertain rollouts, thereby eliminating advantage bias. Furthermore, a Dynamic Uncertainty Reward Adjustment mechanism is introduced to calibrate uncertainty weights in real-time according to model evolution and instance difficulty. Experimental results in mathematical reasoning and general tasks demonstrate that UCPO effectively resolves the reward imbalance, significantly improving the reliability and calibration of the model beyond their knowledge boundaries.

Language model (LM)-based embodied agents are increasingly deployed in real-world settings. Yet, their adaptability remains limited in dynamic environments, where constructing accurate and flexible world models is crucial for effective reasoning and decision-making. To address this challenge, we extend the Mixture-of-Experts (MoE) paradigm to embodied agents. While conventional MoE architectures modularize knowledge into expert components with pre-trained routing, they remain rigid once deployed, making them less effective for adapting to unseen domains in dynamic environments. We therefore propose Test-time Mixture of World Models (TMoW), a framework that enhances adaptability to unseen and evolving domains. TMoW updates its routing function over world models at test time, unlike conventional MoE where the function remains fixed, enabling agents to recombine existing models and integrate new ones for continual adaptation. It achieves this through (i) multi-granular prototype-based routing, which adapts mixtures across object- to scene-level similarities, (ii) test-time refinement that aligns unseen domain features with prototypes during inference, and (iii) distilled mixture-based augmentation, which efficiently constructs new models from few-shot data and existing prototypes. We evaluate TMoW on VirtualHome, ALFWorld, and RLBench benchmarks, demonstrating strong performance in both zero-shot adaptation and few-shot expansion scenarios, and showing that it enables embodied agents to operate effectively in dynamic environments.

Generative medical AI now appears fluent and knowledgeable enough to resemble clinical intelligence, encouraging the belief that scaling will make it safe. But clinical reasoning is not text generation. It is a responsibility-bound process under ambiguity, incomplete evidence, and longitudinal context. Even as benchmark scores rise, generation-centric systems still show behaviours incompatible with clinical deployment: premature closure, unjustified certainty, intent drift, and instability across multi-step decisions. We argue these are structural consequences of treating medicine as next-token prediction. We formalise Clinical Contextual Intelligence (CCI) as a distinct capability class required for real-world clinical use, defined by persistent context awareness, intent preservation, bounded inference, and principled deferral when evidence is insufficient. We introduce Meddollina, a governance-first clinical intelligence system designed to constrain inference before language realisation, prioritising clinical appropriateness over generative completeness. Meddollina acts as a continuous intelligence layer supporting clinical workflows while preserving clinician authority. We evaluate Meddollina using a behaviour-first regime across 16,412+ heterogeneous medical queries, benchmarking against general-purpose models, medical-tuned models, and retrieval-augmented systems. Meddollina exhibits a distinct behavioural profile: calibrated uncertainty, conservative reasoning under underspecification, stable longitudinal constraint adherence, and reduced speculative completion relative to generation-centric baselines. These results suggest deployable medical AI will not emerge from scaling alone, motivating a shift toward Continuous Clinical Intelligence, where progress is measured by clinician-aligned behaviour under uncertainty rather than fluency-driven completion.

Large Language Models (LLMs) show remarkable capabilities, yet their stochastic next-token prediction creates logical inconsistencies and reward hacking that formal symbolic systems avoid. To bridge this gap, we introduce a formal logic verification-guided framework that dynamically interleaves formal symbolic verification with the natural language generation process, providing real-time feedback to detect and rectify errors as they occur. Distinguished from previous neuro-symbolic methods limited by passive post-hoc validation, our approach actively penalizes intermediate fallacies during the reasoning chain. We operationalize this framework via a novel two-stage training pipeline that synergizes formal logic verification-guided supervised fine-tuning and policy optimization. Extensive evaluation on six benchmarks spanning mathematical, logical, and general reasoning demonstrates that our 7B and 14B models outperform state-of-the-art baselines by average margins of 10.4% and 14.2%, respectively. These results validate that formal verification can serve as a scalable mechanism to significantly push the performance boundaries of advanced LLM reasoning.

Automated peer review has evolved from simple text classification to structured feedback generation. However, current state-of-the-art systems still struggle with "surface-level" critiques: they excel at summarizing content but often fail to accurately assess novelty and significance or identify deep methodological flaws because they evaluate papers in a vacuum, lacking the external context a human expert possesses. In this paper, we introduce ScholarPeer, a search-enabled multi-agent framework designed to emulate the cognitive processes of a senior researcher. ScholarPeer employs a dual-stream process of context acquisition and active verification. It dynamically constructs a domain narrative using a historian agent, identifies missing comparisons via a baseline scout, and verifies claims through a multi-aspect Q&A engine, grounding the critique in live web-scale literature. We evaluate ScholarPeer on DeepReview-13K and the results demonstrate that ScholarPeer achieves significant win-rates against state-of-the-art approaches in side-by-side evaluations and reduces the gap to human-level diversity.

Gliomas are placing an increasingly clinical burden on Sub-Saharan Africa (SSA). In the region, the median survival for patients remains under two years, and access to diagnostic imaging is extremely limited. These constraints highlight an urgent need for automated tools that can extract the maximum possible information from each available scan, tools that are specifically trained on local data, rather than adapted from high-income settings where conditions are vastly different. We utilize the Brain Tumor Segmentation (BraTS) Africa 2025 Challenge dataset, an expert annotated collection of glioma MRIs. Our objectives are: (i) establish a strong baseline with nnUNet on this dataset, and (ii) explore whether the celebrated "grokking" phenomenon an abrupt, late training jump from memorization to superior generalization can be triggered to push performance without extra labels. We evaluate two training regimes. The first is a fast, budget-conscious approach that limits optimization to just a few epochs, reflecting the constrained GPU resources typically available in African institutions. Despite this limitation, nnUNet achieves strong Dice scores: 92.3% for whole tumor (WH), 86.6% for tumor core (TC), and 86.3% for enhancing tumor (ET). The second regime extends training well beyond the point of convergence, aiming to trigger a grokking-driven performance leap. With this approach, we were able to achieve grokking and enhanced our results to higher Dice scores: 92.2% for whole tumor (WH), 90.1% for tumor core (TC), and 90.2% for enhancing tumor (ET).

Large Language Models (LLMs) are typically evaluated for safety under single-shot or low-budget adversarial prompting, which underestimates real-world risk. In practice, attackers can exploit large-scale parallel sampling to repeatedly probe a model until a harmful response is produced. While recent work shows that attack success increases with repeated sampling, principled methods for predicting large-scale adversarial risk remain limited. We propose a scaling-aware Best-of-N estimation of risk, SABER, for modeling jailbreak vulnerability under Best-of-N sampling. We model sample-level success probabilities using a Beta distribution, the conjugate prior of the Bernoulli distribution, and derive an analytic scaling law that enables reliable extrapolation of large-N attack success rates from small-budget measurements. Using only n=100 samples, our anchored estimator predicts ASR@1000 with a mean absolute error of 1.66, compared to 12.04 for the baseline, which is an 86.2% reduction in estimation error. Our results reveal heterogeneous risk scaling profiles and show that models appearing robust under standard evaluation can experience rapid nonlinear risk amplification under parallel adversarial pressure. This work provides a low-cost, scalable methodology for realistic LLM safety assessment. We will release our code and evaluation scripts upon publication to future research.

The Semantic Gap Problem (SGP) in Computer Vision (CV) arises from the misalignment between visual and lexical semantics leading to flawed CV dataset design and CV benchmarks. This paper proposes that classification principles of S.R. Ranganathan can offer a principled starting point to address SGP and design high-quality CV datasets. We elucidate how these principles, suitably adapted, underpin the vTelos CV annotation methodology. The paper also briefly presents experimental evidence showing improvements in CV annotation and accuracy, thereby, validating vTelos.

The integration of Large Language Models (LLMs) into network operations (AIOps) is hindered by two fundamental challenges: the stochastic grounding problem, where LLMs struggle to reliably parse unstructured, vendor-specific CLI output, and the security gap of granting autonomous agents shell access. This paper introduces MCP-Diag, a hybrid neuro-symbolic architecture built upon the Model Context Protocol (MCP). We propose a deterministic translation layer that converts raw stdout from canonical utilities (dig, ping, traceroute) into rigorous JSON schemas before AI ingestion. We further introduce a mandatory "Elicitation Loop" that enforces Human-in-the-Loop (HITL) authorization at the protocol level. Our preliminary evaluation demonstrates that MCP-Diag achieving 100% entity extraction accuracy with less than 0.9% execution latency overhead and 3.7x increase in context token usage.

Large Language Models (LLMs) exhibit substantial parameter redundancy, particularly in Feed-Forward Networks (FFNs). Existing pruning methods suffer from two primary limitations. First, reliance on dataset-specific calibration introduces significant data dependency and computational overhead. Second, being predominantly static, they fail to account for the evolving subset of knowledge neurons in LLMs during autoregressive generation as the context evolves. To address this, we introduce DART, i.e., Dynamic Attention-Guided Runtime Tracing), a lightweight, training-free method that performs on-the-fly context-based pruning. DART monitors shifts in attention score distributions to infer context changes, dynamically updating neuron-level masks to retain salient parameters. Across ten benchmarks, DART outperforms prior dynamic baseline, achieving accuracy gains of up to 14.5% on LLAMA-3.1-8B at 70% FFN sparsity. Furthermore, DART achieves up to 3x better ROUGE-L scores with respect to static-masked pruning on summarization tasks, with its performance comparable to the original dense models. We conclusively demonstrate that the proposed framework effectively adapts to diverse semantic contexts, preserves model capabilities across both general and domain-specific tasks while running at less than 10MBs of memory for LLAMA-3.1-8B(16GBs) with 0.1% FLOPs overhead. The code is available at https://github.com/seeder-research/DART.

The application of data-driven remaining useful life (RUL) prediction has long been constrained by the availability of large amount of degradation data. Mainstream solutions such as domain adaptation and meta-learning still rely on large amounts of historical degradation data from equipment that is identical or similar to the target, which imposes significant limitations in practical applications. This study investigates PEFT-MuTS, a Parameter-Efficient Fine-Tuning framework for few-shot RUL prediction, built on cross-domain pre-trained time-series representation models. Contrary to the widely held view that knowledge transfer in RUL prediction can only occur within similar devices, we demonstrate that substantial benefits can be achieved through pre-training process with large-scale cross-domain time series datasets. A independent feature tuning network and a meta-variable-based low rank multivariate fusion mechanism are developed to enable the pre-trained univariate time-series representation backbone model to fully exploit the multivariate relationships in degradation data for downstream RUL prediction task. Additionally, we introduce a zero-initialized regressor that stabilizes the fine-tuning process under few-shot conditions. Experiments on aero-engine and industrial bearing datasets demonstrate that our method can achieve effective RUL prediction even when less than 1\% of samples of target equipment are used. Meanwhile, it substantially outperforms conventional supervised and few-shot approaches while markedly reducing the data required to achieve high predictive accuracy. Our code is available at https://github.com/fuen1590/PEFT-MuTS.

Diffusion language models (Diffusion-LMs) introduce an explicit temporal dimension into text generation, yet how this structure can be leveraged to control generation diversity for exploring multiple valid semantic or reasoning paths remains underexplored. In this paper, we show that Diffusion-LMs, like diffusion models in image generation, exhibit a temporal division of labor: early denoising steps largely determine the global semantic structure, while later steps focus on local lexical refinement. Building on this insight, we propose Time-Annealed Perturbation Sampling (TAPS), a training-free inference strategy that encourages semantic branching early in the diffusion process while progressively reducing perturbations to preserve fluency and instruction adherence. TAPS is compatible with both non-autoregressive and semi-autoregressive Diffusion backbones, demonstrated on LLaDA and TraDo in our paper, and consistently improves output diversity across creative writing and reasoning benchmarks without compromising generation quality.

Test-Time Training offers a promising way to improve the reasoning ability of large language models (LLMs) by adapting the model using only the test questions. However, existing methods struggle with difficult reasoning problems for two reasons: raw test questions are often too difficult to yield high-quality pseudo-labels, and the limited size of test sets makes continuous online updates prone to instability. To address these limitations, we propose TTCS, a co-evolving test-time training framework. Specifically, TTCS initializes two policies from the same pretrained model: a question synthesizer and a reasoning solver. These policies evolve through iterative optimization: the synthesizer generates progressively challenging question variants conditioned on the test questions, creating a structured curriculum tailored to the solver's current capability, while the solver updates itself using self-consistency rewards computed from multiple sampled responses on both original test and synthetic questions. Crucially, the solver's feedback guides the synthesizer to generate questions aligned with the model's current capability, and the generated question variants in turn stabilize the solver's test-time training. Experiments show that TTCS consistently strengthens the reasoning ability on challenging mathematical benchmarks and transfers to general-domain tasks across different LLM backbones, highlighting a scalable path towards dynamically constructing test-time curricula for self-evolving. Our code and implementation details are available at https://github.com/XMUDeepLIT/TTCS.

Digital health (DH) software is increasingly deployed to populations where many end users live with one or more health conditions. Yet, DH software development teams frequently operate using implicit, incorrect assumptions about these users, resulting in products that under-serve the specific requirements imposed by their age and health conditions. Consequently, while software may meet clinical objectives on paper, it often fails to be inclusive during actual user interaction. To address this, we propose \textbf{\textit{HealthMag}}, a tool inspired by GenderMag designed to help better elicit, model and evaluate requirements for digital health software. We developed HealthMag through systematic mapping and calibration following the InclusiveMag framework. Furthermore, we integrated this with a calibrated version of an existing AgeMag method to create a dual-lens approach: \textbf{\textit{Elderly HealthMag}}, designed to aid requirements, design and evaluation of mHealth software for senior end users. We demonstrate application and utility of Age HealthMag via cognitive walkthroughs in identifying inclusivity biases in current senior user-oriented digital health applications.

With the wide application of machine learning techniques in practice, privacy preservation has gained increasing attention. Protecting user privacy with minimal accuracy loss is a fundamental task in the data analysis and mining community. In this paper, we focus on regression tasks under $ε$-label differential privacy guarantees. Some existing methods for regression with $ε$-label differential privacy, such as the RR-On-Bins mechanism, discretized the output space into finite bins and then applied RR algorithm. To efficiently determine these finite bins, the authors rounded the original responses down to integer values. However, such operations does not align well with real-world scenarios. To overcome these limitations, we model both original and randomized responses as continuous random variables, avoiding discretization entirely. Our novel approach estimates an optimal interval for randomized responses and introduces new algorithms designed for scenarios where a prior is either known or unknown. Additionally, we prove that our algorithm, RPWithPrior, guarantees $ε$-label differential privacy. Numerical results demonstrate that our approach gets better performance compared with the Gaussian, Laplace, Staircase, and RRonBins, Unbiased mechanisms on the Communities and Crime, Criteo Sponsored Search Conversion Log, California Housing datasets.

The optimization problems in realistic world present significant challenges onto optimization algorithms, such as the expensive evaluation issue and complex constraint conditions. COBRA optimizer (including its up-to-date variants) is a representative and effective tool for addressing such optimization problems, which introduces 1) RBF surrogate to reduce online evaluation and 2) bi-stage optimization process to alternate search for feasible solution and optimal solution. Though promising, its design space, i.e., surrogate model pool and selection standard, is still manually decided by human expert, resulting in labor-intensive fine-tuning for novel tasks. In this paper, we propose a learning-based adaptive strategy (COBRA++) that enhances COBRA in two aspects: 1) An augmented surrogate pool to break the tie with RBF-like surrogate and hence enhances model diversity and approximation capability; 2) A reinforcement learning-based online model selection policy that empowers efficient and accurate optimization process. The model selection policy is trained to maximize overall performance of COBRA++ across a distribution of constrained optimization problems with diverse properties. We have conducted multi-dimensional validation experiments and demonstrate that COBRA++ achieves substantial performance improvement against vanilla COBRA and its adaptive variant. Ablation studies are provided to support correctness of each design component in COBRA++.

Recent advancements have increasingly focused on leveraging large language models (LLMs) to construct autonomous agents for complex problem-solving tasks. However, existing approaches predominantly employ a single-agent framework to generate search branches and estimate rewards during Monte Carlo Tree Search (MCTS) planning. This single-agent paradigm inherently limits exploration capabilities, often resulting in insufficient diversity among generated branches and suboptimal planning performance. To overcome these limitations, we propose Synergistic Multi-agent Planning with Heterogeneous langauge model assembly (SYMPHONY), a novel multi-agent planning framework that integrates a pool of heterogeneous language model-based agents. By leveraging diverse reasoning patterns across agents, SYMPHONY enhances rollout diversity and facilitates more effective exploration. Empirical results across multiple benchmark tasks show that SYMPHONY achieves strong performance even when instantiated with open-source LLMs deployable on consumer-grade hardware. When enhanced with cloud-based LLMs accessible via API, SYMPHONY demonstrates further improvements, outperforming existing state-of-the-art baselines and underscoring the effectiveness of heterogeneous multi-agent coordination in planning tasks.

The Little-Parks (LP) effect is a quantum phenomenon in which the superconducting transition temperature of a superconducting cylinder (or ring) oscillates periodically as a function of the magnetic flux threading the loop. Recently, multiple experiments have observed half-quantum flux shifts in measurements of LP oscillations, where the oscillations are globally shifted by half a flux quantum compared to conventional cases, a behavior referred to as a $π$-ring. Such observations are commonly linked to unconventional pairing symmetries. In this work, we demonstrate that half-quantum flux shifts can arise in two-dimensional (2D) superconducting rings without invoking unconventional pairing symmetry, provided that vortices near the Berezinskii-Kosterlitz-Thouless (BKT) transition are taken into account. Specifically, based on the vortex-charge duality theory near the BKT transition, we map the problem onto a Coulomb gas model, in which the magnetic flux is represented as a pair of opposite boundary charges (or vortices) at the two edges. The screening of these boundary charges by thermally excited vortex-antivortex pairs is investigated through explicit Monte Carlo simulations. Importantly, we demonstrate that the oscillation of the free-vortex density as a function of magnetic flux can exhibit an anomalous half-quantum flux shift, depending on the geometry of the sample. Our work thus predicts the LP oscillations induced by vortices in 2D superconducting rings near the BKT transition, which provides a new mechanism for generating $π$-rings.

We investigate the influence of capacitive coupling on the detection of anyon braiding in a single-edge interferometer realized in the fractional quantum Hall regime. In this setup, a quantum point contact bends a single edge into a loop, where tunneling occurs at the open end and is controlled by the QPC voltage. In contrast with previously studied two-edge geometries, the weak backscattering regime is dominated by the first-order perturbative term, allowing quantum transport quantities to factorize into a non-universal prefactor and a braiding-induced contribution that provides direct access to the universal statistical angle $πλ$. While previous analyses neglected edge-to-edge capacitance, we show that capacitive effects, which are known to play a crucial role in mesoscopic capacitors, modify both the current and the current cross-correlations. Using a two-point Green's function formalism augmented by Dyson's equation to include the charging energy, we quantify how the fluctuations of the cross-correlations depend simultaneously on $λ$ and on the capacitance of the loop. Our results indicate that a reliable extraction of the statistical angle requires a parallel measurement of the loop capacitance, which can be implemented via a charged gate coupled to the junction.

Critical states in quasiperiodic systems defy the conventional dichotomy between extended and localized states. In this work, we demonstrate that non-Hermiticity fundamentally reshapes this paradigm by giving rise to an exactly solvable quasiperiodic critical phase with no energy selectivity. We introduce a non-Hermitian quasiperiodic lattice based on a modulated Hatano-Nelson model and uncover a new universality class of quasiperiodic skin criticality, in which all eigenstates share an identical multifractal spatial structure. Through a nonunitary gauge transformation, the system is mapped onto a disorder-free lattice, enabling exact analytical solutions for the full spectrum and eigenstates. As a consequence, the inverse participation ratio is strictly energy-independent and controlled solely by a global phase. We further show that this criticality persists in multiband lattices, establishing a general and analytically controlled framework for non-Hermitian quasiperiodic critical phenomena.

Spin transfers in magnetic multilayers offers a promising pathway toward ultrafast, energy-efficient spintronic devices. In this study, we investigate the interfacial spin pumping and temperature-dependent spin current exchange in a Cr1.12Te2/Cu/Ni80Fe20 (Py)(FM1/NM/FM2) trilayer structure. Using broadband and cryogenic ferromagnetic resonance (FMR) measurements, we investigate key magnetization dynamical parameters, including the effective Gilbert damping factor, effective magnetic fields, interfacial spin mixing conductance, and spin current density. Efficient spin angular momentum transfers from Py to Cr1.12Te2 are observed at room temperature. At lower temperatures, the enhanced linewidth reflects temperature dependent spin pumping effects occurring at distinct precession frequencies of the ferromagnetic layers. Notably, the absence of interfacial Damping indicates that spin pumping can be modulated by controlling the net spin current flow. These findings offer critical insight into temperature-dependent tunable spin transport mechanisms in magnetic multilayers, highlighting their potential for next-generation spintronic applications.

Microbes constantly interact with their environment by depleting and transforming food sources. Theoretical studies have mainly focused on Lotka-Volterra models, which do not account for food source dynamics. In contrast, consumer-resource models, which consider food source dynamics, are less explored. In particular, it is still unclear what physical mechanisms control oscillatory dynamics at a single population level, a phenomenon which can only be captured by a consumer-resource model. Here, we present a minimalistic consumer-resource model of a single microbial population with growth and death dynamics, consuming a continuously replenishing substrate. Our model reveals that decaying oscillations can occur around steady state if and only if the timescale of microbial adaptation to food supply changes exceeds the death timescale. This interplay of timescales allows us to rationalize the emergence of oscillatory dynamics when adding various biophysical ingredients to the model. We find that microbial necromass recycling or complementary use of multiple food sources reduces the parameter range for oscillations and increases the decay rate of oscillations. Requiring multiple simultaneous food sources has the opposite effect. Essentially, facilitating growth reduces the likelihood of oscillations around a fixed point. We further demonstrate that such damped oscillatory behavior is correlated with persistent oscillatory behavior in a noisy environment. We hope our work will motivate further investigations of consumer-resource models to improve descriptions of environments where food source distributions vary in space and time.

We present a thermodynamically consistent model for chemically driven running drops on a solid substrate with reversible substrate adsorption of a wettability-changing chemical species. We consider drops confined to a vertical gap, thereby allowing us to first obtain a gradient dynamics description of the closed system, corresponding to a set of coupled dynamical equations for the drop profile and the chemical concentration profiles of species on the substrate and in both fluids (drop, ambient medium). Chemostatting the species in the drop and the ambient medium, we then derive a reduced model for the dynamics of the drop and the adsorbate on the substrate. When the externally imposed chemical potentials are distinct, the system is driven away from thermodynamic equilibrium, allowing for sustained drop self-propulsion across the substrate due to a wettability contrast maintained by chemical reactions. We numerically study the resulting running drops and show how they emerge from drift-pitchfork bifurcations.

Antiskyrmions are unstable in ferromagnetic systems with isotropic bulk or interfacial Dzyaloshinskii-Moriya interaction (DMI). We develop a continuum model for the shrinking dynamics of antiskyrmions in bulk DMI systems, using the Landau-Lifshitz-Gilbert equation for the time derivative of the magnetization field. Owing to the structure of their azimuthal angle, or helicity, elliptic antiskyrmions are energetically favored over circular ones. To capture this feature, we parametrize the magnetization field with a triangular radial profile and an elliptic in-plane shape. This ansatz yields four coupled dynamical equations governing time evolution of the semi-axes, helicities, and rotation angles. In the absence of the DMI, circular antiskyrmions shrink isotropically, exhibiting a crossover from exponential decay to square-root collapse. Initially elliptic antiskyrmions are driven towards circularity. For finite DMI, the semi-axes dynamics couples to the helicity and rotation, where the theory predicts a rotation angle following by half of the slope of the helicity evolution which is linear in time. Only at small semi-axes a cross-over to a logarithmic divergence occurs. The shrinking dynamics of the antiskyrmion size is found to be accompanied by quadrupole-like oscillations. Numerical simulations on the lattice support the predictions from the continuum model.

Classical and quantum spin ice models are amongst the most popular settings for the study of spin liquid physics. $N-$state Potts ice models have been constructed that generalize spin ice, hosting multiple emergent $\text{U}(1)$ gauge fields and excitations charged under non-trivial combinations of these fields. We present a general treatment of classical $N-$state Potts ices relating their properties to the $\mathfrak{su}(N)$ Lie algebras, and demonstrate how the properties of charged excitations in the classical model can be related to this symmetry group. We also introduce quantum generalizations of the Potts Ice models, and demonstrate how charge flavor changing interactions unique to $N>2$ models dominate their low energy physics. We further show how symmetries inherited from the $\mathfrak{su}(N)$ can lead to flux vacuum frustration, greatly modifying the dynamical properties of charged excitations.

Drawing inspiration from swarm intelligence, we show that short-range attractive interactions between thermally driven Brownian quasiparticles enable energy-efficient optimization. As quasiparticles can be generated directly within a material, the swarm size can be adjusted with minimal energy overhead. Using an optimization task defined by a spatially varying temperature landscape, we quantitatively show that interacting swarms reliably identify global optima and significantly outperform non-interacting searchers within a well-defined regime of interaction strength and swarm size. This improvement arises from emergent cooperative behavior, where local interactions guide the swarm toward high-quality solutions without central coordination. To link our physical model to experimental realizations, we coarse-grain the quasiparticle dynamics onto a sensor lattice and generate trajectories emulating particle-tracking measurements. We further show that the interacting swarm adapts robustly to landscapes that evolve over time. These findings establish interacting Brownian quasiparticles as a physical platform for scalable and energy-efficient unconventional computing.

Single-photon emitters (SPEs) hosted by two-dimensional (2D) semiconducting materials are envisioned for next-generation quantum applications. However, SPE creation in 2D semiconductors on rigid substrates like SiO2/Si via nanoindentation is a technological gap, critical for interfacing SPEs with photonic circuits and cavities. Here, we report a protocol for deterministically creating SPEs in monolayer WSe2 on SiO2/Si substrates using a sharp diamond AFM (atomic force microscope) tip. A displacement-controlled indentation process is developed, allowing indent depths > 150 nm necessary for creating SPEs. Sharp defect peaks (~200 μeV) are observed in cryogenic (4K) photoluminescence (PL) spectrum at nanoindented sites and are stable upto ~ 120K. 76% of sites exhibit sharp defect-bound peaks confirmed by power-dependent, temperature-dependent, and time-resolved PL (TRPL). AFM and PL mapping link these peaks to indent periphery. The peaks show sub-linewidth spectral jitter, no blinking, and single-photon nature in second-order autocorrelation measurements. SPEs can be switched on/off, and background emissions suppressed using electrical gating. Gate-voltage dependent TRPL indicate that SPE dynamics can be tuned, depending on nature of SPE, pointing the way to higher-purity SPEs. Our work is directly applicable to other 2D materials and photonic circuit/cavity compatible rigid substrates and is a significant step for scalable SPE technologies.

We explore possible synchronization in two-dimensional (2D) locally coupled discrete-state oscillators under thermal fluctuations, using the self-rotating $q$-state clock model as a prototype. Large-scale Monte Carlo simulations reveal that for $q \ge q_c$ (with $q_c = 5$), the system undergoes two-step Berezinskii-Kosterlitz-Thouless (BKT) transitions: first from a disordered phase to a critical synchronized phase, and then to a spatiotemporal pattern phase. The latter includes oscillatory droplet states that survive in finite systems and a thermodynamically stable spiral wave state. Notably, the synchronized phase features algebraically decaying spatial correlations, alongside divergent coherence time, thus realizing a continuous time crystal; while it vanishes when $q < q_c$. Mean-field theory supports the existence of the synchronized phase, but predicts a lower critical value $q_c^{MF} = 4$.

It is shown that the widely used two-dimensional model of random networks of metallic nanowires or carbon nanotubes significantly overestimates the number of contacts between elements compared to real systems, which, within the mean-field approach, leads to overestimated estimates of electrical conductivity, especially when the contact resistances between conductors make the main contribution to the electrical conductivity of the system. In the case of a two-dimensional model, the electrical conductivity of the system depends quadratically on the number density of conductors, whereas in the case of a three-dimensional model this dependence is linear.

We show that recent numerical findings of universal scaling relations in systems of noisy, aligning self-propelled particles by Kürsten [Kürsten, arXiv:2402.18711v2 [cond-mat.soft] (2025)] can robustly be explained by perturbation theory and known results for the Mathieu equation with purely imaginary parameter. In particular, we highlight the significance of a cascade of exceptional points that leads to non-trivial fractional scaling exponents in the singular-perturbation limit of high activity. Crucially, these features are rooted in the Fokker-Planck operator corresponding to free self-propulsion. This can be viewed as a dynamical phase transition in the dynamics of noisy active matter.

Quantized vortices are fundamental topological excitations of quantum fluids. We report single-shot interferometric measurements of spontaneous vortex nucleation in a room-temperature organic exciton-polariton condensate. From hundreds of independent realizations we find random vortex-core positions and unbiased circulation, consistent with intrinsically stochastic, unpinned defect formation. The mean vortex number scales with pump power above threshold with an exponent consistent with Kibble-Zurek freeze-out in a driven-dissipative condensate. Using reconstructed phase maps we obtain single-shot flow fields, compute the incompressible component, and extract kinetic-energy spectra. Vortex-containing realizations develop a robust Kolmogorov-like segment with Einc(k) proportional to k^(-5/3) over a finite k range, indicating the onset of turbulent spectral scaling in a quantum fluid of light. These results establish single-shot access to phase and flow as a direct route to quantifying stochastic defect formation and emerging turbulence in polariton condensates.

The relationship between the topological architecture of polymer networks and their macroscopic rupture remains a fundamental challenge in polymer physics. Recent coarse-grained simulations have revealed that the dependence of stretch at break (λ_b) on node functionality and reaction conversion can be unified into a universal master curve when plotted against the cycle rank density (ξ). However, a theoretical derivation explaining this universality has been lacking. This study proposes a simple mechanical model to describe the ξ-dependence of fracture strain. The polymer network is modeled as a mechanical system consisting of a sequence of springs representing localized highly stretched strands and the surrounding unstretched network. By relating the stiffness contrast between these regions to the network connectivity defined by ξ, an analytical expression for the stretch at break is derived: λ_b\propto\sfrac{\left(2ξ+5\right)}{\left(2ξ+1\right)}. The proposed model is validated against phantom chain simulations using both Gaussian and finite extensibility (FENE) springs. The theoretical prediction shows excellent agreement with simulation data, providing a physical basis for the phenomenological universality observed in polymer network rupture.

We solve the problem of the spin quantum Hall transition on random networks using a mapping to classical percolation that focuses on the boundary of percolating clusters. Using tools of two-dimensional quantum gravity, we compute critical exponents that characterize this transition and confirm that these are related to the exponents for the regular (square) network through the KPZ relation. Our results demonstrate the relevance of the geometric randomness of the networks and support conclusions of numerical simulations of random networks for the integer quantum Hall transition.

Many-body and complex systems, both classical and quantum, often exhibit slow, nonlinear relaxation toward stationary states due to the presence of metastable configurations and environmental fluctuations. Nonlinear relaxation in a wide variety of natural systems proceeds through metastable states, which arise in condensed-matter physics as well as in fields ranging from cosmology and biology to high-energy physics. Moreover, noise-induced phenomena play a central role in shaping the dynamics of such systems far from equilibrium. This review develops a unifying perspective centered on noise-assisted stabilization and the statistical properties of metastable dynamics. We first discuss escape processes driven by Lévy flights in smooth metastable potentials, emphasizing the emergence of nonmonotonic residence-time behavior. We then connect these concepts to stochastic resistive switching in memristive devices, where noise-induced effects can enhance stability and reproducibility. We further examine driven dissipative quantum bistability, showing how the interplay between external driving and system-environment coupling reshapes escape pathways and lifetimes. Finally, we outline how switching-time statistics in current-biased Josephson junctions can provide an experimentally accessible strategy for axion detection, based on an axion-induced resonant-activation signature.

We examine the reaction of a homogeneous spherical fluid vesicle to the force exerted by a rigid circular ring located at its equator in the linear regime. We solve analytically the linearized first integral of the Euler-Lagrange equation subject to the global constraints of fixed area and volume, as well as to the local constraint imposed by the ring. We determine the first-order perturbations to the generating curve of the spherical membrane, which are characterized by the difference of the radii of the membrane and the ring, and by a parameter depending on the physical quantities of the membrane. We determine the total force that is required to begin the deformation of the membrane, which gives rise to a discontinuity in the curvature of the membrane across the ring.

We theoretically study the emergence of topological magnons in electron-magnon coupled systems. The magnon dispersion in a ferromagnet usually possesses an effective time reversal symmetry in the absence of Dzyaloshinskii-Moriya (DM) interaction, preventing the appearance of topological magnons. When a spin system is coupled to itinerant electrons, we find that the magnon band structure of the spin system experiences time-reversal symmetry breaking with the electron-magnon interaction via the exchange coupling, where topological magnons arise without requiring strong DM. Specifically, we consider a heterostructure consisting of a ferromagnetic insulator and a transition metal dichalcogenide (TMD) monolayer and investigate topological gap opening in magnon bands. Our findings reveal that even trivial ferromagnets can host topological magnons via coupling to itinerant electronic systems.

Techniques for detecting critical phenomena -- phase transitions where correlation length diverges and small perturbations have large effects -- have been developed across at least eight fields of application over nine decades. We document this convergence pattern. The physicist's correlation length $ξ$, the cardiologist's DFA scaling exponent $α$, the financial analyst's Hurst exponent $H$, and the machine learning engineer's spectral radius $χ$ all measure correlation decay rate, detecting the same critical signatures under different notation. Citation analysis reveals minimal cross-domain awareness during the formative period (1987--2010): researchers in biomedicine, finance, machine learning, power systems, and traffic flow developed equivalent techniques independently, each with distinct notation and terminology. We present Metatron Dynamics, a framework derived from distributed systems engineering, as a candidate ninth independent discovery -- strengthening the convergence pattern while acknowledging that as authors of both the framework and this analysis, external validation would strengthen this claim. Correspondence testing on the 2D Ising model confirms that measures from multiple frameworks correctly identify the critical regime at $T_c = 2.269$. We argue that repeated independent discovery establishes criticality mathematics as fundamental public knowledge, with implications for cross-disciplinary education and research accessibility. Because these findings affect fields beyond mathematics and physics, we include a plain-language summary in Appendix B for non-specialist readers.

These are the notes for the 4.5-hour course with the same title that I delivered in August 2025 at the Les Houches summer school ``Exact Solvability and Quantum Information''. In these notes I pedagogically introduce the setting of brickwork quantum circuits and show that it provides a useful framework to study non-equilibrium quantum many-body dynamics in the presence of local interactions. I first show that brickwork quantum circuits evolve quantum correlations in a way that is fundamentally similar to local Hamiltonians, and then present examples of brickwork quantum circuits where, surprisingly, one can compute exactly several relevant dynamical and spectral properties in the presence of non-trivial interactions.

A minimal model of a graphene topological insulator heterostructure is considered, where a moire superlattice modulates the Rashba spin orbit interaction. In the spin degenerate, spin orbit free limit, the reduced Brillouin zone contains flat, spin degenerate moire minibands, with periodicity determined by superlattice folding. The inclusion of spin orbit interaction lifts the spin degeneracy and reduces the effective spectral periodicity by a factor of two. Through spin orbit interaction, the moire potential entangles spin, sublattice, and leg degrees of freedom, reshaping the miniband structure in momentum space and generating emergent helicity spectral functions. As the Rashba coupling is renormalized by the moire pattern, it induces helicity fragmentation, in which the helicity weight is distributed across a dense manifold of moire minibands, forming an extended network of helicity carrying states and significantly enhancing helicity fluctuations at the bare response level. The emergence of Dirac like miniband crossings at finite spin orbit interaction demonstrates that moire heterostructures can support relativistic quasiparticles through band reconstruction. This model provides a microscopic mechanism by which proximity induced spin orbit coupling can be amplified via moire engineering.

While software implementations of neural networks have driven significant advances in computation, the von Neumann architecture imposes fundamental limitations on speed and energy efficiency. Neuromorphic networks, with structures inspired by the brain's architecture, offer a compelling solution with the potential to approach the extreme energy efficiency of neurobiological systems. Photonic neuromorphic networks (PNNs) are particularly attractive because they leverage the inherent advantages of light, namely high parallelism, low latency, and exceptional energy efficiency. Previous PNN demonstrations have largely focused on device-level functionalities or system-level implementations reliant on supervised learning and inefficient optical-electrical-optical (OEO) conversions. Here, we introduce a purely photonic deep PNN architecture that enables online, unsupervised learning. We propose a local feedback mechanism operating entirely in the optical domain that implements a Hebbian learning rule using non-volatile phase-change material synapses. We experimentally demonstrate this approach on a non-trivial letter recognition task using a commercially available fiber-optic platform and achieve a 100 percent recognition rate, showcasing an all-optical solution for efficient, real-time information processing. This work unlocks the potential of photonic computing for complex artificial intelligence applications by enabling direct, high-throughput processing of optical information without intermediate OEO signal conversions.

In this study, a new alternative model algorithm has been proposed for assembling amorphous structures, unifying the bosonic paradigm applicable at low temperatures with crystalline models relevant at room and higher temperatures. Physical meaning of main model parameters is determined together with an explanation for the appearing bosonic peak using the random graph theory. Numerically, statistical atomic distribution in a multiphase amorphous system is provided without the melting simulation of base crystals, and the mean energy function has been determined analytically. The calculated table data are in good agreement with neutronography measurements of the actual amorphous alloy in its solid state. Programme optimisations were also implemented, and we outlined several effective steps to achieve the higher processing speed. The proposed programme code can be used for potential test assembling and simulations of amorphous systems with sorting by the optimal atomic content or proportion (i.e. glass forming ability).

Cavity magnonics investigates hybrid systems where magnons interact coherently with photons, providing a platform to harness light-matter interaction in magnetic materials. Progress in this field hinges on achieving stronger and tunable nonlinear effects, which are essential for controlling magnon dynamics and frequency conversion. Here, we demonstrate the magnon Kerr effect in an anisotropic magnonic system comprising a 200~nm-thick yttrium iron garnet film strongly coupled to a three-dimensional microwave resonator. The strong shape anisotropy significantly enhances the magnon Kerr effect compared to a sphere of equivalent volume, while the cavity enables sensitive probing of magnetization dynamics. We demonstrate continuous tunability of the magnitude and sign of the Kerr shift by controlling the static orientation of the magnetization. Input-output modeling of the magnon-photon interaction provides a consistent description of our system and Kerr coefficients matching the experimental results. Our findings demonstrate a scalable approach to enhancing Kerr anharmonicity in hybrid magnon-photon systems while preserving strong coupling.

I examine the physical foundations of temperature and thermal equilibrium by identifying photons as the fundamental agents that establish and maintain the characteristic energy scale $E_c = k_B T$ in ordinary matter. While classical thermodynamics successfully describes equilibrium phenomenologically, the realization of thermal distributions requires concrete microscopic mechanisms provided by quantum electrodynamics. We derive the Boltzmann distribution from a minimal differential scaling postulate and show that sustaining thermal equilibrium demands continuous photon exchange with average energy $\langle hν\rangle = 2.701\,E_c$, quantifying the energetic throughput necessary to counter radiative losses. Entropy production is shown to arise naturally from inelastic photon scattering that converts high-energy photons into many lower-energy quanta, thereby increasing accessible microstates and driving irreversible evolution toward equilibrium. We establish physical criteria distinguishing genuine thermal equilibrium from purely formal temperature assignments and demonstrate that the classical notion of an infinite thermal reservoir emerges as an effective idealization within a hierarchy of dynamically maintained photon baths. This photonic framework complements phenomenological thermodynamics by providing its microscopic foundation and clarifies the physical meaning of temperature as an emergent collective property of photon-mediated energy exchange.

We study the statistical properties of trainable agents moving in discrete space. After introducing the mathematical framework, we first analyze the dynamics of two completely random walkers, mutually competing in a chaser-target interaction scheme. The statistics of the encounters is analytically obtained and the predictions tested versus numerical simulations. We then move forward to extend the baseline case to agents capable of learning and adapting to an external reward signal, using reinforcement learning. Smart walkers morph the statistics of the encounter, to maximize their cumulated reward, as confirmed by combined numerical and analytical insights. More interestingly, configuration entropy proves a reliable proxy to gauge the acquired ability of the agents to cope with the assigned task when no other information about them (i.e. reward signal, policy, etc) is present. We further test the proposed measure of learned skills by operating the Stockfish chess engine against a quasi-random untrained opponent. The obtained conclusions corroborate our claim.

We show that Andreev spin qubits can be realized in a Josephson junction based on the helical edge states of a two-dimensional topological insulator (quantum spin Hall system) proximized by superconducting films, in the presence of magnetic doping. We demonstrate that the electrical dipole transitions between the Andreev spin states induced by the magnetic doping can be harnessed to optically manipulate the Andreev spin qubit by microwave radiation pulses. We numerically simulate the realization of NOT and Hadamard quantum logic gates, and discuss implementations in realistic setups.

Quantum many-body systems display an extraordinary degree of complexity, yet many of their features are universal: they depend not on microscopic details, but on a few fundamental physical aspects such as symmetries. A central challenge is to distill these universal characteristics from model-specific ones. Random quantum states sampled from a uniform distribution, the Haar measure, provide a powerful framework for capturing this typicality. Here, we experimentally study the entanglement and symmetries of random many-body quantum states generated by evolving simple product states under ergodic Floquet models. We find excellent agreement with the predictions from the Haar-random state ensemble. First, we measure the Rényi-2 entanglement entropy as a function of the subsystem size, observing the Page curve. Second, we probe the subsystem symmetries using entanglement asymmetry. Finally, we measure the moments of partially transposed reduced density matrices obtained by tracing out part of the system in the generated ensembles, thereby revealing distinct entanglement phases. Our results offer an experimental perspective on the typical entanglement and symmetries of many-body quantum systems.

We investigate the effect of translational and rotational inertia on motility-induced phase separation in underdamped active dumbbells and identify the emergence of four distinct kinetic temperatures across the coexisting phases-unlike in overdamped systems. We find that the dilute, gas-like phase consistently exhibits a higher translational kinetic temperature than the dense, liquid-like phase, with this difference amplified by increasing the rotational inertia. Rotational kinetic temperatures display a similar trend, with the dense phase remaining colder than the dilute phase; however, in this case the temperature difference grows with translational inertia and activity, while becoming practically independent of rotational inertia. This counterintuitive behavior arises from the interplay of activity-driven collisions with both translational and rotational inertia in the coexisting phases. Our results highlight the critical role of translational and rotational inertia in shaping the kinetic temperature landscape of motility-induced phase separation and offer new insights into the nonequilibrium thermodynamics of active matter.

In a one-dimensional ferromagnet-superconductor nanowire, magnetism can suppress superconductivity except where the Zeeman field is suppressed, for example domain wall superconductivity (DWS) near magnetic domain walls or multi-domain-averaged superconductivity (MDAS) in multi-domain states where the net magnetization over the coherence length averages to nearly zero. Here we study full-shell InAs/EuS/Al nanowires using scanning SQUID magnetometry and transport, and find superconductivity in the Al shell only when the EuS is in a multi-domain state, consistent with both DWS and MDAS, and absent in the saturated single-domain state. Scanning SQUID magnetometry further shows that the EuS magnetic texture is position dependent and reconfigurable by small changes in external magnetic field, including moving a well-defined domain wall at $\approx$5.5 $μ$m/mT with sub-mT fields, implying that any associated localized superconducting region would likewise be movable. Such magnetic texture controlled superconductivity along a nanowire may be useful for topological qubits, Andreev spin qubits, superconducting logic, and memory devices.

Nonlinear magnetic response driven by time-periodic magnetic fields offers a distinct route to probe spin-resolved quantum geometry beyond conventional electric-field-driven nonlinear effects. While linear magnetic responses depend on the Zeeman quantum geometric tensor, the influence of generalized spin-rotation quantum geometries on nonlinear responses has not been established. Here, we develop a microscopic quantum-kinetic framework to elucidate how the Zeeman and spin-rotation quantum geometric tensors govern nonlinear gyrotropic magnetic transport in two-dimensional systems. We derive second-order gyrotropic magnetic currents and reveal a distinct geometric separation: the off-diagonal sector is controlled by the Zeeman symplectic and metric connections, whereas the diagonal sector is dictated by the spin-rotation quantum metric and Berry curvature. This identifies the spin-rotation quantum geometric tensor as a fundamental geometric quantity unique to the nonlinear regime. Applying our theory to massless Dirac fermions, hexagonally warped topological insulator surface states, tilted massive Dirac fermions, and parity-time symmetric CuMnAs, we demonstrate how specific symmetries selectively activate conduction and displacement channels. Our findings link spin-resolved quantum geometry to nonlinear magnetic transport, offering design principles for engineering tailored nonlinear magnetic responses in optoelectronic and spintronic devices.

Transition metal dichalcogenides (TMDs) hold relevance for spin-triplet superconducting phases due to their inherent Ising spin-orbit coupling, but the majority of studies have so far focused on oversimplified models. In this work, we consider a TMD monolayer using a three-orbital model with anisotropic couplings and investigate the emergent superconducting properties when it is placed in proximity to a conventional spin-singlet $s$-wave superconductor. We find that the multiorbital nature of the TMDs lead to superconducting gaps not only at zero energy, but also at higher energies, so-called hybridization gaps, which exhibit a complex structure due to the anisotropic couplings, challenging their spectral measurement. Moreover, we find that the inherent Ising spin-orbit coupling induces a spin splitting and a spin polarization along the $z$-direction, which correlates with the emergence of mixed spin-triplet superconducting pairs. These spin-triplet pair correlations appear in the monolayer as a proximity-induced effect due to the impact of the Ising spin-orbit field on conventional spin-singlet $s$-wave superconductivity. Taking realistic parameters for a $\text{MoS}_2$ monolayer, we show that the Ising field is strong enough to induce spin-triplet pair correlations of the same magnitude as their spin-singlet counterparts. We also include Rashba spin-orbit coupling, naturally emerging in a heterostructure and find that it induces equal spin-triplet superconducting pairs that compete with the mixed spin-triplet pairs induced by the Ising spin-orbit coupling. Our findings help understand the superconducting properties of TMDs in proximity to conventional superconductors.

The rapid rise of artificial intelligence has led to an unsustainable growth in energy consumption. This has motivated progress in neuromorphic computing and physics-based training of learning machines as alternatives to digital neural networks. Many theoretical studies focus on simple architectures like all-to-all or densely connected layered networks. However, these may be challenging to realize experimentally, e.g. due to connectivity constraints. In this work, we investigate the performance of the widespread physics-based training method of equilibrium propagation for more realistic architectural choices, specifically, locally connected lattices. We train an XY model and explore the influence of architecture on various benchmark tasks, tracking the evolution of spatially distributed responses and couplings during training. Our results show that sparse networks with only local connections can achieve performance comparable to dense networks. Our findings provide guidelines for further scaling up architectures based on equilibrium propagation in realistic settings.

Each cycle of a quantum computation requires a quantum state initialisation. For semiconductor-based quantum platforms, initialisation is typically performed via slow microwave processes and usually requires cooling to temperatures where only the lowest quantum level is occupied. In silicon, boron atoms are the most common impurities. They bind holes in orbitals including an effective spin-3/2 ground state as well as excited states analogous to the Rydberg series for hydrogen. Here we show that initialisation temperature demands may be relaxed and speeds increased over a thousand-fold by importing, from atomic physics, the procedure of optical pumping via excited orbital states to preferentially occupy a target ground state spin. Spin relaxation within the orbital ground state of unstrained silicon is too fast to measure for conventional pulsed microwave technology, except at temperatures below 2 K, implying a need not only for fast state preparation but also fast state readout. Circularly polarised ~10 THz photon pulses from a free electron laser meet both needs at temperatures above 3 K: a 9 ps pulse enhances the population of one spin eigenstate for the "1s"-like ground state orbital, and the second interrogates this imbalance in spin population. Using parameters given by our data, we calculate that it should be possible to initialise 99% of spins for boron in strained silicon within 250 ps at 3 K. The speedup of both state preparation and measurement gained for THz rather than microwave photons should be explored for the many other solid state quantum systems hosting THz excitations potentially useful as intermediate states.

We introduce the concept of \textit{defect relative entropy} as a measure of distinguishability within the space of defects. We compute the defect relative entropy for conformal/topological defects, deriving a universal formula in conformal field theories (CFTs) on a circle. This formula reduces to the Kullback-Leibler divergence. Furthermore, we provide a detailed expression of the defect relative entropy for diagonal CFTs with specific topological defect choices, utilizing the theory's modular $\mathcal{S}$ matrix. We also present a general formula for the \textit{ defect sandwiched Rényi relative entropy} and the \textit{defect fidelity}. Through explicit calculations in specific models, including the Ising model, the tricritical Ising model, and the $\widehat{su}(2)_{k}$ WZW model, we have made an intriguing finding: zero defect relative entropy between reduced density matrices associated with certain topological defect. Notably, we introduce the concept of the \textit{defect relative sector}, representing the set of topological defects with zero defect relative entropy.

The evaluation of the electrostatic potential is fundamental to the study of condensed phase systems. We discuss the calculation of the relevant lattice summations by Ewald-type techniques. A model charge density is introduced, that cancels multipole moments of the crystalline charge distribution up to a desired order, for accelerating convergence of the Ewald sums. The method is applicable to calculations of bulk systems, employing arbitrary unit cells in a classical or quantum context, and with arbitrary basis functions to represent the charge density. The approach clarifies a decades-old implementation in the CRYSTAL code.

We study the behaviour of a thin fluid filament (a rivulet) flowing in an air-filled Hele-Shaw cell. Transverse and longitudinal deformations can propagate on this rivulet, although both are linearly attenuated in the parameter range we use. On this seemingly simple system, we impose an external acoustic forcing, homogeneous in space and harmonic in time. When the forcing amplitude exceeds a given threshold, the rivulet responds nonlinearly, adopting a peculiar pattern. We investigate the dance of the rivulet both experimentally using spatiotemporal measurements, and theoretically using a model based on depth-averaged Navier-Stokes equations. The instability is due to a three-wave resonant interaction between waves along the rivulet, the resonance condition fixing the pattern wavelength. Although the forcing is additive, the amplification of transverse and longitudinal waves is effectively parametric, being mediated by the linear response of the system to the homogeneous forcing. Our model successfully explains the mode selection and phase-locking between the waves, it notably allows us to predict the frequency dependence of the instability threshold. The dominant spatiotemporal features of the generated pattern are understood through a multiple-scale analysis.

Plastic deformation in amorphous solids is carried by localized shear transformations that self-organize into avalanches. In amorphous carbon modeled with a machine-learned interatomic potential, we find that the energetics and organization of these avalanches can be resolved by systematically following the underlying energy landscape. With a pseudo-arclength numerical continuation framework, we decompose avalanches into constituent shear transformations and determine their strain-dependent energetics. Our analysis shows that, prior to onset, avalanches have a latent structure that consists of well-separated local minima. We further demonstrate that arclength continuation yields an event driven framework for following avalanche dynamics, eliminating time-step effects on statistical avalanche properties such as distributions of stress drops.

We study the response of Chua's circuit driven by a chaotic signal of variable time-scale. We observe that when the frequency of the drive is significantly lower than that of the response and the driving strength is above a threshold, the Chua's circuit exhibits multiple stable attractors. The features of the attractors change as the driving strength ε increases, for instance the attractors are double-scroll at low ε and are single-scroll when ε is high. We also investigate generalized synchronization(GS) between the drive and the response systems by employing the auxiliary system approach. When the drive is much slower than the response, we observe different scenarios of remote synchronization(RS) between response and auxiliary units. In addition to complete synchrony between response and auxiliary systems indicating GS between drive and response, we notice that the response and auxiliary units can be lag synchronized and can also have correlated trajectories indicating novel forms of RS. The slow drive can induce multistability between these RS states which disappears as the frequency of drive increases and become equivalent to the response Chua's ciruit.

Exact unstable solutions of the Navier-Stokes equations are thought to underpin the dynamics of turbulence, but are usually computed in minimal computational domains. Here, we extend this dynamical systems approach to spatially extended turbulent flows featuring multiple interacting 'substructures', and show how new simple invariant solutions can be constructed by spatial tiling of exact solutions from small-box calculations. Candidate solutions are found via gradient-based optimization of a scalar loss function which targets autorecurrence in spatially-masked regions of the flow. We apply these ideas to a vertically-extended Kolmogorov flow, where we first identify large numbers of relative periodic orbits (RPOs) which are combinations of high-dissipation, small-box solutions with laminar patches. We then show that vertically-stacked combinations of pairs of distinct small-box RPOs can form robust guesses for dynamically-relevant two-tori in the larger domain. Finally, we show how our optimization procedure can identify 'turbulent' trajectories which locally shadow a small-box RPO for multiple periods in a subdomain. These small-box combinations are possible as the flow spends prolonged periods in a regime where it can be effectively considered as a pair of weakly-coupled small-box systems, due to shielding effects associated with higher-dissipation flow structures.

Traditional compartmental models, including SIR, SEIR, and SIRS frameworks, remain the analytical standard for epidemic forecasting. However, real-world data validation consistently reveals significant predictive failures, such as peak underestimations of up to 50%. This research identifies a persistent fundamental methodological error: the calibration of prevalence-based (stock) models using raw daily incidence (flow) data without proper transformation. We propose an integrated protocol utilizing an exponentially weighted convolution to reconstruct active cases from reported incidence: $I(t) \approx \frac{1}{p} \int_{0}^{t} NDC(τ) e^{-γ(t-τ)} dτ$. This transformation accounts for the recovery rate $γ$ and the ascertainment rate $p$. We demonstrate that increasing structural complexity, such as adding latency (SEIR) or waning immunity (SIRS), fails to resolve the incidence-prevalence gap. Simulation results show that without the proposed universal pre-processor, these advanced models inherit the systematic biases of misaligned data types, leading to significant errors in estimating latent periods and the "heavy tail" of endemicity. The proposed convolution transformation must serve as a universal prerequisite for any compartmental framework, bridging the gap between clinical reporting and mechanistic modeling.

An active research line within the broader field of network science is the one concerning link prediction. Close in scope to network reconstruction, link prediction targets specific connections with the aim of uncovering the missing ones, as well as predicting those most likely to emerge in the future, from the available information. In this paper, we consider two families of methods, i.e. those rooted in statistical physics and those based upon machine learning: the members of the first family identify missing links as the most probable non-observed ones, the probability coefficients being determined by solving maximum-entropy benchmarks over the accessible network structure; the members of the second family, instead, associate the presence of single edges to explanatory node-specific variables. Running likelihood-based models such as the Configuration Model, or one of its many fitness-based variants, in parallel with the Gradient Boosting Decision Tree algorithm reveals that the former's accuracy is comparable to (and sometimes slightly higher than) the latter's. Such a result confirms that white-box algorithms are viable competitors to the currently available black-box ones, being computationally faster and more interpretable than the latter.

Mechanical models of brain tissue are a beneficial tool to simulate neurosurgical interventions, disease progression, or brain development. However, the accuracy and predictive capacity of such a model relies on a precise experimental characterization of the tissue's mechanical behavior. Such a characterization is yet limited by inconsistent or contradictory experimental responses reported in the literature, particularly when measurements are performed in different time or length scales. Although brain tissue has been extensively investigated in previous studies, the combination of experimental findings from different scales has received limited attention. In this study, we combine ex vivo mechanical responses of porcine brain tissue obtained at different time scales in a mechanical model. We investigated the mechanical behavior of three different brain regions in the quasi-static domain with multi-modal large strain rheometer measurements and at high frequencies with magnetic resonance elastography (MRE). A comparative analysis of the mechanical parameters obtained from both experimental techniques demonstrated consistent regional variations in the viscoelastic behavior across the two domains. However, the mechanical behavior changes from a higher elasticity in the quasi-static and low frequency domain to a dominating viscosity at high frequencies. Based on the quasi-static and the high frequency behavior, we calibrated a fractional Kelvin-Voigt model and consequently unified the two responses in a single mechanical model to obtain a comprehensive characterization of the tissue's mechanical behavior.

The Martian brain terrain (MBT), characterized by its unique brain-like morphology, is a potential geological archive for finding hints of paleoclimatic conditions during its formation period. The morphological similarity of MBT to self-organized patterned ground on Earth suggests a shared formation mechanism. However, the lack of quantitative descriptions and robust physical modeling of self-organized stone transport jointly limits the study of the thermal and aqueous conditions governing MBT's formation. Here we established a specialized quantitative system for extracting the morphological features of MBT, taking a typical region located in the northern Arabia Terra as an example, and then employed a numerical model to investigate its formation mechanisms. Our simulation results accurately replicate the observed morphology of MBT, matching its key geometric metrics with deviations $<10\%$. Crucially, however, we find that the self-organized transport can solely produce relief $<0.5$ m, insufficient to explain the formation of MBT with average relief of $3.29 \pm 0.65$ m. We attribute this discrepancy to sculpting driven by late-stage sublimation, constraining cumulative subsurface ice loss in this region to $\sim 3$ meters over the past $\sim 3$ Ma. These findings demonstrate that MBT's formation is a multi-stage process: initial patterning driven by freeze-thaw cycles (implying liquid water) followed by vertical sculpting via sublimation (requiring a dry environment). This evolution provides physical evidence for the transition of the ancient Martian climate from a wetter period to a colder hyper-arid state.

Species interactions through cross-feeding via leakage and uptake of chemicals are important in microbial communities, and play an essential role in the coexistence of diverse species. Here, we study a simple dynamical model of a microbial community in which species interact by competing for the uptake of common metabolites that are leaked by other species. The model includes coupled dynamics of species populations and chemical concentrations in the medium, allowing for a variety of uptake and leakage networks among species. Depending on the structure of these networks, the system exhibits different attractors, including fixed points, limit cycles, low-dimensional chaos, and high-dimensional chaos. In the fixed-point and limit-cycle cases, the number of coexisting species is bounded by the number of exchangeable chemicals, consistent with the well-known competitive exclusion principle. In contrast, in the low-dimensional chaotic regime, the number of coexisting species exhibits noticeable but limited excess over this limit. Remarkably, in the high-dimensional chaotic regime, a much larger number of species beyond this limit coexist persistently over time. In this case, the rank-abundance distribution is broader than exponential, as often observed in real ecosystems. The population dynamics displays intermittent switching among quasi-stationary states, while the chemical dynamics explore most of the high dimensions. We find that such high-dimensional chaos is ubiquitous when the number of uptake chemicals is moderately larger than the number of leaked chemicals. Our results identify high-dimensional chaos with intermittent switching as a generic dynamical mechanism that stabilizes coexistence in interacting systems. We discuss its relevance to sustaining diverse microbial communities with leak-uptake cross-feeding.

This work shows that the recently discovered operator contraction identity for solving the discreet Path Integral of the harmonic oscillator can be applied equally to fermions in any dimension. This then yields an exactly solvable model for studying the sign problem where the Path Integral Monte Carlo energy at any time step for any number of fermions is known analytically, or can be computed numerically. It is found that repulsive/attractive pairwise interaction shifts the sign problem to larger/smaller imaginary time, but does not make it more severe than the non-interacting case. More surprisingly, for closed-shell number of fermions, the sign problem goes away at large imaginary time. Fourth-order and newly found variable-bead algorithms are used to compute ground state energies of quantum dots with up to 110 electrons and compared to results obtained by modern neural networks.

Understanding water-metal interactions is central to disciplines spanning catalysis, electrochemistry, and atmospheric science. Monolayer ice phases are well established on hydrophilic surfaces, where strong water-substrate interactions stabilize ordered hydrogen-bond networks. In contrast, their formation on hydrophobic metals has been deemed ther-modynamically unfavourable, with water typically assembling into amorphous films, three-dimensional crystallites, or interlocked bilayer ice. Here, we demonstrate the synthesis of a monolayer ice phase on the hydrophobic Au(111) surface using a low-energy-electron-assisted growth method. Combined experimental characterizations including low-energy electron diffraction, angle-resolved photoemission spectroscopy, and X-ray photoelectron spectroscopy, complemented by first-principles calculations, prove that the monolayer ice phase composes of intact water molecules. This approach provides a generalizable strategy for stabilizing ordered two-dimensional ice on inert substrates and offers new insight into the interplay between water and low-energy electrons at hydrophobic interfaces.

Multiplex contagion dynamics display localization phenomena in which spreading activity concentrates on a subset of layers, as well as delocalized regimes where layers behave collectively. We investigate how these regimes are encoded in temporal correlations of node activity. By deriving a closed-form mean-field expression for node autocorrelations in a contact-based social contagion multiplex model and validating it through simulations, we show that lag-one autocorrelations act as sensitive indicators of both activation and localization transitions. Our results establish temporal correlations as lightweight, structure-agnostic probes of multiplex spreading dynamics, particularly valuable in partially observable systems.

Despite the growing utility of Large Language Models (LLMs) for simulating human behavior, the extent to which these synthetic personas accurately reflect world and moral value systems across different cultural conditionings remains uncertain. This paper investigates the alignment of synthetic, culturally-grounded personas with established frameworks, specifically the World Values Survey (WVS), the Inglehart-Welzel Cultural Map, and Moral Foundations Theory. We conceptualize and produce LLM-generated personas based on a set of interpretable WVS-derived variables, and we examine the generated personas through three complementary lenses: positioning on the Inglehart-Welzel map, which unveils their interpretation reflecting stable differences across cultural conditionings; demographic-level consistency with the World Values Survey, where response distributions broadly track human group patterns; and moral profiles derived from a Moral Foundations questionnaire, which we analyze through a culture-to-morality mapping to characterize how moral responses vary across different cultural configurations. Our approach of culturally-grounded persona generation and analysis enables evaluation of cross-cultural structure and moral variation.

Smartwatches are widely used to estimate caloric expenditure for weight management, clinical decision making, and public health monitoring. These devices combine photoplethysmography, accelerometry, and proprietary algorithms. However, prior studies report substantial error, and the influence of moderators such as skin tone and body fat percentage (BF) remains underexamined. This study tested whether smartwatch brand, BF, and Fitzpatrick skin type (III to V) predict caloric expenditure error relative to indirect calorimetry. Fifty eight Hispanic adults completed a single laboratory visit including a ten minute recumbent cycling protocol with alternating two minute moderate and vigorous intensity intervals, bracketed by rest and recovery. Participants wore four consumer devices: Apple Watch Series 8, Fitbit Sense 2, Samsung Galaxy Watch 5, and Garmin Forerunner 955. Energy expenditure was measured using a COSMED K5 metabolic system. After device specific data quality filtering, valid participant device pairings ranged from 44 to 52 per brand. One sample tests showed significant mean bias for three devices: Apple, Garmin, and Samsung. Fitbit showed no significant overall bias, although this depended on device specific outlier removal. Mean bias varied by brand, with Garmin and Samsung showing the largest overestimations. Mixed effects models revealed significant effects of device and BF, as well as a device by BF interaction, with physical activity energy expenditure error increasing as adiposity increased. Overall, common smartwatches substantially misestimate caloric expenditure compared with indirect calorimetry. Error varies by brand and worsens with higher body fat, highlighting limitations of current consumer wearables and the need for improved accuracy across diverse body types.

Wearable devices are widely used for heart rate (HR) monitoring, yet their accuracy across diverse body compositions and skin tones remains uncertain. This study evaluated four wrist worn devices (Apple, Fitbit, Samsung, Garmin) in 58 Hispanic adults with Fitzpatrick skin types III to V during a cycling protocol alternating moderate (0.64 to 0.76 HRmax) and vigorous (0.77 to 0.95 HRmax) intensities. Criterion HR was obtained using a Polar H10 ECG, and accuracy was assessed using mean absolute error, mean absolute percentage error (MAPE), bias, and intraclass correlation coefficients. All devices showed significant deviation from criterion measures. Apple and Garmin demonstrated the lowest error, whereas Fitbit and Samsung exhibited greater inaccuracies. Higher BMI and darker skin tones were associated with increased MAPE. These biases disproportionately affect higher risk populations, underscoring the need for improved algorithms to ensure equitable health monitoring.

In this paper we have revisited the energy-momentum squared gravity theory, by taking into account the second derivative of the matter Lagrangian with respect to the metric, encapsulating relations originated from thermodynamical grounds. After obtaining the scalar tensor representation of the energy-momentum squared gravity with the new corrections, we have analyzed the physical implications by relying on the linear stability theory. The results show that the current cosmological system is compatible with the expansion of the Universe for some specific matter Lagrangians, explaining the emergence of matter domination era, approaching the late time accelerated expansion era close to the de-Sitter phenomenology.

GPS mobility data is a valuable source of behavioral measurement which is subject to systematic biases including the over- or under-representation of demographic groups, and variations in the quality of location sampling across time. In this paper, we address the challenge of temporal bias in mobility data, which can skew the representation of mobility behaviors due to the event-based nature of location data sampling. We use the American Time Use Survey (ATUS) to assess the accuracy of a place-based measure of economic segregation drawn from large-scale mobility data across 11 U.S. cities. We show that comparisons with high quality time use surveys such as the ATUS can validate behavioral insights from mobility data, while quantifying uncertainty and highlighting areas of relative instability in analytical findings. We also propose a temporal re-weighting method that can complement existing bias-mitigation techniques to improve the accuracy of conclusions drawn from GPS-based mobility data.

A packed community of exponentially proliferating microbes will spread in size exponentially. However, due to nutrient depletion, mechanical constraints, or other limitations, exponential proliferation is not indefinite, and the spreading slows. Here, we theoretically explore a fundamental question: is it possible to infer the dominant limitation type from the spreading dynamics? Using a continuum active fluid model, we consider three limitations to cell proliferation: intrinsic growth arrest (e.g., due to sporulation), pressure from other cells, and nutrient access. We find that memoryless growth arrest still results in superlinear (accelerating) spreading, but at a reduced rate. In contrast, pressure-limited growth results in linear (constant-speed) spreading in the long-time limit. We characterize how the expansion speed depends on the maximum growth rate, the limiting pressure value, and the effective fluid friction. Interestingly, nutrient-limited growth results in a phase transition: depending on the nutrient supply and how efficiently nutrient is converted to biomass, the spreading can be either superlinear or sublinear (decelerating). We predict the phase boundary in terms of these parameters and confirm with simulations. Thus, our results suggest that when an expansion slowdown is observed, its dominant cause is likely nutrient depletion. More generally, our work suggests that cell-level growth limitations can be inferred from population-level dynamics, and it offers a methodology for connecting these two scales.

We address the problem of how individuals can integrate efficiently their private behavior with information provided by others within a group. To this end, we consider the model of collective search introduced in [https://doi.org/10.1103/PhysRevE.102.012402], under a minimal setting with no olfactory information. Agents combine a private exploratory behavior and a social imitation consisting in aligning to their neighbors, and weigh the two contributions with a single ``trust" parameter that controls their relative influence. We find that an optimal trust parameter exists even in the absence of olfactory information, as was observed in the original model. Optimality is dictated by the need to explore the minimal region of space that contains the target. An optimal trust parameter emerges from this constraint because it it tunes imitation, which induces a collective mechanism of inertia affecting the size and path of the swarm. We predict the optimal trust parameter for cohesive groups where all agents interact with one another. We show how optimality depends on the initialization of the agents and the unknown location of the target, in close agreement with numerical simulations. Our results may be leveraged to optimize the design of swarm robotics or to understand information integration in organisms with decentralized nervous systems such as cephalopods.

Determinations of the Hubble constant based on the local distance ladder remain in significant tension with early-Universe inferences from the cosmic microwave background. While this tension is often discussed in terms of new physics or unmodeled systematics, the role of the assumed priors on the model parameters has received comparatively little attention. Recently, Desmond et al. (2025) pointed out that the commonly adopted flat prior on distance moduli upweights smaller distances and systematically favors high inferred values of the Hubble constant. Motivated by this observation, we perform a comprehensive Bayesian recalibration of the distance ladder, applying physically motivated priors uniformly to all distances, including the Milky Way Cepheids, which are incorporated directly into the joint fit. Together with a conservative treatment of the Gaia EDR3 residual parallax offset, the Hubble constant shifts from $H_0 = 73.0 \pm 1.0 \, \mathrm{km/s/Mpc}$ to $H_0 = 70.6 \pm 1.0 \, \mathrm{km/s/Mpc}$, reducing the Hubble tension from $5 \, σ$ to $2 \, σ$. Our results show that the assumed priors -- often treated as innocuous defaults -- may play a central role in the Hubble tension. Because all local distance ladders rely on the calibration of distances, similar prior-driven effects are expected to arise across distance-ladder methods.

Understanding urban mobility requires models that capture how people interact with and navigate the built environment. We present a scalable, generalizable agent-based framework in which daily schedules emerge from the interplay between mandatory (e.g., work, school) and flexible (e.g., errands, food, leisure) activities, driven by evolving individual needs. The results of our model are validated against empirical patterns from the 2017 U.S. National Household Travel Survey, including activity distributions, origin-destination flows, and trip-chain length distributions. We introduce a normalized similarity metric to quantify agreement between simulated and empirical patterns. Most cities achieve scores above 0.80, demonstrating strong alignment without the need for city-specific calibration. The model scales efficiently to over 20 million agents, enabling full-population simulations of large metropolitan areas. This combination of universality and scalability enables scenario analysis for infrastructure stress testing, disaster recovery, innovation diffusion, and disease spread in urban systems.

In the aftermath of large-scale disasters, the scarcity of resources and the paralysis of infrastructure raise severe challenges to effective post-disaster recovery. Efficient coordination between shelters and victims plays a crucial role in building community resilience, yet the evolution of two-layer behavioral feedback between these two groups through network coupling remains insufficiently understood. Here, this study develops a two-layer network to capture the cross-layer coupling between shelters and victims. The upper layer uses a post-disaster emergency resource redistribution model within the framework of the public goods game, while the lower layer adopts a cooperative evolutionary game to describe internal victim interactions. Monte Carlo simulations on scale-free networks reveal threshold effects of incentives: moderate public goods enhancement and subsidies promote cooperation, whereas excessive incentives induce free-riding. In contrast, credible and well-executed punishment effectively suppresses defection. Targeted punishment of highly connected shelters significantly enhances cooperation under resource constraints. A comparative analysis using a network generated from the actual coordinates of Beijing shelters confirms the model's generality and practical applicability. The findings highlight the importance of calibrated incentives, enforceable sanctions, and structural targeting in fostering robust cooperation across organizational and individual levels in post-disaster environments.

In this work, we present Multimodal Equivariant Inverse Design Network (MEIDNet), a framework that jointly learns structural information and materials properties through contrastive learning, while encoding structures via an equivariant graph neural network (EGNN). By combining generative inverse design with multimodal learning, our approach accelerates the exploration of chemical-structural space and facilitates the discovery of materials that satisfy predefined property targets. MEIDNet exhibits strong latent-space alignment with cosine similarity 0.96 by fusion of three modalities through cross-modal learning. Through implementation of curriculum learning strategies, MEIDNet achieves ~60 times higher learning efficiency than conventional training techniques. The potential of our multimodal approach is demonstrated by generating low-bandgap perovskite structures at a stable, unique, and novel (SUN) rate of 13.6 %, which are further validated by ab initio methods. Our inverse design framework demonstrates both scalability and adaptability, paving the way for the universal learning of chemical space across diverse modalities.

Humans can infer accurate mechanical outcomes from only a few observations, a capability known as mechanics intuition. The mechanisms behind such data-efficient learning remain unclear. Here, we propose a reinforcement learning framework in which an agent encodes continuous physical observation parameters into its state and is trained via episodic switching across closely related observations. With merely two or three observations, the agent acquires robust mechanics intuition that generalizes accurately over wide parameter ranges, substantially beyond the training data, as demonstrated on the brachistochrone and a large-deformation elastic plate. We explain this generalization through a unified theoretical view: it emerges when the learned value function enforces Bellman consistency across neighboring task parameters, rendering the Bellman residual stationary with respect to physical variations. This induces a smooth policy that captures a low-dimensional solution manifold underlying the continuum of tasks. Our work establishes episodic switching as a principled route to artificial mechanics intuition and offers a theoretical link to similar generalization abilities in biological learners.

A new approach to incorporating coupling elements into a generalized coupled mode theory is presented. The simplest model of coupling of a structured waveguide with an external RF power source and load through loops and transmission lines was used. Even such a simple model significantly complicated the system of coupled equations. It turned into a coupled integro-differential system of the Barbashin type with degenerate kernels. Since the integral kernels are degenerate, this system is reduced to three independent systems of differential equations. Instead of solving a system of coupled integro-differential equations, we need to find solutions to three systems of ordinary differential equations. Two systems describe the distribution of the field excited by one loop and the specified value of the excitation current in it. In the first system the loop is located at the section's input, and in the second, at the section's output. The third system does not depend on the loop parameters at all. It describes the distribution of the field excited by an electron beam in a section without loops. Based on the developed analytical model, the computer code was developed for matching the loop couplers for the uniform accelerating sections of X-band. The calculation results were used to simulate the non-uniform section. Without additional matching, we obtained an input reflection coefficient of 8E-3.

Cycles are ubiquitous in various networks such as social, biological, and technological systems, where they play a significant functional and dynamical role. This paper proposes a node similarity measure based on minimal simple cycles, referred to as cycle similarity. Specifically, the metric quantifies the similarity between two nodes by considering the minimal cycles that connect them through their neighboring nodes, with an upper bound imposed on the cycle size to ensure computational feasibility. We then systematically examine the effectiveness and applicability of this similarity measure through two fundamental tasks: link prediction and community detection. To address the scarcity of cycles in link prediction, an edge-addition correction strategy is introduced, whereby the existence of a candidate edge is hypothetically assumed before computing node similarity. Experimental results demonstrate that this correction leads to improved performance on datasets including karate, INT, PPI, and Grid. In hierarchical community detection using cycle similarity, we find that the significance of cyclic structures (reflected by Z-scores), the presence of pendant nodes with degree one, and the existence of cut vertices are the primary factors influencing the algorithm's performance.

The regeneration of oral and craniofacial bone defects ranging from minor periodontal and peri-implant defects to large and critical lesions imposes a substantial global health burden. Conventional therapies are associated with several limitations, highlighting the development of a unique treatment strategy, such as tissue engineering. A well-designed scaffold for bone tissue engineering should possess biocompatibility, biodegradability, mechanical strength, and osteoconductivity. For this purpose, mesoporous silica nanoparticles (MSNs) were synthesized and incorporated at different ratios (10, 20, and 30%) into alginate/chitosan (Alg/Chit)-based porous composite scaffolds fabricated through the freeze-drying method. The MSN incorporation significantly improved the mechanical strength of the scaffolds while showing a negligible decreasing effect on the porosity. All of the samples showed desirable swelling behaviors, which is beneficial for cell attachment and proliferation. The MSN-containing scaffolds indicated a decreased hydrolytic degradation in an MSN percentage-dependent manner. The fabricated scaffolds did not depict cytotoxic characteristics. The Alg/Chit/MSN30 scaffolds not only showed noncytotoxic properties, but also increased the cell viability significantly compared to the control group. The biomineralization properties of the MSN-containing nanocomposite scaffolds were significantly higher than the Alg/Chit composite, suggesting the potential of these nanoparticles for bone tissue engineering applications. Taken together, it is concluded that the Alg/Chit/ MSN30 scaffolds are considerable substances for bone tissue regeneration, and MSN has a great tissue engineering potential in addition to its extensive biomedical applications.

Self-supervised fMRI foundation models have shown promising transfer performance, yet most rely on predefined region-level parcellations that discard fine-grained voxel information and introduce atlas-dependent biases. We propose Omni-fMRI, an atlas-free foundation model that operates directly on voxel-level signals. To enable scalable pretraining on 49,497 fMRI sessions across nine datasets, Omni-fMRI introduces a dynamic patching mechanism that substantially reduces computational cost while preserving informative spatial structure. To support reproducibility and fair comparison, we establish a comprehensive benchmark suite spanning 11 datasets and a diverse set of resting-state and task-based fMRI tasks. Experimental results demonstrate that Omni-fMRI consistently outperforms existing foundation models, providing a scalable and reproducible framework for atlas-free brain representation learning. Code and logs are available.

Tactile localization is the seemingly simple ability to 'tell' where a touch has occurred. However, how this ability is assessed, and what conclusions are drawn from experiments, depends on the theoretical ideas that inspire the research. Here, we review both theoretical frameworks and methodological approaches based on a systematic web-based literature search on tactile localization. After presenting current theories of tactile localization, we discuss task characteristics that differentiate current methodology for tactile localization into at least 8 distinct types of experimental tasks. We describe these tasks, discuss their, often implicit, underlying assumptions and cognitive requirements, and relate them to the theoretical approaches. We then compare, in an exemplary manner, the tactile localization results reported by a subset of studies and demonstrate how some methods are associated with specific biases, illustrating that the choice of experimental method significantly affects the conclusions drawn from the results. Our review suggests that the field currently lacks a clear concept of the specific processes induced by the various experimental tasks and, thus, calls for concerted efforts to clarify and unify currently diverse, fragmented, and partly inconsistent theoretical underpinnings of tactile spatial processing, flanked by dedicated data sharing to allow across-study analysis.

Dynamic models are widely used to mathematically describe biological phenomena that evolve over time. One important area of application is leukaemia research, where leukaemia cells are genetically modified in preclinical studies to explore new therapeutic targets for reducing leukaemic burden. In advanced experiments, these studies are often conducted in mice and generate time-resolved data, the analysis of which may reveal growth-inhibiting effects of the investigated gene modifications. However, the experimental data is often times evaluated using statistical tests which compare measurements from only two different time points. This approach does not only reduce the time series to two instances but also neglects biological knowledge about cell mechanisms. Such knowledge, translated into mathematical models, expands the power to investigate and understand effects of modifications on underlying mechanisms based on experimental data. We utilise two population growth models -- an exponential and a logistic growth model -- to capture cell dynamics over the whole experimental time horizon and to consider all measurement times jointly. This approach enables us to derive modification effects from estimated model parameters. We demonstrate that the exponential growth model recognises simulated scenarios more reliably than the other candidate model and than a statistical test. Moreover, we apply the population growth models to evaluate the efficacy of candidate gene knockouts in patient-derived xenograft (PDX) models of acute leukaemia.

Recent advances in metagenomics have revealed macroecological patterns or "laws" describing robust statistical regularities across microbial communities. Stochastic logistic models (SLMs), which treat species as independent -- akin to ideal gases in physics -- and incorporate environmental noise, reproduce many single-species patterns but cannot account for the pairwise covariation observed in microbiome data. Here we introduce an interacting stochastic logistic model (ISLM) that minimally extends the SLM by sampling an ensemble of random interaction networks chosen to preserve these single-species laws. Using dynamical mean-field theory, we map the model's phase diagram -- stable, chaotic, and unbounded-growth regimes -- where the transition from stable fixed-point to chaos is controlled by network sparsity and interaction heterogeneity via a May-like instability line. Going beyond mean-field theory to account for finite communities, we derive an estimator of an effective stability parameter that quantifies distance to the edge of instability and can be inferred from the width of the distribution of pairwise covariances in empirical species-abundance data. Applying this framework to synthetic data, environmental microbiomes, and human gut cohorts indicates that these communities tend to operate near the edge of instability. Moreover, gut communities from healthy individuals cluster closer to this edge and exhibit broader, more heterogeneous associations, whereas dysbiosis-associated states shift toward more stable regimes -- enabling discrimination across conditions such as Crohn's disease, inflammatory bowel syndrome, and colorectal cancer. Together, our results connect macroecological laws, interaction-network ensembles, and May's stability theory, suggesting that complex communities may benefit from operating near a dynamical phase transition.

The BioModels database is one of the premier databases for computational models in systems biology. The database contains over 1000 curated models and an even larger number of non-curated models. All the models are stored in the machine-readable format, SBML. Although SBML can be translated into the human readable Antimony format, analyzing the models can still be time consuming. In order to bridge this gap, a LLM (large language model) assistant was created to analyze the BioModels and allow interaction between the user and the model using natural language. By doing so, a user can easily and rapidly extract the salient points in a given model. Our analysis workflow involved 'chunking' BioModels and converting them to plain text using llama3, and then embedding them in a ChromaDB database. The user-provided query was also embedded, and a similarity search was performed between the query and the BioModels in ChromaDB to extract the most relevant BioModels. The BioModels were then used as context to create the most accurate output in the chat between the user and the LLM. This approach greatly minimized the chance of hallucination and kept the LLM focused on the problem at hand.

The efficacy of anti-cancer therapies is severely limited by the emergence of drug resistance. While genetic drivers are well-characterized, growing evidence suggests that non-genetic mechanisms, particularly those involving drug-tolerant persisters (DTPs), play a pivotal role in solid tumor relapse. To elucidate the evolutionary dynamics of DTP-induced resistance, we develop a stochastic agent-based model (ABM) of solid tumor evolution that couples macroscopic population dynamics with microscopic epigenetic state inheritance during the cell cycle. Our simulations accurately reproduce the temporal progression of relapse observed in experimental studies, capturing the dynamic transition from sensitive cells to DTPs, and ultimately to stable resistant phenotypes under prolonged therapy. By explicitly modeling the epigenetic plasticity of individual cells, our model bridges the gap between cellular heterogeneity and population-level tumor evolution. Furthermore, we performed \textit{in silico} clinical trials using virtual patient cohorts to evaluate therapeutic outcomes, demonstrating that optimized adaptive treatment strategies can significantly delay tumor relapse compared to standard dosing. This study provides a quantitative framework for dissecting DTP-driven resistance mechanisms and designing more effective, biologically informed therapeutic strategies.

Large language models (LLMs) trained with next-word-prediction have achieved success as clinical foundation models. Representations from these language backbones yield strong linear probe performance across biomedical tasks, suggesting that patient semantics emerge from next-token prediction at scale. However, this paradigm treats patients as a document to be summarized rather than a dynamical system to be simulated; a patient's trajectory emerges from their state evolving under interventions and time, requiring models that simulate dynamics rather than predict tokens. To address this, we introduce SMB-Structure, a world model for structured EHR that grounds a joint-embedding prediction architecture (JEPA) with next-token prediction (SFT). SFT grounds our model to reconstruct future patient states in token space, while JEPA predicts those futures in latent space from the initial patient representation alone, forcing trajectory dynamics to be encoded before the next state is observed. We validate across two large-scale cohorts: Memorial Sloan Kettering (23,319 oncology patients; 323,000+ patient-years) and INSPECT (19,402 pulmonary embolism patients). Using a linear probe evaluated at multiple points along the disease trajectory, we demonstrate that our training paradigm learns embeddings that capture disease dynamics not recoverable by autoregressive baselines, enabling SMB-Structure to achieve competitive performance on complex tasks characterized by high patient heterogeneity. Model weights are available at https://huggingface.co/standardmodelbio/SMB-v1-1.7B-Structure.

Phylogenetic reconciliation seeks to explain host-symbiont co-evolution by mapping parasite trees onto host trees through events such as cospeciation, duplication, host switching, and loss. Finding an optimal reconciliation that ensures time feasibility is computationally hard when timing information is incomplete, and the complexity remains open when host switches are restricted by a fixed maximum distance $d$. While the case $d=2$ is known to be polynomial, larger values are unresolved. In this paper, we study the cases $d=3$ and $d=4$. We show that although arbitrarily large cycles may occur, it suffices to check only bounded-size cycles (we provide a complete list), provided the reconciliation satisfies acyclicity (i.e., time-feasibility) in a stronger sense. These results do not resolve the general complexity, but highlight structural properties that advance the understanding of distance-bounded reconciliations.

Generalisation refers to the ability of a machine learning (ML) model to successfully apply patterns learned from training data to new, unseen data. Quantum devices in the current Noisy Intermediate-Scale Quantum (NISQ) era are inherently affected by noise, which degrades generalisation performance. In this work, we derive upper and lower margin-based generalisation bounds for Quantum Kernel-Assisted Support Vector Machines (QSVMs) under local depolarising noise. These theoretical bounds characterise noise-induced margin decay and are validated via numerical simulations across multiple datasets, as well as experiments on real quantum hardware. We further justify the focus on margin-based measures by empirically establishing margins as a reliable indicator of generalisation performance for QSVMs. Additionally, we motivate the study of local depolarising noise by presenting empirical evidence demonstrating that the commonly used global depolarising noise model is overly optimistic and fails to accurately capture the degradation of generalisation performance observed in the NISQ era.

Although photons are robust, room-temperature carriers well suited to quantum machine learning, the absence of photon-photon interactions hinder the realization of memory functionalities that are critical for capturing long-range context. Recently, measurement-based implementations of photonic quantum memristors (PQMRs) have enabled tunable non-Markovian responses. However, their memory remains confined to local elements, in contrast to biological or artificial networks where memory is shared across the system. Here, we propose a scalable PQMR network that enables measurement-based memory sharing. Each memristive node updates its internal state using the history of its own and neighbouring quantum states, thereby realizing distributed memory. By modelling each node as a photonic quantum memtransistor, we demonstrate pronounced enhancements in both classical and quantum hysteresis at the device level, as well as enhanced network-level quantum hysteresis. Implemented as a quantum reservoir, the architecture achieves improved Fashion-MNIST classification accuracy and confidence via increased data separability. Our approach paves the way toward high-capacity quantum machine learning using memristive devices compatible with linear-optical quantum computing.

Phase sensing with entangled multiqubit states in the presence of noise is a central theme of modern quantum metrology. The present work investigates Dicke state superposition probes for quantum phase sensing under parameter encoding generated by one- and two-body interaction Hamiltonians. A class of N-qubit Dicke superposition states that exhibit near-Heisenberg scaling, of the quantum Fisher information, while maintaining significantly enhanced robustness to dephasing noise compared to GHZ, W-superposition, and balanced Dicke states, under unitary encodings generated by one-body interaction Hamiltonians are identified. For two-body interactions, Dicke superposition probes optimizing the quantum Fisher information are identified, and their performance under phase-damping, amplitude-damping, and global depolarizing noise is explored. Within this family, certain Dicke superpositions are found to combine super-Heisenberg scaling with improved resilience to phase damping relative to Fisher information optimal probes. These results establish tailored near-optimal Dicke-state superposition probes as versatile and noise-resilient resources for Heisenberg and super-Heisenberg quantum phase sensing governed by one- and two-body interactions.

We demonstrate high-fidelity, tunable, and ultrafine-resolution on-chip frequency beamsplitters using a quantum frequency processor based on an integrated pulse shaper with six spectral channels. Near-ideal Hadamard gate performance is achieved, with fidelity F > 0.9995 and modified success probability P > 0.9621 maintained across frequency spacings from 2-5 GHz and down to as few as four spectral pulse shaper channels. The system's support of frequency spacings as narrow as 2 GHz significantly surpasses prior bulk demonstrations and enables arbitrary splitting ratios via spectral phase or modulation index control. These results establish a scalable and resource-efficient platform for integrated frequency-bin quantum photonics, opening new directions in quantum information processing, including densely parallel single-qubit operations and multidimensional gate implementations.

Gravitational memory is a zero-frequency effect associated with a permanent change in the asymptotic spacetime metric induced by radiation. While its universal manifestation is a net change of proper distances, gravitational-wave detectors are intrinsically insensitive to the final offset and can only probe the associated transition. A central challenge for any claim of detection therefore lies in defining a physically meaningful and operationally robust model of this time-dependent signal, which is uniquely attributable to gravitational memory and distinguishable from purely oscillatory radiation. In this work, we propose a general solution to this challenge. Building on a self-contained review of the theory of gravitational memory, we discuss a theoretical framework for defining and modeling a gravitational memory rise, in particular applicable to compact binary coalescences. Specializing to space-based detectors, we analyze the response of LISA to gravitational radiation including a memory contribution, with particular emphasis on mergers of supermassive black hole binaries, which offer the most promising prospects for a first single-event detection. The framework developed here provides the theoretical foundation for statistically well-defined hypothesis testing between memory-free and memory-full radiation and quantitative assessments of detection prospects. As such, these results establish a principled pathway toward a future observational claim of gravitational memory.

This paper investigates the photon ring and shadow structure of the Reissner-Nordström black hole in the scalar-tensor-vector gravitational framework. The black hole is characterized by the ( MOG) parameter (α) and the charge (Q). The study finds that as (α) increases, the event horizon radius (r_h), photon sphere radius (r_{ph}), and critical impact parameter (b_{ph}) all increase, while these decrease as (Q) increases. The innermost stable circular orbit radius (r_{isco}) exhibits similar monotonic behavior. Ray-tracing shows that as (Q) increases, the impact parameter (b) interval between the lensing ring and photon ring widens; (b_{\text{ph}}) is non-degenerate, and the photon ring radius is uniquely determined by (α) and (Q). Using $EHT$ constraints on (SgrA^*) and (M87^*), the bounds on (α) and (Q) are derived. For (Q = 0), (0.5), and (1), the allowed ranges are (α\in [0, 0.06]), ([0, 0.11]), and ([0.19, 0.36]), respectively. Radiative simulations show that for fixed (Q), larger (α) leads to a larger, non-degenerate photon ring. The Schwarzschild case is approached only when both (α) and (Q) are small. This provides a computational basis for testing modified black holes and offers a non-degenerate observational criterion for distinguishing quantum gravity models, consistent with current $EHT$ data. Future observations with $ngEHT$ and multi-band polarization can further test this. The results suggest that studying the photon ring structure of a Reissner-Nordström black hole in scalar-tensor-vector gravity provides a unique optical diagnostic for potential quantum-gravity tests and black hole properties.

We directly observe and spatially map an optical analog of the Magnus effect, where intrinsic spin-orbit-like coupling of light generates a spin-dependent transverse displacement of the atom-light interaction profile for a $^{40}$Ca$^+$ ion. Probed on a quadrupole transition using a tightly focused beam, we observe displacements of the maximum in the profile of the effective interaction by several 100 nm originating from intrinsic longitudinal electric field components beyond the paraxial approximation. The tight focus of the beam induces additional transverse polarization gradients, which we characterize through a phase-sensitive measurement and spatial maps for different beam configurations. The results establish the physical basis of polarization-gradient interactions relevant to optical tweezer-based quantum control.

We employ the wavelet formalism of quantum field theory to study field theories in the nonperturbative Hamiltonian framework. Specifically, we make use of Daubechies wavelets in momentum space. These basis elements are characterised by a resolution and a translation index that provides for a natural nonperturbative infrared and ultraviolet truncation of the quantum field theory. As an application, we consider the $φ^4$ theory and demonstrate the emergence of the well-known nonperturbative strong-coupling phase transition in the $m^2>0$ sector.

Magic states are a foundational resource for universal quantum computation. To survive in a realistic noisy environment, magic states must be prepared fault-tolerantly and protected by a quantum error-correcting code. The recent discovery of highly efficient quantum low-density parity-check codes, together with efficient logic gates, lays the groundwork for low-overhead fault-tolerant quantum computation. This motivates the search for fast and parallel protocols for logical magic state preparation to enable universal quantum computation. Here, we introduce a fast code surgery procedure that performs a fault-tolerant measurement of many transversal logic gates in parallel. This is achieved by performing a generalized gauging measurement on a quantum code that supports a higher-form transversal gate. The time overhead of our procedure is constant, and the qubit overhead is linear. The procedure inherits fault-tolerance properties from the base code and the structure of the higher-form transversal gate. When applied to codes that support higher-form Clifford gates our procedure achieves fast and fault-tolerant preparation of many magic states in parallel. This motivates the search for good quantum low-density parity-check codes that support higher-form Clifford gates.

We present exact solutions of the Einstein-$SU(2)$ non-linear sigma model in $3+1$ spacetime dimensions, describing bumpy black holes and black branes. Using an Ansatz for superfluid pion multi-vortices, the matter sector reduces to a first-order BPS system, while the Einstein equations reduce to a Liouville equation with a smooth source governing the horizon deformation. These solutions describe horizons of different constant curvatures, with nontrivial bumpy geometries protected by an integer topological invariant, namely the vorticity, which also controls the number of bumps and the black hole thermodynamics. Remarkably, such horizons arise in a minimal and physically motivated matter model, without invoking exotic fields or modified gravity. The physical implications of these results in holography and astrophysics are briefly described.

We identify a new cosmological coincidence that parallels the well-known matter/dark-energy coincidence: the present-epoch transition of the universe from a weakly coupled (collisionless) to a strongly coupled (collisional) gravitational regime. Within a cosmological model containing equal amounts of positive and negative Bondi masses -- consistent with the weak equivalence principle and momentum conservation -- we show that this coupling transition naturally coincides with the shift from a coasting to an accelerating expansion. A linear response analysis of the corresponding Vlasov-Poisson system reveals that mixed positive-negative mass configurations are always unstable, with growth rates that increase at shorter wavelengths, thereby driving the system toward strong coupling. Using long-time, exact one-dimensional N-body simulations, we demonstrate that the universe undergoes three successive expansion phases: an initial ballistic regime, an intermediate random-walk acceleration driven by sporadic Bondi encounters, and finally a uniformly accelerating phase triggered by the formation of stable positive/negative mass pairs. The onset of this last phase occurs precisely when the coupling parameter crosses unity, linking the two cosmological coincidences through a single dynamical mechanism. These results suggest that cosmic acceleration may arise from the nonlinear dynamics of a gravitationally neutral mixed-mass universe, without invoking dark energy or a cosmological constant.

We improve on our version of the second law of thermodynamics as a deterministic theorem for quantum spin systems in two basic aspects. The first concerns the general statement of the second law: spontaneous changes in an adiabatically closed system will always be in the direction of increasing mean entropy, which rises to a maximal value. Two specific examples concern the transition from pure to mixed states in two different universality classes of dynamics in one dimension, one being the exponential model, the other the Dyson model, the dynamics of the latter exhibiting strong graphical evidence of quantum chaos, as a consequence of the results of Albert and Kiessling on the Cloitre function.

Bell conditions for local realism are critically revisited. In particular for optical experiments I criticize Bell's proposed response of detectors to signals as extremely idealized. More physical conditions are proposed, whence a realistic local model of an optical experiment is possible which violates the Clauser-Horne (Bell) inequality. The possibility rests on the existence of a coincidence-time loophole in the experiments.

We propose and investigate a scheme for the steady-state emission of quantum-correlated, telecom-band light from a single multilevel atom. By appropriately tuning the frequency of a pair of lasers, a two-photon transition is continually driven to an atomic excited state that emits photons at the desired wavelength. We show that resonantly coupling a cavity mode to the telecom transition can enhance the rate of emission while retaining the antibunched counting statistics that are characteristic of atomic light sources. We also explore coupling a second, independent cavity mode to the atom, which increases the telecom emission rate and introduces quantum correlations between the cavity modes. A model for the hyperfine structure of a single cesium atom is then described and numerically integrated to demonstrate the viability of implementing the scheme with a modern cavity QED system.

We develop a quantum theory for the scattering of squeezed coherent light by a dissipative dielectric slab. Using the Green-function quantization approach, we derive the transformation of the field quadratures and show how dispersion, absorption, and multiple reflections distort the incident squeezing. We find that the slab can selectively attenuate or amplify quadrature noise depending on the slab parameters and provide expressions for the output power spectra.

Quantum error mitigation (QEM) infers noiseless expectation values by combining outcomes from intentionally modified, noisy variants of a target quantum circuit. Unlike quantum error correction, QEM requires no additional hardware resources and is therefore routinely employed in experiments on contemporary quantum processors. A central limitation of QEM is its substantial sampling overhead, which necessitates long execution times where device noise may drift, potentially compromising the reliability of standard mitigation protocols. QEM strategies based on agnostic noise amplification (ANA) are intrinsically resilient to such noise variations, but their sampling cost remains a major practical bottleneck. Here we introduce a mitigation framework that combines virtual noise scaling with a layered mitigation architecture, yielding orders of magnitude reduction in runtime overhead compared to conventional zero-noise extrapolation post-processing. The proposed approach is compatible with dynamic circuits and can be seamlessly integrated with error detection and quantum error correction schemes. In addition, it naturally extends to ANA-based mitigation of mid-circuit measurements and preparation errors. We validate our post-processing approach by applying it to previously reported experimental data, where we observe a substantial improvement in mitigation efficiency and accuracy.

This paper highlights cosmologically viable sine and cosine hyperbolic evolution functions in the framework of $f(Q,\mathcal{L}_m)$ gravity. The models have been tested to check the behavior of the equation of state (EoS) parameter under the variation of parametric values. The EoS parameter experiences a quintessence phase, and is approaching to $-1$ at late time. The models are showing inclined behaviour with the $Λ$CDM model at the late time. The viability of both the models is retested using the widely accepted energy conditions in both cases. The violation of the strong energy condition admits the accelerating behaviour of the models. The same has been explained through the analysis of the profile of deceleration parameter, which concretely supports the evidence that the models explain early deceleration to late time acceleration phenomena.

Undecidability, a hallmark of Gödel incompleteness theorems, has recently emerged in quantum many-body physics through the spectral gap problem. We demonstrate how this logical limitation can be holographically transmitted to a class of gravitational theories via the AdS/CFT correspondence. By embedding a translationally invariant spin Hamiltonian with undecidable gap status into a large-N gauge theory, we generate an AdS dual in which the selection of dominant bulk saddle (Poincaré AdS or AdS soliton) is itself undecidable. Consequently, under standard semiclassical holographic assumptions, even determining which smooth spacetime geometry emerges from quantum gravity can be beyond the limits of computability.

We argue that the polynomial degeneracies of curvature invariants can act as geometric selection rules for spacetime singularities in modified theories of gravity. The degeneracies arise purely from the algebraic structure of Riemannian geometry and impose non-trivial constraints on the effective energy-momentum tensor. We derive these constraints for metric $f(R)$ gravity and a wide class of scalar-tensor theories to show that a singularity formation is generally occluded by curvature and/or scalar-induced anisotropies. Therefore, formation of a singularity in modified theories of gravity is not always a generic outcome but can occur only along algebraically admissible branches selected by Riemannian curvature invariants.

Classical mechanics admits multiple equivalent formulations, from Newton's equations to the variational Lagrange-Hamilton framework and the scalar Hamilton-Jacobi (HJ) theory. In the HJ formulation, classical ensembles evolve through the continuity equation for a real density $ρ= R^{2}$ coupled to Hamilton's principal function $S$. Here we develop a complementary formulation, the Hamilton-Jacobi-Schrödinger (HJS) theory, by embedding the pair $(R,S)$ into a single complex field. Starting from a completely general complex ansatz $ψ= f(R,S) e^{i g(R,S)}$, and imposing two minimal structural requirements, we obtain a unique map $ψ= R e^{iS/κ}$ together with a linear HJS equation whose $|κ| \to 0$ limit reproduces the HJ formulation exactly. Remarkably, when $\mathrm{Re}(κ)\neq 0$, essential features of quantum mechanics, including superposition, operator algebra, commutators, the Heisenberg uncertainty principle, Born's rule, and unitary evolution, arise naturally as consistency conditions. HJS thus provides a unified mathematical viewpoint in which classical and quantum dynamics appear as different limits of a single underlying structure.

Accretion discs that are tilted with respect to their compact hosts can warp out-of-plane through general relativistic frame-dragging. Warp influences disc dynamics in ways that have been studied extensively, especially as regards instabilities that might lead to rapid angular-momentum cancellation between neighbouring rings of fluid and mass infall. We provide a review of warped-disc phenomena here, revisiting key hydrodynamical assumptions that impact calculations of the shear viscosity controlling instability thresholds. Relativistic effects at the level of gas-parcel orbits are included, as are external Lorentz forces applied by the compact primary's magnetic field. Semi-analytic analysis reveals that intense magnetic fields can bring about new branches of warp modes and avoided crossings that significantly reduce the perpendicular viscosity at sub-Eddington accretion rates. Critical strengths required for misaligned torques to tear a thin disc may thus relax for systems like neutron star X-ray binaries or radio-loud active galactic nuclei.

We investigate the one-loop effective action for a test scalar field in a general Friedmann-Lemaître-Robertson-Walker (FLRW) background, specifically focusing on quantum corrections up to the second order in the interaction strength. By employing the Schwinger-Keldysh formalism, we derive the equation of motion for the field expectation value, which incorporates not only the standard local radiative corrections but also novel nonlocal features: a memory term and a stochastic noise term. We identify all ultraviolet divergent structures within these nonlocal terms and provide a consistent renormalization procedure. To analyze the physical impact of these terms, we apply a local approximation under the assumption of slowly-varying fields, by which the memory term acts as a negative contribution to the drift coefficient. As a concrete application, we consider a massive $φ^4$ theory and show that these one-loop corrections lead to a suppression of the field variance in the infrared regime compared to the tree-level results.

Conventionally, dispersion forces mediated by quantum vacuum fluctuations are known to exhibit universal distance scalings, with retardation typically leading to a faster decay of the interaction. Here, we show that this expectation fails for intrinsically discrete systems. Using the microscopic scattering approach, we study the Casimir-Polder interaction between two atomic arrays, and uncover an unconventional distance scaling in which the force crosses over from a faster decay at short separations to a slower decay in the retarded regime. This behavior originates from the discrete lattice structure and can be consistently understood within the scattering picture. Extending our analysis to Rydberg atomic arrays, we predict an even stronger deviation from conventional scaling and propose an experimentally feasible scheme for direct measurement. Our results provide a new platform for exploring dispersion forces beyond the continuum limit.

In this work, we introduce a unified method to characterize and measure multipartite entanglement using the framework of thermodynamics. A family of the new entanglement measures is proposed: \textit{ergotropic-gap concentratable entanglement}. Furthermore, we establish that ergotropic-gap concentratable entanglement constitutes a well-defined entanglement measure within a specific parameter regime, satisfying key properties including continuity, majorization monotonicity and monogamy. We demonstrate the utility of this measure by showing it effectively distinguishes between multi-qubit Greenberger-Horne-Zeilinger states and W states. It also proves effective in detecting entanglement in specific classes of four-partite star quantum network states.

Traversable wormhole solutions are explored in $f(\mathcal{R},\mathbb{T})$ gravity, a curvature--matter extension in which $\mathcal{R}$ is the Ricci scalar and $\mathbb{T}$ denotes the trace of the energy--momentum tensor. To generate explicit wormhole models, we prescribe holographic dark-energy densities based on entropy formalism proposed by Rényi, Moradpour, and Bekenstein--Hawking, namely \[ ρ_{\textit{R}} = \fracα{4α_1 r^4 c^2 κ}\ln\!\left(1+πα_1 r^2\right), \qquad ρ_{\textit{M}} = \fracα{4πr^2 c^2 κ\left(πα_1 r^2 + 1\right)}, \qquad ρ_{\textit{BH}} = \fracα{4 c^2 κr^2}, \] with $α$ and $β$ carrying dimensions of $L^{-2}$. The corresponding shape functions obtained from the field equations satisfy the standard throat and flare-out requirements for traversability. We then study how varying $α$ and $β$ affects (i) the balance of forces associated with equilibrium and (ii) the status of the energy conditions. In particular, the null energy condition is found to be violated, indicating that exotic matter (or an effective exotic sector) is required to support the wormhole geometry. The spatial structure of the solutions is further visualized through embedding surfaces.

Entanglement concurrence is an important bipartite entanglement measure that has found wide applications in quantum technologies. In this work, inspired by unified entropy, we introduce a two-parameter family of entanglement measures, referred to as the unified $(q,s)$-concurrence. Both the standard entanglement concurrence and the recently proposed $q$-concurrence emerge as special cases within this family. By combining the positive partial transposition and realignment criteria, we derive an analytical lower bound for this measure for arbitrary bipartite mixed states, revealing a connection to strong separability criteria. Explicit expressions are obtained for the unified $(q,s)$-concurrence in the cases of isotropic and Werner states under the constraint $q>1$ and $qs\geq 1$. Furthermore, we explore the monogamy properties of the unified $(q,s)$-concurrence for $q\geq 2$, $0\leq s\leq 1$ and $1\leq qs\leq 3$, in qubit systems. In addition, we derive an entanglement polygon inequality for the unified $(q,s)$-concurrence with $q\geq 1$ and $qs\geq 1$, which manifests the relationship among all the marginal entanglements in any multipartite qudit system.

Quantum $(r,δ)$-locally recoverable codes ($(r,δ)$-LRCs) are the quantum version of classical $(r,δ)$-LRCs designed to recover multiple failures in large-scale distributed and cloud storage systems. A quantum $(r,δ)$-LRC, $Q(C)$, can be constructed from an $(r,δ)$-LRC, $C$, which is Euclidean or Hermitian dual-containing. This article is devoted to studying how to get quantum $(r,δ)$-LRCs from BCH and homothetic-BCH codes. As a consequence, we give pure quantum $(r,δ)$-LRCs which are optimal for the Singleton-like bound.

Ultraspinning black holes have attracted considerable attention due to their super-entropic nature, and previous analyses -- mostly restricted to neutral cases and high-temperature regimes -- have suggested that such black holes are always thermodynamically unstable. In this work, we revisit the thermodynamic stability of ultraspinning black holes by performing a systematic analysis of the heat capacity in different ensembles over the full range of the horizon radius $r_H$, which were missed in earlier temperature-based analyses. We demonstrate for the first time that, contrary to earlier claims, ultraspinning black holes can admit thermodynamically stable regions, whose existence crucially depends on the spacetime dimension, the solution branch, and the presence of charge. In addition, we present the first application of the revised reverse isoperimetric inequality to ultraspinning black holes. Despite the violation of the original reverse isoperimetric inequality in this super-entropic regime, we find that the revised inequality remains applicable and imposes nontrivial constraints on the allowed parameter space, including an upper bound on the ultraspinning parameter $μ$, strengthened lower bounds on the mass $m$, and upper bounds on both the charge $q$ and the AdS radius $l$. To ensure the consistency of the thermodynamic description, the conserved charges and the first law in the ultraspinning limit are derived using the Iyer-Wald formalism together with integrability conditions.

Accurate classification of multipartite entanglement in high-dimensional quantum systems is crucial for advancing quantum communication and information processing. However, conventional methods are resource-intensive, and even many machine-learning-based approaches necessitate large training datasets, creating a significant experimental bottleneck for data acquisition. To address this challenge, we propose a hybrid neural network architecture integrating Convolutional and Bidirectional Long Short-Term Memory networks (CNN-BiLSTM). This design leverages CNNs for local feature extraction and BiLSTMs for sequential dependency modeling, enabling robust feature learning from minimal training data. We investigate two fusion paradigms: Architecture 1 (flattening-based) and Architecture 2 (dimensionality-transforming). When trained on only 100 samples, Architecture 2 maintains classification accuracies exceeding 90% for both 3-qubit and 4-qubit systems, demonstrating rapid loss convergence within tens of epochs. Under full-data conditions (400 000 samples), both architectures achieve accuracies above 99.97%. Comparative benchmarks reveal that our CNN-BiLSTM models, especially Architecture 2, consistently outperform standalone CNNs, BiLSTMs, and MLPs in low-data regimes, albeit with increased training time. These results demonstrates that the tailored CNN-BiLSTM fusion significantly alleviates experimental data acquisition burden, offering a practical pathway toward scalable entanglement verification in complex quantum systems.

We analyze the self-thermalization dynamics of the $M$-site Bose-Hubbard model in terms of the single-particle density matrix that is calculated by using the pseudoclassical approach. It is shown that a weak inter-particle interaction, which suffices to convert the integrable system of non-interacting bosons into a chaotic system, has a negligible effect on the thermal density matrix given by the Bose-Einstein distribution. This opens the door for equilibration where the two coupled Bose-Hubbard systems, which are initially in different thermal states, relax to the same thermal state. When we couple these two subsystems by using a lattice of the length $L\ll M$, we numerically calculate the quasi-stationary current of Bose particles across the lattice and show that its magnitude is consistent with the solution of the master equation for the boundary driven $L$-site Bose-Hubbard model.

We investigate the chaos in the dynamics of a probe massless particle confined by the harmonic potential near the horizon of the dyonic $\rm{AdS_4}$-Reissner-Nordström black hole. The total energy of the particle, chemical potential and magnetic field in this system serving as independently adjustable parameters tune nonlinearity and phase-space structure. By analyzing the trajectories on the Poincaré section and evaluating the Lyapunov exponents, we obtain the dynamical phase diagrams of the chaos and find their counteracting regulatory role: at low energy, chaos is enhanced and the Lyapunov exponent $λ_L$ violates its upper bound (i.e. surface gravity) in the extremal black hole limit(combined paramete $Γ=3$); at high energy, the same extremal limit suppresses chaos, with $λ_L$ dropping to zero and a regular dynamical corridor emerging along $Γ=3$ in the dynamical phase diagrams. These results establish a direct mapping between black hole thermodynamics and microscopic chaos, offering new insights into the AdS/QCD correspondence and nonlinear dynamics in strongly curved spacetimes.

Quantum sensing leverages quantum resources to surpass the standard quantum limit, yet many existing protocols rely on the preparation of complex entangled states and Hamiltonian engineering, posing challenges for universality and scalability. Here, we report an experimental realization of a universal protocol, known as Butterfly Metrology, proposed in [arXiv:2411.12794], demonstrating a scrambling-based approach for quantum-enhanced sensing on a superconducting quantum processor. By exploiting many-body information scrambling, we observe quantum-enhanced sensitivity to an encoded phase beyond the standard quantum limit, with a scaling consistent with a factor-of-two of the Heisenberg limit for system sizes of up to 10 qubits. Importantly, we experimentally establish a connection between the enhanced sensitivity and the dynamics of the out-of-time-order correlator (OTOC), and show that the buildup of scrambling-induced genuine multipartite entanglement underlies the observed sensitivity enhancement. Our results demonstrate a scalable and practical approach for quantum-enhanced sensing in interacting many-body quantum systems.

Quantum low-density parity-check (qLDPC) codes promise constant-rate, linear-distance families with bounded-weight checks, and recent work has realized transversal or constant-depth non-Clifford gates on various (often non-LDPC) codes. However, no explicit \emph{qubit} qLDPC family is known that simultaneously has constant rate, linear distance, bounded stabilizer weight, and a native \emph{magic-state fountain} that prepares many non-Clifford resource states in constant depth. We take a structural approach and identify coding-theoretic conditions under which a CSS qLDPC family necessarily supports a constant-depth $\CCZ$ magic-state fountain. The key ingredients are: (i) an algebraic notion of \emph{magic-friendly triples} of $X$-type logical operators, defined by pairwise orthogonality and a triple-overlap form controlling diagonal $\CCZ$ phases, and (ii) a 3-uniform hypergraph model of physical $\CCZ$ circuits combined with a packing lemma that turns large collections of such triples with bounded overlaps into bounded-degree hypergraphs. Our main theorem shows that if a CSS code family on $n$ qubits admits $Ω(n^{1+γ})$ magic-friendly triples whose supports have bounded per-qubit participation, then there exists a constant-depth circuit of physical $\CCZ$ gates implementing $Ω(n^γ)$ logical $\CCZ$ gates in parallel while preserving distance up to a constant factor. For asymptotically good qLDPC families such as quantum Tanner codes, this reduces the existence of a native $\CCZ$ magic-state fountain to a concrete combinatorial problem about counting and distributing magic-friendly triples in the logical $X$ space.

We explore the formation of Dicke states. A system consisting of two two-level atoms located in the right Rindler wedge, has investigated to determine the conditions under which the superradiant or subradiant state can be formed. The dynamics of N two-level atoms forming symmetric state has also been analyzed and showed that the probability to excite any one atom of a collection of N atoms is related to the probability of exciting a single atom. We derive the analytical expression for the joint excitation probability which demonstrates the the interference effect. These findings provide new insights into the behavior of quantum systems in non-inertial frames and contribute to the broader understanding of relativistic quantum information theory.

An important distinction in our understanding of capacities of classical versus quantum channels is marked by the following question: is there an algorithm which can compute (or even efficiently compute) the capacity? While there is overwhelming evidence suggesting that quantum channel capacities may be uncomputable, a formal proof of any such statement is elusive. We initiate the study of the hardness of computing quantum channel capacities. We show that, for a general quantum channel, it is QMA-hard to compute its quantum capacity, and that the maximal-entanglement-assisted zero-error one-shot classical capacity is uncomputable.

Extremely large mass-ratio inspirals (XMRIs), consisting of a brown dwarf orbiting a supermassive black hole, emit long-lived and nearly monochromatic gravitational waves in the millihertz band and constitute a promising probe of strong-field gravity and black-hole properties. However, dedicated data-analysis pipelines for XMRI signals have not yet been established. In this work, we develop, for the first time, a hierarchical semi-coherent search pipeline for XMRIs tailored to space-based gravitational-wave detectors, with a particular focus on the TianQin mission. The pipeline combines a semi-coherent multi-harmonic $\mathcal{F}$-statistic with particle swarm optimization, and incorporates a novel eccentricity estimation method based on the relative power distribution among harmonics. We validate the performance of the pipeline using simulated TianQin data for a Galactic center XMRI composed of a brown dwarf and Sgr A*. For a three-month observation, the pipeline successfully recovers the signal and achieves high-precision parameter estimation, including fractional uncertainties of $<10^{-6}$ in the orbital frequency, $\lesssim10^{-3}$ in the eccentricity, $\lesssim2\times10^{-3}$ in the black-hole mass, and $\lesssim10^{-3}$ in the black-hole spin. Our framework establishes a practical foundation for future XMRI searches with space-based detectors and highlights the potential of XMRIs as precision probes of stellar dynamics and strong-field gravity in the vicinity of supermassive black holes.

Learning high-dimensional quantum systems is a fundamental challenge that notoriously suffers from the curse of dimensionality. We formulate the task of predicting quantum evolution in the linear response regime as a specific instance of learning a Complex-Valued Linear Dynamical System (CLDS) with sector-bounded eigenvalues -- a setting that also encompasses modern Structured State Space Models (SSMs). While traditional system identification attempts to reconstruct full system matrices (incurring exponential cost in the Hilbert dimension), we propose Quantum Spectral Filtering, a method that shifts the goal to improper dynamic learning. Leveraging the optimal concentration properties of the Slepian basis, we prove that the learnability of such systems is governed strictly by an effective quantum dimension $k^*$, determined by the spectral bandwidth and memory horizon. This result establishes that complex-valued LDSs can be learned with sample and computational complexity independent of the ambient state dimension, provided their spectrum is bounded.

The final state of a binary black hole merger is predicted with high precision by numerical relativity, but could there be a simple thermodynamic principle within general relativity that governs the selection of the remnant? Using post-Newtonian relations between the mass M (including the binding energy) and angular momentum J of quasi-circular, nonspinning binaries, we uncover a puzzling result: When the binary's instantaneous M and J are mapped to those of a hypothetical Kerr black hole, the corresponding entropy exhibits a maximum during the evolution. This maximum occurs at values of M and J strikingly close to those of the final remnant predicted by numerical relativity. Consistent behavior is observed when using the relation between M and J obtained from numerical relativity evolution. Although this procedure is somewhat ad hoc, the agreement between the masses and spins of the final state obtained from numerical relativity and the results of this maximum entropy procedure is remarkable, with agreement to within a few percent when using either post-Newtonian or numerical relativity results for M and J. These findings allow us to propose an entropy maximization conjecture for binary black hole mergers, hinting that thermodynamic principles may govern the selection of the final black hole state.

Verifying whether two quantum circuits are equivalent is a central challenge in the compilation and optimization of quantum programs. We introduce \textsc{Manjushri}, a new automated framework for scalable quantum-circuit equivalence checking. \textsc{Manjushri} uses local projections as discriminative circuit fingerprints, implemented with weighted binary decision diagrams (WBDDs), yielding a compact and efficient symbolic representation of quantum behavior. We present an extensive experimental evaluation that, for random 1D Clifford+$T$ circuits, explores the trade-off between \textsc{Manjushri} and \textsc{ECMC}, a tool for equivalence checking based on a much different approach. \textsc{Manjushri} is much faster up to depth 30 (with the crossover point varying from 39--49, depending on the number of qubits and whether the input circuits are equivalent or inequivalent): when inputs are equivalent, \textsc{Manjushri} is about 10$\times$ faster (or more); when inputs are inequivalent, \textsc{Manjushri} is about 8$\times$ faster (or more). For both kinds of equivalence-checking outcomes, \textsc{ECMC}'s success rate out to depth 50 is impressive on 32- and 64-qubit circuits: on such circuits, \textsc{ECMC} is almost uniformly successful. However, \textsc{ECMC} struggled on 128-qubit circuits for some depths. \textsc{Manjushri} is almost uniformly successful out to about depth 38, before tailing off to about 75\% at depth 50 (falling to 0\% at depth 48 for 128-qubit circuits that are equivalent). These results establish that \textsc{Manjushri} is a practical and scalable solution for large-scale quantum-circuit verification, and would be the preferred choice unless clients need to check equivalence of circuits of depth $>$38.

Product constructions constitute a powerful method for generating quantum CSS codes, yielding celebrated examples such as toric codes and asymptotically good low-density parity check (LDPC) codes. Since a CSS code is fully described by a chain complex, existing product formalisms are predominantly homological, defined via the tensor product of the underlying chain complexes of input codes, thereby establishing a natural connection between quantum codes and topology. In this Letter, we introduce the \textit{quantum bootstrap product} (QBP), an approach that extends beyond this standard homological paradigm. Specifically, a QBP code is determined by solving a consistency condition termed the ``bootstrap equation''. We find that the QBP paradigm unifies a wide range of important codes, including general hypergraph product (HGP) codes of arbitrary dimensions and fracton codes typically represented by the X-cube code. Crucially, the solutions to the bootstrap equation yield chain complexes where the chain groups and associated boundary maps consist of multiple components. We term such structures \textit{fork complexes}. This structure elucidates the underlying topological structures of fracton codes, akin to foliated fracton order theories. Beyond conceptual insights, we demonstrate that the QBP paradigm can generate self-correcting quantum codes from input codes with constant energy barriers and surpass the code-rate upper bounds inherent to HGP codes. Our work thus substantially extends the scope of quantum product codes and provides a versatile framework for designing fault-tolerant quantum memories.

We study grey-body factors and Hawking radiation of higher-dimensional regular black holes arising in quasi-topological gravity. These spacetimes incorporate infinite-curvature corrections that remove the central singularity while preserving an event horizon and a well-defined semiclassical description. We show that, for all considered regular black hole models, the transmission of radiation and the corresponding Hawking evaporation are significantly suppressed compared to the singular black hole solutions of General Relativity.

Modern supply chains must balance high-speed logistics with environmental impact and security constraints, prompting a surge of interest in AI-enabled Internet of Things (AIoT) solutions for global commerce. However, conventional supply chain optimization models often overlook crucial sustainability goals and cyber vulnerabilities, leaving systems susceptible to both ecological harm and malicious attacks. To tackle these challenges simultaneously, this work integrates a quantum-inspired reinforcement learning framework that unifies carbon footprint reduction, inventory management, and cryptographic-like security measures. We design a quantum-inspired reinforcement learning framework that couples a controllable spin-chain analogy with real-time AIoT signals and optimizes a multi-objective reward unifying fidelity, security, and carbon costs. The approach learns robust policies with stabilized training via value-based and ensemble updates, supported by window-normalized reward components to ensure commensurate scaling. In simulation, the method exhibits smooth convergence, strong late-episode performance, and graceful degradation under representative noise channels, outperforming standard learned and model-based references, highlighting its robust handling of real-time sustainability and risk demands. These findings reinforce the potential for quantum-inspired AIoT frameworks to drive secure, eco-conscious supply chain operations at scale, laying the groundwork for globally connected infrastructures that responsibly meet both consumer and environmental needs.

A finite sequence of numbers is perfect if it has zero periodic autocorrelation after a nontrivial cyclic shift. In this work, we study quaternionic perfect sequences having a one-to-one correspondence with the binary sequences arising in Williamson's construction of quaternion-type Hadamard matrices. Using this correspondence, we devise an enumeration algorithm that is significantly faster than previously used algorithms and does not require the sequences to be symmetric. We implement our algorithm and use it to enumerate all circulant and possibly non-symmetric Williamson-type matrices of orders up to 21; previously, the largest order exhaustively enumerated was 13. We prove that when the blocks of a quaternion-type Hadamard matrix are circulant, the blocks are necessarily pairwise amicable. This dramatically improves the filtering power of our algorithm: in order 20, the number of block pairs needing consideration is reduced by a factor of over 25,000. We use our results to construct quaternionic Hadamard matrices of interest in quantum communication and prove they are not equivalent to those constructed by other means. We also study the properties of quaternionic Hadamard matrices analytically, and demonstrate the feasibility of characterizing quaternionic Hadamard matrices with a fixed pattern of entries. These results indicate a richer set of properties and suggest an abundance of quaternionic Hadamard matrices for sufficiently large orders.

We study the photon-graviton pair production induced by the propagation of a classical electromagnetic (EM) wave in a Minkowskian spacetime. In our model, the gravitational field is described in terms of the quantized graviton field, whereas the electromagnetic field is split into a classical drive (a linearly or circularly polarized electromagnetic wave) and a quantum fluctuation field. We analyze the time evolution of the quantum state showing that, among other outcomes, the propagation of the EM wave can generate Bell states in the photon-graviton polarization basis. We finally discuss the possibility to observe entangled photons in artificial and natural scenarios.

In this contribution, we investigate the scattering cross sections of black holes immersed in perfect fluid dark matter (PFDM). We present both the classical and semi-classical scattering cross sections for different values of the parameter that characterizes the PFDM contribution. Our results show that the presence of dark matter increases the classical scattering cross section and modifies the width of the interference fringes in the semi-classical regime. In addition, the scattering cross section is also computed using the partial wave method for the black holes considered, exhibiting similar qualitative behavior. These findings suggest that the effects of dark matter surrounding black holes may play an important role in black holes phenomenology, particularly in certain regions near the black hole.

We address the problem of measuring nonclassicality in continuous-variable bosonic systems without having access to a known reference signal. To this end, we construct broader classes of criteria for nonclassicality which allow us to investigate quantum phenomena regardless of the quantumness of selected subsystems. Such witnesses are based on the notion of partial normal ordering. This approach is applied to balanced homodyne detection using arbitrary, potentially nonclassical local oscillator states, yet only revealing the probed signal's quantumness. Our framework is compared to standard techniques, and the robustness and advanced sensitivity of our approach is shown. Therefore, a widely applicable framework, well-suited for applications in quantum metrology and quantum information, is derived to assess the quantum features of a photonic system when a well-defined coherent laser as a reference state is not available in the physical domain under study.

Characterizing errors in quantum circuits is essential for device calibration, yet detecting rare error events requires a large number of samples. This challenge is particularly severe in calibrating fault-tolerant, error-corrected circuits, where logical error probabilities are suppressed to higher order relative to physical noise and are therefore difficult to calibrate through direct logical measurements. Recently, Wagner et al. [PRL 130, 200601 (2023)] showed that, for phenomenological Pauli noise models, the logical channel can instead be inferred from syndrome measurement data generated during error correction. Here, we extend this framework to realistic circuit-level noise models. From a unified code-theoretic perspective and spacetime code formalism, we derive necessary and sufficient conditions for learning the logical channel from syndrome data alone and explicitly characterize the learnable degrees of freedom of circuit-level Pauli faults. Using Fourier analysis and compressed sensing, we develop efficient estimators with provable guarantees on sample complexity and computational cost. We further present an end-to-end protocol and demonstrate its performance on several syndrome-extraction circuits, achieving orders-of-magnitude sample-complexity savings over direct logical benchmarking. Our results establish syndrome-based learning as a practical approach to characterizing the logical channel in fault-tolerant quantum devices.

We propose a protocol to measure impulses beyond the standard quantum limit. The protocol reduces noise in all three spatial dimensions and consists of squeezing a mechanical system's state via a series of jumps in the frequency of the harmonic potential. We quantify how decoherence in a realistic system of an optically levitated, dielectric nanoparticle limits the ultimate sensitivity. We predict that $\sim$10 dB of squeezing is achievable with current technology, enabling quantum-enhanced detection of weak impulses.

Hairy black hole solutions are constructed within quantum-inspired nonlocal quadratic gravity. Nonlocal effects induce Yukawa screening, which shifts the event horizon inward and modifies both the Bekenstein--Hawking entropy and the Hawking temperature, while also renormalizing the chemical potential. Nonlocal corrections also reduce the magnitude of the negative specific heat, making small hairy black holes more stable, with Helmholtz and Gibbs free energies consistent with the absence of first-order phase transitions. The nonlocal spin-2 propagator contains, in addition to the massless graviton, a massive pole with positive residue and positive norm. Consequently, hairy black holes in nonlocal quadratic gravity are free of ghost instabilities at the quadratic and classical levels, in the effective field theory.

In the paper our aim was to study the properties of a new version of coherent states whose argument is a linear combination of two special singular square 2 x 2 matrix, having a single nonzero element, equal to 1, and two labeling complex variables as developing coefficients. We have shown that this new version of coherent states satisfies all the conditions imposed on coherent states, both of pure, as well as the mixed (thermal) states characterized by the density operator. As applications, we examined the connection between these coherent states and the notions of qubits and von Neuman entropy.

In this work, we propose a deep learning-based approach for quantum entanglement and discord classification using convolutional autoencoders. We train models to distinguish entangled from separable bipartite states for $d \times d$ systems with local dimension $d$ ranging from two to seven, which enables identification of bound and free entanglement. Through extensive numerical simulations across various quantum state families, we demonstrate that our model achieves high classification accuracy. Furthermore, we leverage the learned representations to generate samples of bound entangled states, the rarest form of entanglement and notoriously difficult to construct analytically. We separately train the same convolutional autoencoders architecture for detecting the presence of quantum discord and show that the model also exhibits high accuracy while requiring significantly less training time.

We develop a systematic analysis of the metaplectic semigroup $\mathrm{Mp}_+(d,\mathbb{C})$ associated with positive complex symplectic matrices, a notion introduced almost simultaneously and independently by Hörmander, Brunet, Kramer, and Howe, thereby extending the classical metaplectic theory beyond the unitary setting. While the existing literature has largely focused on propagators of quadratic evolution equations, for which results are typically obtained via Mehler formulas, our approach is operator-theoretic and symplectic in spirit and adapts techniques from the standard metaplectic group $\mathrm{Mp}(d,\mathbb{R})$ to a substantially broader framework that is not driven by differential problems or particular propagators. This point of view provides deeper insight into the structure of the metaplectic semigroup, and allows us to investigate its generators, polar decomposition, and intertwining relations with complex conjugation and with the Wigner distribution. We then exploit these structural results to characterize, from a metaplectic perspective, classes of time-frequency representations satisfying prescribed structural properties. Finally, we discuss further implications for parabolic equations with complex quadratic Hamiltonians, we study the boundedness of their propagators on modulation spaces, we obtain estimates in time of their operator norms. Finally, we apply our theory to the study of propagation of Wigner singularities.

We investigate whether late-time modifications of gravity in the teleparallel framework can impact the current tension in the Hubble constant $H_0$, focusing on $f(T)$ cosmology as a minimal and well-controlled extension of General Relativity. We consider three representative $f(T)$ parametrisations that recover the teleparallel equivalent of General Relativity at early times and deviate from it only at late epochs. The models are confronted with unanchored Pantheon+ Type~Ia supernovae, DESI DR2 baryon acoustic oscillations, compressed Planck cosmic microwave background distance priors, and redshift-space distortion data, allowing us to jointly probe the background expansion and the growth of cosmic structures. Two of the three models partially shift the inferred value of $H_0$ towards local measurements, while the third worsens the discrepancy. This behaviour is directly linked to the effective torsional dynamics, with phantom-like regimes favouring higher $H_0$ and quintessence-like regimes producing the opposite effect. A global statistical comparison shows that the minimal $f(T)$ extensions considered here are not favoured over $Λ$CDM by the combined data. Nevertheless, our results demonstrate that late-time torsional modifications can non-trivially redistribute current cosmological tensions among the background and growth sectors.

Inflationary perturbations are quantum in origin. Yet, when computing cosmological observables, they are often treated as classical stochastic fields. Do they nevertheless retain quantum birthmarks? A hallmark of genuinely quantum behaviour is quantum interferences, arising from phase coherence between distinct branches of the wavefunction. Such interference is diagnosed by the non-positivity of the Wigner function, and according to Hudson's theorem, the only pure states with positive Wigner functions are Gaussian states. Consequently, any departure from Gaussianity necessarily implies a non-positive Wigner function, precluding a description in terms of a classical distribution. This motivates us to compute the Wigner function of curvature perturbations, accounting for primordial non-Gaussianities, using the EFT of inflation. We find that the Wigner function develops pronounced interference fringes on super-Hubble scales, and in particular, its negativity grows as $a^2$ in ultra-slow-roll backgrounds. These results demonstrate that quantum effects can remain significant at late times, and that squeezing alone does not ensure classicality, contrary to standard lore. This suggests that the prospects for detecting genuinely quantum signatures of the universe's origins in cosmological observables may be less bleak than previously thought.

While no gravitational-wave detection of subsolar mass black holes has been confirmed to date, a number of candidate detections invite us to speculate on the origin of such black holes should a detection be confirmed. It is generally assumed that the observation of a black hole with subsolar mass $M_{\rm obs}$ would provide strong evidence for primordial black holes (PBHs). The mass $M_{\rm PBH}$ of the PBH, however, does not necessarily have to be equal to $M_{\rm obs}$, as it would in what we term a ``direct PBH scenario". Instead, a black hole of mass $M_{\rm obs}$ may form in a capture of a much smaller primordial black hole, $M_{\rm PBH} \ll M_{\rm obs}$, by a dwarf star of mass $M_* \simeq M_{\rm obs}$, followed by the total consumption of the star by the PBH. We provide some rough estimates and demonstrate that such an ``indirect PBH scenario" may also lead to significant populations of black holes with mass $M_{\rm obs}$, especially in dwarf galaxies, and may be able to explain rare subsolar mass events.

We show that Scalar-Tensor-Vector Gravity (STVG-MOG) is observationally equivalent to the standard model $Λ$CDM cosmological model for all probes that depend on isotropic and linear gravitational dynamics, including galaxy rotation curves, cluster lensing, the linear matter power spectrum P(k), $σ_8$, baryon acoustic oscillations, and the cosmic microwave background (CMB). This degeneracy arises from the scale-dependent effective gravitational coupling $G_{\mathrm{eff}}$, which ensures identical background evolution, transfer functions, and linear growth. Consequently, all early-universe, low and intermediate scale cosmological observables are equally well described by STVG-MOG without invoking non-baryonic dark matter. We argue that the equivalence implies that isotropic cosmological data alone cannot establish the physical existence of dark matter. The degeneracy is broken only by observables sensitive to large-scale, anisotropic gravitational response. In particular, recent measurements of enhanced radio-galaxy and quasar number-count dipoles at gigaparsec scales probe a regime where $G_{\mathrm{eff}}$ departs from its $Λ$CDM limit, allowing STVG-MOG to generate anisotropic bulk flows, while preserving consistency with all isotropic constraints. These observations provide a concrete pathway for empirically distinguishing modified gravity from particle dark matter.

We perform fully nonlinear, spherically symmetric numerical simulations of superhorizon false-vacuum-domain (FVD) collapse in a coupled gravity-scalar-fluid system to study primordial black hole (PBH) formation during delayed first-order phase transitions (FOPTs). Using adaptive mesh refinement to resolve the bubble wall, we identify three dynamical outcomes: type B (supercritical) PBHs with an interior baby universe and a bifurcating trapping horizon, type A (subcritical) PBHs with an apparent horizon formed by direct wall collapse, and dispersal with no PBH formation. To separate these three cases, we evaluate two commonly used PBH-formation criteria: the time scale ratio $t_\mathrm{H}/t_\mathrm{V}$ (horizon crossing time versus vacuum-energy domination time) and the local density contrast $δ(t_\mathrm{H})$ at horizon crossing. For the parameter space explored, we find that $t_\mathrm{H}/t_\mathrm{V}$ is a more robust predictor of outcome: type B PBHs form when $t_\mathrm{H}/t_\mathrm{V} \gtrsim 1$ (critical range $\sim 1.1 - 1.6$ in our survey), type A PBHs arise when $t_\mathrm{H}/t_\mathrm{V}$ is below this threshold but remains above a lower bound (typical range $\sim 0.35 - 0.7$), and no-PBH dispersal occurs when $t_\mathrm{H}/t_\mathrm{V}$ falls below this lower bound. When a clear thin-wall FVD boundary exists, $δ(t_\mathrm{H})$ can correspondingly distinguish different outcomes (roughly $δ_c(t_\mathrm{H}) \sim 1 - 1.7$ for type B and $δ_c(t_\mathrm{H}) \sim 0.35 - 0.5$ for type A), but is highly sensitive to wall structure and model details and thus less universal. These results offer new insights into the dynamics of FVD collapse, quantify practical PBH-formation thresholds, and pave the way for precise predictions of PBH abundance from delayed FOPTs.

Differential geometries derived from tensor decompositions have been extensively studied and provided the foundations for a variety of efficient numerical methods. Despite the practical success of the tensor ring (TR) decomposition, its intrinsic geometry remains less understood, primarily due to the underlying ring structure and the resulting nontrivial gauge invariance. We establish the quotient geometry of TR decomposition by imposing full-rank conditions on all unfolding matrices of the core tensors and capturing the gauge invariance. Additionally, the results can be extended to the uniform TR decomposition, where all core tensors are identical. Numerical experiments validate the developed geometries via tensor ring completion tasks.

We consider the problem of covert communication over the entanglement-assisted (EA) bosonic multiple access channel (MAC). We derive a closed-form achievable rate region for the general EA bosonic MAC using high-order phase-shift keying (PSK) modulation. Specifically, we demonstrate that in the low-photon regime the capacity region collapses into a rectangle, asymptotically matching the point-to-point capacity as multi-user interference vanishes. We also characterize an achievable covert throughput region, showing that entanglement assistance enables an aggregate throughput scaling of \(O(\sqrt{n} \log n)\) covert bits with the block length $n$ for both senders, surpassing the square-root law as in the point-to-point case. Our analysis reveals that the joint covertness constraint imposes a linear trade-off between the senders throughput.

The Bravyi-König (BK) theorem is an important no-go theorem for the dynamics of topological stabiliser quantum error correcting codes. It states that any logical operation on a $D$-dimensional topological stabiliser code that can be implemented by a short-depth circuit acts on the codespace as an element of the $D$-th level of the Clifford hierarchy. In recent years, a new type of quantum error correcting codes based on Pauli stabilisers, dubbed Floquet codes, has been introduced. In Floquet codes, syndrome measurements are arranged such that they dynamically generate a codespace at each time step. Here, we show that the BK theorem holds for a definition of Floquet codes based on locally conjugate stabiliser groups. Moreover, we introduce and define a class of generalised unitaries in Floquet codes that need not preserve the codespace at each time step, but that combined with the measurements constitute a valid logical operation. We derive a canonical form of these generalised unitaries and show that the BK theorem holds for them too.

Recent studies have shown that dark matter halos can support regular black holes or compact stars by assuming an anisotropic energy-momentum tensor. In this paper, we extend the analysis to the dark matter halo as an isotropic perfect fluid. By employing galactic dark matter profiles-specifically the Einasto and Dehnen models-as the mass-energy density source, we numerically solve the Einstein field equations and find a class of non-singular, horizonless compact star solutions. Moreover, these configurations remain stable against axial perturbations while satisfying the dominant energy condition.

Protecting quantum information through quantum error correction (QEC) is a cornerstone of future fault-tolerant quantum computation. However, current QEC-protected logical qubits have only achieved coherence times about twice those of their best physical constituents. Here, we show that the primary barrier to higher QEC gains is ancilla-induced operational errors rather than intrinsic cavity coherence. To overcome this bottleneck, we introduce error-detectable universal control of bosonic modes, wherein ancilla relaxation events are detected and the corresponding trajectories discarded, thereby suppressing operational errors on logical qubits. For binomial codes, we demonstrate universal gates with fidelities exceeding $99.6\%$ and QEC gains of $8.33\times$ beyond break-even. Our results establish that gains beyond $10\times$ are achievable with state-of-the-art devices, establishing a path toward fault-tolerant bosonic quantum computing.

Gravitational lensing of gravitational wave (GW) will become the next frontier in studying cosmology and gravity. While time-delay cosmography using quadruply lensed GW events associated with optical images of the lens systems can provide precise measurement of the Hubble constant ($H_0$), they are considered to be much rarer than doubly lensed events. In this work, we analyze time-delay cosmography with doubly lensed GW events for the first time. We generate mock doubly lensed GW events with designed sensitivity of the LIGO-Virgo-KAGRA (LVK) O5 network, with LIGO post-O5 upgrade, and with Einstein Telescope (ET) + Cosmic Explorer (CE) respectively, and select the events that can be associated with future galaxy surveys. Over 1000 realizations, we find an average of 0.2(2.4) qualified events with the LVK O5(post-O5) network. Whereas with the ET+CE network, we find an average of 73.2 qualified events over 100 realizations. Using the Singular Isothermal Sphere (SIS) lens model, we jointly estimate waveform parameters and the impact parameter with doubly lensed GW signals, and then forecast the constraints on cosmological parameters and modified GW propagation by combining time-delay cosmography and the standard siren approach. The average posterior gives a constraint on $H_0$ with a relative uncertainty of $14\%$, $10\%$ and $0.42\%$ in the $Λ$CDM model for the LVK O5, LVK post-O5, and ET+CE network, respectively. While the LVK network gives uninformative constraints on the $(w_0,w_a)$ dynamical dark energy model, the ET+CE network yields a moderate constraint of $w_0=-1.02^{+0.31}_{-0.22}$ and $w_a=0.48^{+0.99}_{-1.54}$. In addition, our method can provide precise constraints on modified GW propagation effects jointly with $H_0$.

We introduce a principle, implicitly contained in Schroedinger's paper (Schr35), which allows a proof of the non-existence of the EPR-locality paradox in the Copenhagen interpretation of quantum mechanics. The paradox is shown to be well-posed already in the simplest example of an entangled state of two spins one-half, independently of the (well-taken) objections by Araki and Yanase that the measurement of spin is not a local measurement. We assume that any measurement results in the collapse of the wave-packet.

Determining when the multiparameter quantum Cramér--Rao bound (QCRB) is saturable with experimentally relevant single-copy measurements is a central open problem in quantum metrology. Here we establish an equivalence between QCRB saturation and the simultaneous hollowization of a set of traceless operators associated with the estimation model, i.e., the existence of complete (generally nonorthogonal) bases in which all corresponding diagonal matrix elements vanish. This formulation yields a geometric characterization: optimal rank-one measurement vectors are confined to a subspace orthogonal to a state-determined Hermitian span. This provides a direct criterion to construct optimal Positive Operator-Valued Measures(POVMs). We then identify conditions under which the partial commutativity condition proposed in [Phys. Rev. A 100, 032104(2019)] becomes necessary and sufficient for the saturation of the QCRB, demonstrate that this condition is not always sufficient, and prove the counter-intuitive uselessness of informationally-complete POVMs.

We construct an exact magnetized generalization of the Zipoy-Voorhees spacetime by applying the magnetic Harrison transformation to a static seed with quadrupolar deformation parameter $k$. The resulting Melvin-Zipoy-Voorhees metric is a solution to the Einstein-Maxwell equations that interpolates between the unmagnetized Zipoy-Voorhees geometry and the Melvin magnetic universe. We analyze the algebraic structure, finding the spacetime to be generically of Petrov type I, and investigate the equatorial dynamics of charged test particles and photons. Our analysis reveals that the external magnetic field $b$ induces a ``Lorentz shift'' in the effective angular momentum, suppressing the potential barrier and causing the Innermost Stable Circular Orbit (ISCO) to migrate inward. In contrast, the radius of the photon ring shifts slightly outward with increasing magnetization.

Hybrid quantum-classical learning models increasingly integrate neural networks with variational quantum circuits (VQCs) to exploit complementary inductive biases. However, many existing approaches rely on tightly coupled architectures or task-specific encoders, limiting conceptual clarity, generality, and transferability across learning settings. In this work, we introduce Quantum LEGO Learning, a modular and architecture-agnostic learning framework that treats classical and quantum components as reusable, composable learning blocks with well-defined roles. Within this framework, a pre-trained classical neural network serves as a frozen feature block, while a VQC acts as a trainable adaptive module that operates on structured representations rather than raw inputs. This separation enables efficient learning under constrained quantum resources and provides a principled abstraction for analyzing hybrid models. We develop a block-wise generalization theory that decomposes learning error into approximation and estimation components, explicitly characterizing how the complexity and training status of each block influence overall performance. Our analysis generalizes prior tensor-network-specific results and identifies conditions under which quantum modules provide representational advantages over comparably sized classical heads. Empirically, we validate the framework through systematic block-swap experiments across frozen feature extractors and both quantum and classical adaptive heads. Experiments on quantum dot classification demonstrate stable optimization, reduced sensitivity to qubit count, and robustness to realistic noise.

Black holes exactly incorporating quantum matter backreaction effects, namely, quantum black holes, are notoriously difficult to construct, let alone study their horizon thermodynamics. Here, we derive the thermodynamics of three-dimensional charged and rotating quantum black holes via the tree-level gravitational partition function. Specifically, we primarily focus on holographic quantum BTZ black holes, dual to $(3+1)$-dimensional accelerating black holes in anti-de Sitter space that localize on Karch-Randall end-of-the-world (ETW) branes. To derive their horizon thermodynamics, we regulate the bulk Euclidean geometry by adding a second ETW brane at asymptotic spatial infinity. We compute the on-shell action of the complexified accelerating black hole in the grand canonical ensemble and derive the quantum BTZ black hole thermodynamics, where the thermal entropy is equal to the generalized entropy. This provides a first principles derivation of the generalized entropy of three-dimensional quantum black holes. Further, we construct charged and rotating quantum black holes in three-dimensional de Sitter and Minkowski space using Randall-Sundrum ETW branes, and compute their horizon thermodynamics.

We consider the evolution of self-gravitating matter fields that may undergo phase transitions, and we connect ideas from phase transition dynamics with concepts from bouncing cosmology. Our framework introduces scattering maps prescribed on two classes of hypersurfaces: a gravitational singularity hypersurface and a fluid-discontinuity hypersurface. By analyzing the causal structures induced by the light cone and the acoustic cone, we formulate a local evolution problem for the Einstein-Euler system in the presence of such interfaces. We explain how suitable scattering relations must supplement the field equations in order to ensure uniqueness and thus yield a complete macroscopic description of the evolution. This viewpoint builds on a theory developed in collaboration with G. Veneziano for quiescent (velocity-dominated) singularities in solutions of the Einstein equations coupled to a scalar field, where the passage across the singular hypersurface is encoded by a singularity scattering map. The guiding question is to identify junction prescriptions that are compatible with the Einstein and Euler equations, in particular with the propagation of constraints. The outcome is a rigid set of universal relations, together with a family of model-dependent parameters. Under physically motivated requirements (general covariance, causality, constraint compatibility, and ultra-locality), we aim to classify admissible scattering relations arising from microscopic physics and characterizing, at the macroscopic level, the dynamics of a fluid coupled to Einstein gravity.

This paper examines the application of a Quantum Support Vector Machine (QSVM) for radarbased aerial target classification using micro-Doppler signatures. Classical features are extracted and reduced via Principal Component Analysis (PCA) to enable efficient quantum encoding. The reduced feature vectors are embedded into a quantum kernel-induced feature space using a fully entangled ZZFeatureMap and classified using a kernel based QSVM. Performance is first evaluated on a quantum simulator and subsequently validated on NISQ-era superconducting quantum hardware, specifically the IBM Torino (133-qubit) and IBM Fez (156-qubit) processors. Experimental results demonstrate that the QSVM achieves competitive classification performance relative to classical SVM baselines while operating on substantially reduced feature dimensionality. Hardware experiments reveal the impact of noise and decoherence and measurement shot count on quantum kernel estimation, and further show improved stability and fidelity on newer Heron r2 architecture. This study provides a systematic comparison between simulator-based and hardware-based QSVM implementations and highlights both the feasibility and current limitations of deploying quantum kernel methods for practical radar signal classification tasks.

Quantum mechanics provides extraordinarily accurate probabilistic predictions, yet the framework remains silent on what distinguishes quantum systems from definite measurement outcomes. This paper develops a measurement-theoretic framework for projective gating. The central object is the $β$-bound, an inequality that controls how much probability assignments can drift when gating and measurement fail to commute. For a density operator $ρ$, projector $F$, and effect $E$, with gate-passage probability $s = {\rm Tr}(ρF)$ and commutator norm $\varepsilon = \|[F, E]\|$, the symmetric partial-gating drift satisfies $|Δp_F(E)| \leq 2 \sqrt{(1 - s)/s} \cdot \varepsilon$. The constant 2 is sharp. We introduce two diagnostic quantities: the coherence witness $W(ρ, F) = \|F ρ(I - F)\|_1$, measuring cross-boundary coherence, and the record fidelity gap $Δ_T(ρ_F, R)$, measuring expectation-value change under symmetrisation. Three experimental vignettes demonstrate falsifiability: Hong--Ou--Mandel interferometry, atomic energy-basis dephasing, and decoherence-induced classicality. The framework is operational and interpretation-neutral, compatible with Everettian, Bohmian, QBist, and collapse approaches. It provides quantitative structure that any interpretation must accommodate, along with a template for experimental tests.

The IceCube neutrino observatory consists of an array of Digital Optical Modules (DOMs) instrumenting one cubic-kilometer of deep glacial ice at the South Pole. DeepCore, a densely-spaced sub-array of DOMs at the bottom central region of IceCube, enables the detection of atmospheric neutrinos with an energy threshold in the GeV range. The high statistics data of DeepCore provides a unique opportunity to perform neutrino oscillation studies as well as explore various sub-leading Beyond the Standard Model (BSM) physics signatures. We consider a well-motivated minimal extension of the Standard Model by an additional anomaly-free, gauged lepton-number symmetry, such as $L_e - L_μ$ or $L_e - L_τ$. These symmetries give rise to flavor-dependent long-range interaction mediated through a very light neutral gauge boson. In this contribution, we present the sensitivity of the IceCube DeepCore detector to search for this flavor-dependent long-range interaction potential with a runtime of 9.3 years.

Earth's mass and internal structure have been primarily studied through gravitational and seismic methods. Neutrinos, however, offer an independent way to explore Earth's interior via matter effects in neutrino oscillations that depend on the electron distribution inside Earth, and hence its matter density. Our study uses atmospheric neutrinos at DeepCore, a densely instrumented sub-detector of the IceCube Neutrino Observatory, to estimate Earth's mass and layer densities. We also assess how the upcoming IceCube Upgrade, with denser instrumentation, could improve these measurements.

We implement modular flavor symmetries within the Standard Model Effective Field Theory (SMEFT) framework, using the flavor group $A_4^{(q)} \times A_4^{(e)}$ with distinct moduli $τ_q$ and $τ_e$, and assigning different modular weights to right-handed quarks using simplest weight assignment. By treating the moduli as non-dynamical spurions, adopting the MFV-like assumption, and neglecting effects associated with $\mathrm{Im}\,τ$, we systematically construct a finite set of independent modular-invariant higher-dimensional operators via the Hilbert-series techniques. In the holomorphic $A_4$ scenario, where all modular forms derive from the weight-2 triplet $Y^{(2)}_{\mathbf{3}}$, we present two equivalent Hilbert-series bases. This establishes that higher-dimensional operators can be formally organized as $[Y_{\mathbf{r}}^{(k_Y)},{Y_{\mathbf{r}'}^{(k_Y')}}^{*},\mathcal{O}]_{\mathbf{1}}$ singlets. We subsequently enumerate all independent operators up to dimension 7 under this assumption and provide explicit constructions for all dimension-5 operators as well as baryon- and lepton-number conserving dimension-6 operators. Relaxing holomorphicity to the non-holomorphic case of polyharmonic Maas forms, considering that non-holomorphic modular forms are not closed under multiplication, adopting the holomorphic organizing idea would generically lead to an infinite proliferation of modular-invariant structures. To retain a finite and complete operator basis, we therefore impose the same minimal formal organizing principle, which reproduces the benchmark Weinberg operator and the corresponding dimension-$6$ operators.

The IceCube detector, using its densely instrumented center, called DeepCore, can detect multi-GeV atmospheric neutrinos. The oscillation pattern of neutrinos is altered due to interactions with ambient electrons as they pass through Earth. The changes in these patterns are influenced by the amount of matter and its specific arrangement. As neutrinos propagate, they retain information about the densities they encounter. Our study demonstrates that IceCube DeepCore can utilize the Earth's matter effects to distinguish between a homogeneous matter density profile and a layered structure density profile of Earth. In this contribution, we present that IceCube DeepCore data equivalent to 9.3 years of observation can reject the homogeneous matter density profile with a confidence level of 1.4$σ$.

The discovery of the non-zero value of $θ_{13}$ has opened an exciting opportunity to probe the Earth's matter effects in three-flavor oscillations of atmospheric neutrinos. These matter effects depend on both neutrino energy and the electron density distributions encountered during their propagation through Earth. In this contribution, we present preliminary sensitivities from the DeepCore detector, a densely instrumented sub-array of the IceCube neutrino observatory at the South Pole, demonstrating its ability to observe these matter effects in atmospheric neutrino oscillations. Using simulated data equivalent to 9.3 years of observations at IceCube DeepCore, we show the sensitivity of the DeepCore to reject the vacuum oscillation hypothesis and align with the Preliminary Reference Earth Model. Additionally, we present the expected improvement in sensitivity for rejecting the vacuum oscillations using the upcoming IceCube Upgrade, a low-energy extension of the IceCube detector.

We study regulator and cutoff artifacts in the quark-meson model at finite temperature and quark chemical potential within the functional renormalization-group approach using the local potential approximation. To this end, we discuss the concept of renormalization-group consistency in effective models, which necessitates a nontrivial parameter-fixing procedure to enable a meaningful comparison of results obtained with different regulators and cutoffs. We employ a standard range of cutoff values used in phenomenological studies and regulators that differ significantly in their analytic properties as well as in their classification according to the principle of strongest singularity. We find that regulator and cutoff dependences are small at low temperatures and quark chemical potentials. At high temperatures and low quark chemical potentials, significant cutoff artifacts arise, whereas the properties of the regulator affect the dynamics in the regime governed by a chiral phase transition of first order at low temperatures and high quark chemical potentials.

The Standard Model (SM) of particle physics provides a successful description of elementary particles and their interactions. However, it does not explain phenomena such as the hierarchy problem or the nature of dark matter. Many Beyond the Standard Model (BSM) theories provide answers to these open questions by adding new heavy resonances, which can be probed directly at colliders. This note summarizes recent resonance searches by the CMS Collaboration, focusing on final states containing top quarks and Higgs bosons.

We present a local subtraction scheme that enables the combined integration of loop momenta and the final-state parton phase space in real-virtual NNLO QCD corrections to cross sections for hadroproduction of electroweak and other colorless states. All initial- and final-state infrared singularities are subtracted at the integrand level in momentum space, yielding a locally finite integral ready for numerical integration in four dimensions. The subtraction terms are all based on the well-understood process of single-Higgs production. The core of our subtraction scheme relies on achieving local factorization in all infrared limits of real and virtual momenta. This necessitates systematic modifications of the original Feynman integrand for loop amplitudes, enabling gauge symmetry cancellations before performing integrations. Our approach provides an essential step toward NNLO cross-section calculations for hadron collider processes, where both loop and phase-space integrations are carried out numerically.

Using skein valued holomorphic curve counting techniques, we give a flow loop formula for the skein valued partition function of the Lagrangian knot complement of a fibered knot (of the $A$-model open topological strings with Lagrangian $A$-branes wrapping the complement) in the cotangent bundle of the three-sphere and in the resolved conifold. For torus knots we show that the partition function in the cotangent bundle localizes on two or three holomorphic annuli and give a corresponding generalized quiver structure for the partition function in the resolved conifold. We connect the formula to the augmentation curve, the representation variety of the knot contact homology algebra of the knot, generated by Reeb chords of its Legendrian conormal and with differential given by holomorphic disks interpolating between words of Reeb chords. The curve admits a quantization as a $q$-difference equation for the generating function of symmetrically colored HOMFLYPT-polynomials of the knot or, geometrically, for the $U(1)$-partition function of the knot conormal. For $(2,2p+1)$-torus knots we show that, after a change of variables, the partition function of the knot complement also satisfies this $q$-difference equation. This gives another geometrically defined coordinate chart for the $D$-module defined by the quantized augmentation polynomial.

We study fluctuations and correlations of conserved charges in QCD using a string-based description of the hadronic mass spectrum. Mesons and baryons are modeled as open relativistic strings with quark-antiquark and quark-diquark endpoints, respectively, leading to an exponential Hagedorn growth of states with a limiting temperature fixed by the string tension. We find that continuous Hagedorn spectra constrained by experimentally established hadrons underestimate net-baryon number fluctuations obtained in lattice QCD calculations. By extracting the Hagedorn string spectrum directly from lattice QCD through a fit to the second-order net-baryon number susceptibility, we obtain a consistent description of a broad set of fluctuations of conserved charges from LQCD with the Hagedorn temperature $T_H \simeq 323~$MeV, without introducing additional free parameters. Our results provide thermodynamic evidence in support of a string quark-diquark picture of baryons in the confined phase of QCD.

We propose a new approach to probe the initial fireball geometry in relativistic heavy-ion collisions using spin polarization. Specifically, we introduce polarization harmonics of open heavy hadrons as a novel observable sensitive to geometric anisotropies. Heavy quarks are produced in early hard scatterings and can acquire spin polarization from the strong, transient electromagnetic fields present at early times. As they propagate through the anisotropic quark-gluon plasma, medium-induced interactions lead to path-length dependent depolarization, imprinting an azimuthally anisotropic polarization pattern. Within the framework of rotational Brownian motion, we show that the resulting polarization harmonics are directly related to the initial spatial eccentricities, thereby establishing heavy-flavor polarization anisotropies as a sensitive and complementary probe of the early-time collision geometry. We present quantitative estimates of the second polarization harmonic associated with the recently observed $D^{*+}$ spin alignment reported by the ALICE Collaboration.

In this note, we will generalize the concept of the holography for the color superconductivity (CSC) phase becomes to d-dimension AdS instead of 6d. The dual field theory live in (d-1)-dimension $SU(N_c)$ and have no confinement phase contradiction with the QCD color superconductivity. And we will try to use holographic model with Einstein-Maxwell gravity in d dimension AdS and we study this phase with $N_c\geq 2$. And after, we will discuss the equation of state of the color superconductivity in $d=4$ case without confinement via holography

We review the chromoelectric and chromomagnetic mechanisms that tentatively lead to stable or metastable multiquark configurations. An alternative interpretation of the dynamics is the quark interchange between hadrons, as illustrated in the case of the fully-charm systems.

We present a model with dipole Pomeron bassed on Regge theory framework for vetor meson photoproduction by arbitarily polaried photons. The accurate helicity amplitude is constructed with free trajectory parameters. This model consistently describes the total and momentum-transfer differential cross sections and he spin-density matrix elements in photoproduction of $ρ^{0}$, by fitting it to the data of $3$ categories simultaneously to determine its $20$ parameters. The agreements with experimental data of our model improve those of the previous models remarkably and predictions for circular SDMEs are made. The model provides key description of the process for our innovative polarimetry of cosmic photons and essential insight to explore the non-perturbative regime and the spin dynamics of strong interaction.

In the article, we study prompt $J/ψ$ production at small transverse momentum within the Transverse Momentum Dependent (TMD) factorization. The Soft Gluon Resummation (SGR) approach is used for modelling of TMD PDFs. The hadronization process of heavy charm quarks is described with the Improved Color Evaporation model (ICEM). In order to obtain spectra of $J/ψ$ at the arbitrary transverse momenta, fixed-order calculations within the collinear parton model (CPM) are also applied and further matched with the TMD calculations using the Inverse-Error Weighting (InEW) scheme. We present results of calculations for a wide range of collision energy with the fitting of the relevant nonperturbative parameter of the ICEM. Predictions for the SPD NICA experiment are also provided alongside comparison with the result of calculation using nonrelativistic QCD (NRQCD) framework.

We present a unified axion model framework that simultaneously addresses the origin of neutrino masses, leptonic flavor structure, the strong CP problem, and dark matter. The model is based on a global $U(1)_{\rm PQ}$ symmetry combined with a modular $S_3$ symmetry and is realized within a novel class of KSVZ-type axion model. Exotic colored fermions and scalars mediate radiative neutrino mass generation at the one loop-level. The PQ charge assignment forbids tree-level neutrino masses and leaves a residual $Z_3$ symmetry that ensures the Dirac nature of neutrinos. In the minimal realization, the neutrino mass matrix is of rank two, predicting one massless neutrino. Consequently, the sum of neutrino masses is constrained for both the normal and inverted hierarchies. We analyze the implications for charged lepton flavor violation and the lepton $g-2$. The axion emerging from this framework dynamically resolves the strong CP problem and accounts for the observed dark matter abundance. Notably, the predicted axion-photon coupling is within reach of upcoming experiments and consistent with existing astrophysical and cosmological bounds.

Matrix elements of the energy-momentum tensor for one-particle states of the $Z$-boson are parameterized in terms of gravitational form factors. One-loop order electroweak corrections to these quantities are calculated. Renormalization and physical interpretation of the obtained results are discussed.

The femtoscopic correlation function has been established in recent years as a high-precision tool for investigating hadron-hadron interactions and exotic states, providing stringent constraints on the dynamics of low-energy strong interactions. However, current research has been predominantly focused on the $s$-wave interaction between hadrons, while studies of higher partial waves remain scarce. We present a general analytical expression for the femtoscopic correlation function in an arbitrary partial wave using the Lippmann-Schwinger equation. This formalism is applied to constrain the $d$-wave $K^-p$ scattering through a combined study of the $K^-p$ correlation function and the $D_{03}$ scattering amplitude of $\bar K N \to \bar K N$ and $\bar K N \to πΣ$ processes, from which the properties of $Λ(1520)$ are extracted and found to be in good agreement with the experimental results. These findings demonstrate the feasibility of determining dynamics between hadrons through femtoscopic correlation functions and scattering amplitudes with higher partial waves.

We study the two-body hadronic weak decays of the $Ω^-$ hyperon with strangeness $S=-3$, including three singly Cabibbo-suppressed decay modes: $Ξ^0 π^-$, $Ξ^-π^0$ and $ΛK^-$. The decay amplitudes at the quark level, arising from $s\to ud \bar{u}$ transitions (direct pion emission and color-suppressed processes) and $su\to ud$ transitions (pole terms), are calculated in the framework of the non-relativistic constituent quark model.The theoretical results show that the $Ξ^0 π^-$ channel is dominated by the color-allowed direct pion emission process, while the $ΛK^-$ channel is well described by one type of pole contribution mediated through intermediate $Ξ$ resonances ($1^2S_{1/2^+}$ and $1^2P_{1/2^-}$ states). However, the contribution from tree-level mechanisms alone to the branching ratio of $Ω^- \to Ξ^- π^0$ is small due to its color-suppressed nature. The discrepancy is resolved by including final state interactions through rescattering processes via intermediate states $Ξ^0π^-$ and $ΛK^-$. This work demonstrates that a unified description of $Ω^-$ hadronic weak decays necessitates the interplay of quark-level weak vertices, baryon pole structures, and long-distance final state rescattering dynamics. With these mechanisms, the obtained branching ratios are in agreement with the high-precision experimental data from the BESIII. Furthermore, these above decays are found to be dominated by the parity-conserving $P$-wave transitions, thus the asymmetry parameters are almost zero.

We discuss the application of a relativistic potential model to the description of the spectrum and radiative transitions in mesons containing at least one heavy quark (b or c). Although the model has a small number of parameters, it is possible to achieve qualitative agreement with all available experimental data, including those that could not be explained by all previous methods. This demonstrates the importance of taking relativistic effects into account. A remarkable property of the relativistic potential model is that the predictions for meson masses and partial widths of radiative transitions remain finite in the limit of zero light quark mass.

We study the entanglement produced in transverse momentum by two-particle scattering at high energy. Employing the S-matrix framework for the derivation of reduced density matrices, we formulate the entanglement entropy for an inelastic scattering as well as an elastic one. We display the formulas of the entanglement entropy in terms of two-body cross sections. We also derive the entanglement density as a function of the transverse momentum. As an application, we then focus on both forward elastic ($pn \to pn$) and inelastic ($pn \to np$) channels scattering allowing for a fruitful comparison of the two reactions with the same proton-neutron content. We evaluate the elastic and inelastic entanglement entropy by using known parameterizations of experimental data for neutron-proton reactions. Comparing those entanglement entropies, we observe that the inelastic scattering produces more overall entanglement than the elastic one in the $pn$ sector.

The ATLAS Collaboration at CERN's Large Hadron Collider is at the forefront of particle physics research and is equally committed to bridging the gap between cutting-edge science and the wider public. Since 2010, the ATLAS Virtual Visits programme has provided live, interactive tours of the ATLAS detector and control room to global audiences in their language, without the need to travel. The programme has grown significantly since its inception, as demonstrated by quantitative data collected since January 2019 and case studies of large-scale implementations in Brazil and Greece. The impact on individual participants is also discussed.

The Standard Model of particle physics explains many natural phenomena yet remains incomplete. Vectorlike quarks and leptoquarks lie at the heart of many extensions to the Standard Model seeking to address the hierarchy problem, or the flavour sector anomalies. These proceedings present the new results from searches with the ATLAS detector at the LHC.

This study examines the spectral functions of neutral ($π_0$) and charged ($π_{\pm}$) pions under a uniform magnetic field using the SU(2) Nambu-Jona-Lasinio (NJL) model with the Ritus method. The analysis highlights the complex interplay of magnetic field effects, thermal influences, and chiral symmetry on meson properties in extreme QCD environments. For $π_0$, whose properties are governed by the behavior of its constituent quarks, magnetic field-induced Landau levels lead to a multi-peak structure in its spectral function, reflecting stable and resonance solutions that evolve with temperature, showing shifts and critical enhancements near chiral restoration. For $π_{\pm}$, cross terms that come from the asymmetry between the constituent quarks introduce Landau cuts alongside Unitary cuts, indicating damping effects, with decay widths narrowing at higher temperatures, suggesting increased stability.

In this work, we investigate the possibility that supersymmetric structures may leave observable imprints in the stochastic gravitational-wave (GW) background generated during the reheating era. To this end, we construct a phenomenological interaction vertex describing the coupling between a single inflaton and the D-term sectors of a pair of chiral and anti-chiral superfields. In contrast to the conventional Yukawa coupling between the inflaton and structureless matter fields, we find that the supersymmetry-preserving chiral multiplet structure leads to a substantial enhancement, by at least one order of magnitude, in the amplitude of the resulting GWs spectrum. Our results therefore suggest that the interplay between reheating-era stochastic GWs and supersymmetric phenomenology merits further exploration and development.

We study the classical double copy for static black string solutions in an Anti--de Sitter (AdS) background. By casting the black string metric into Kerr--Schild form over a cylindrical AdS geometry, we construct the corresponding single and zeroth copies. The single copy describes a gauge field satisfying Maxwell-like equations and sourced by an effective line of color charge, while the zeroth copy is given by a scalar field conformally coupled to the AdS background. We also extend the analysis to charged black strings, identifying the associated modifications in the gauge sector. These results show that the classical double copy consistently applies to extended gravitational objects in curved spacetimes.

Neutrino oscillations continue to provide one of the most promising avenues for uncovering physics beyond the Standard Model. In particular, beyond-standard-model neutrino matter interactions may perturb neutrino oscillations in matter, leading to an observable signal in long baseline oscillation experiments. Moreover, such interactions can be a possible explanation of the rising tension between T2K and NOvA's $δ_{\text{CP}}$ measurements. We examine IceCube DeepCore's sensitivity to these Non-Standard Interactions (NSI) by employing a model-independent NSI parameterization, and examine IceCube DeepCore's ability to comment on NSI being the cause of the T2K-NOvA $δ_{\text{CP}}$ tension.

The Spin Physics Detector (SPD) is a universal detector in the one of two interaction points of the NICA collider under construction at JINR, Dubna. SPD plans to study the spin structure of the proton and deuteron and other spin-related phenomena using a unique possibility to operate with polarized proton and deuteron beams at a collision energy up to 27 GeV and a luminosity up to $10^{32}$ cm$^{-2}$ s$^{-1}$. As the main goal, the experiment aims to provide access to the gluon TMD PDFs in the proton and deuteron, as well as the gluon transversity distribution and tensor PDFs in the deuteron, via the measurements of specific single and double spin asymmetries using different complementary probes such as charmonia, open charm, and prompt photon production processes. Other polarized and unpolarized physics is possible, especially at the first stage of NICA operation with reduced luminosity and collision energy of the proton and ion beams. Construction of the first stage of the SPD facility is included in the JINR seven-year development plan for 2024-2030. The physics program of the SPD project and the design of the SPD setup are presented.

Lattice QCD has historically produced energy results interpretable as either estimates relying on implicit assumptions about asymptotic behavior or one-sided upper bounds. New Lanczos methods providing two-sided bounds with less-restrictive assumptions are introduced and quantified in a high-statistics calculation with unphysical quark masses. Two-sided bounds without spectral assumptions provide sub-percent constraints on the nucleon mass. Other bounds, which assume all states in a given energy window are resolved, provide meaningful two-sided constraints on nucleon-nucleon scattering phase shifts.

Excited-state effects lead to hard-to-quantify systematic uncertainties in lattice quantum chromodynamics (LQCD) spectroscopy calculations when computationally accessible imaginary times are smaller than inverse excitation gaps, as often arises for multi-hadron systems with signal-to-noise problems. Lanczos residual bounds address this by providing two-sided constraints on energies that do not require assumptions beyond Hermiticity, but often give very conservative systematic uncertainty estimates. Here, a more-constraining set of gap bounds is introduced for hadron spectroscopy. These bounds provide tighter constraints whose validity requires an explicit assumption about an energy gap. Exactly solvable lattice field theory correlators are used to test the utility of residual and gap bounds at finite and infinite statistics. Two-sided bounds and other analysis methods are then applied to a high-statistics LQCD calculation of nucleon-nucleon scattering at $m_π\sim 800$ MeV. Generalized eigenvalue problem (GEVP) and Lanczos energy estimators are compatible when applied to the same correlator data, but analyses including different interpolating operators show statistically significant inconsistencies. However, two-sided bounds from all operators are consistent. Under the assumption that the number of energy levels below $NΔ$ and $ΔΔ$ thresholds is the same as for non-interacting nucleons, gap bounds are sufficient to constrain nucleon-nucleon scattering amplitudes at phenomenologically relevant precision. Lanczos methods further reveal that energy-eigenstate estimates from previously studied asymmetric correlators have not converged over accessible imaginary times. Nevertheless, data-driven examples demonstrate why assumptions are required to draw conclusions about the natures of two-nucleon ground states at these masses.

MicroBooNE is a liquid argon time projection chamber (LArTPC) neutrino detector located along the Fermilab Booster Neutrino Beam and 8 degrees off-axis to the Neutrinos at the Main Injector beam. MicroBooNE collected data from both beams accumulating a large neutrino-argon scattering dataset containing hundreds of thousands of events. Understanding neutrino-argon interactions is crucial for the next generation of neutrino oscillation experiments including DUNE. MicroBooNE has developed pioneering methodologies and novel reconstruction tools in order to benchmark models at very high sensitivity across the interaction phase space, including for ultra-rare channels. This proceeding presents an overview of the most recent MicroBooNE neutrino interaction results. These measurements span inclusive, CC0$π$, and rare channels including $Λ$, $K^+$ and $η$ production, providing invaluable datasets for constraining backgrounds and improving the modeling of neutrino scattering critical for the broader LArTPC neutrino physics program.

Recently, the KM3NeT Collaboration reported the detection of an ultra-high energy ($\sim 220$ PeV) neutrino event, KM3-230213A. In this work, we perform a detailed investigation into whether this event could originate from the diffuse neutrino flux produced by a class of pulsar-powered optical transients. In particular, we consider populations of ordinary supernovae (SNe), super-luminous supernovae (SLSNe), and luminous fast blue optical transients (LFBOTs) with a newly formed magnetar as the central engine. We discuss both the thermal electromagnetic and non-thermal neutrino emission from such sources. We scan the parameter space of the dipolar magnetic field strength and the initial spin period to determine characteristic optical emission properties and lightcurve timescales of these transients. Additionally, our scan identifies which classes of these transients can reproduce the required diffuse flux level and neutrino energies. Combining our results, we conclude that a diffuse neutrino flux from a population of LFBOTs can explain the KM3NeT event. Therefore, pulsar-powered optical transients may serve as promising sources for the current and upcoming high-energy and ultra-high energy neutrino telescopes.

We present a phenomenological study of partonic spin correlations and $\mathcal{CP}$ properties in $H\to gg$ decay channel at future lepton colliders. We investigate two classes of observables: Lund observable defined based on subjets and four-point energy-energy correlator (E4C) between particles inside two jets. Our results show that the E4C with energy weighted to the power of \(n=4\) achieves the strongest sensitivity to the spin correlations of gluons from Higgs boson decay. Under the assumption of ideal identification of different gluon splitting modes, we estimate that future lepton colliders operating at \(\sqrt{s}=240~\mathrm{GeV}\) with an integrated luminosities of \(5.6~\mathrm{ab}^{-1}\) can successfully probe gluon spin correlations, while \(20~\mathrm{ab}^{-1}\) of data can probe the $\mathcal{CP}$-mixing angle in the \(Hgg\) coupling to \(\lesssim 0.03π\) using E4C. We outline strategies for extending this framework to realistic detector-level analyses, which can provide a new pathway for the precision test of Standard Model and searches for new physics.

The primordial abundance of $^4$He plays a central role in big-bang nucleosynthesis (BBN) and in the cosmic microwave background (CMB). The LBT $Y_{\rm p}$ Project's new measurement of the primordial $^4$He mass fraction $Y_{\rm p} =0.2458 \pm 0.0013$ is the most precise determination to date. In this paper, we combine our new $Y_{\rm p}$ value with the latest primordial deuterium measurement, and assess the consequences for cosmology. For Standard BBN, where the number of light neutrino species is fixed at $N_ν=3$, the single free parameter is the cosmic baryon density; the CMB measures this independently, with results consistent with each other. Combining $Y_{\rm p}$ , D/H, BBN, and the CMB, gives the cosmic baryon-to-photon ratio $η= (6.120 \pm 0.038) \times 10^{-10}$, corresponding to a baryon density parameter $Ω_{\rm B} h^2 = 0.02236 \pm 0.00014$. We then allow $N_ν$ to vary and thus measure relativistic species present during nucleosynthesis. We find $η= (6.101 \pm 0.044) \times 10^{-10}$ or $Ω_{\rm B} h^2= 0.02229 \pm 0. 00016$, and $N_ν= 2.925 \pm 0.082$, and for $N_ν\ge 3$, $ΔN_ν= N_ν-3 \le 0.125$ (95\% CL) during BBN and the CMB. Our results demonstrate consistency with the Standard Model of particle physics, and with the standard cosmology that links BBN at $\sim 1 \ \rm sec$ and the CMB at $\sim 400,000$ yr.

CMB limits on cold-dark-matter isocurvature are often interpreted as excluding the simultaneous realization of high-scale inflation and large QCD axion decay constants in pre-inflationary Peccei--Quinn (PQ) scenarios. We show that this conclusion can be evaded by exploiting \emph{field-space geometry}. For a minimal complex PQ scalar with a $U(1)$-symmetric potential and nonlinear sigma-model kinetic term $dσ^{2}=dR^{2}+f^{2}(R)\,dθ^{2}$, a curved target-space metric endows the axion fluctuation with a time-dependent geometric mass during inflation, suppressing isocurvature without explicit PQ breaking and without extreme radial displacements. Specializing to a hyperbolic metric $f(R)=L\sinh(R/L)$ with curvature scale $L$, we find that for $R\gtrsim L$ the canonically normalized angular mode can be generically $\mathcal{O}(H_{\rm inf})$-heavy during radial slow-roll, dynamically damping CMB-scale fluctuations while producing a characteristic blue-tilted isocurvature spectrum. As a result, inflationary Hubble scales as large as $H_{\rm inf}\sim 10^{13}\,\mathrm{GeV}$ can be compatible with $f_a\sim 10^{14}$--$10^{16}\,\mathrm{GeV}$, reopening parameter space usually regarded as excluded. We present numerical benchmarks and a semi-analytic template that relates the scale-dependence of isocurvature to the geometric lever arm $R/L$, providing a direct phenomenological probe on PQ field-space geometry.

Ultra-peripheral collisions (UPCs) involving relativistic heavy ions are a unique laboratory to study quantum correlations. The intense electromagnetic fields generate high rates of photonuclear interactions, including events involving multiple photon exchange. Multiple photon exchange can result in the production of multiple vector mesons and/or nuclear excitations. These interactions share a common impact parameter, so the photons have the same linear polarization. The shared polarization entangles the particles, leading to unique quantum correlations. The decays of these vector excitations are sensitive to this polarization, allowing for the study of these correlations. This letter will compare classical and quantum calculations of the correlations between these azimuthal directions. The two approaches predict vert different angular correlations. The differences are akin to those seen with polarized photons in tests of Bells inequality. Uniquely, UPC photoproduction can produce final states containing three or more particles, all entangled with the same polarization. These more complex states exhibit additional unique phenomenology, allowing new tests of multi-particle entanglement.

Strong-field quantum electrodynamics offers a unique window into non-perturbative phenomena such as vacuum pair production, in which electron--positron pairs are created from the vacuum in the presence of intense electromagnetic fields. The LUXE experiment at DESY is designed to probe this regime using collisions between a high-intensity laser and the 16.5 GeV electron beam of the European XFEL. Future accelerator infrastructures, such as linear colliders, could extend these studies to even higher intensity and energy scales. Additionally, high-energy photons produced in such interactions can be used in beam-dump experiments to search for new physics.

We introduce the Celestial Energy-Energy Correlator (cEEC), an infrared and collinear safe observable that makes the celestial conformal symmetry of four-dimensional scattering manifest. The cEEC is defined as a correlation function of Average Null Energy operators measured on boost eigenstates, and takes the form of a four-point function in a fictitious two-dimensional CFT on the celestial sphere. An important feature of the cEEC is that it smoothly interpolates between different key regimes of perturbative gauge theory and gravity, such as the collinear limit, the Sudakov limit, and the Regge limit. We compute the cEEC to the first non-trivial order in $\mathcal{N}=4$ super Yang-Mills, pure Yang-Mills, Einstein gravity, and $\mathcal{N}=8$ supergravity. In $\mathcal{N}=8$ supergravity, the cEEC is uniquely determined by celestial symmetries and boundary data, demonstrating that bootstrap methods can yield closed-form results for this class of observables.

The NL-$e$EDM experiment searches for a non-zero electric dipole moment of the electron $d_e$ ($e$EDM) in the ground state of barium monofluoride (BaF). A beam of BaF from a supersonic expansion source is probed with the spin precession method presented in \cite{Boeschoten2024}. This method permits the extraction of an $e$EDM value as well as values for parameters causing a possible systematic bias leading to a false $e$EDM. The currently achievable sensitivity is limited by statistics collected in a period of 34 hours and yields an $d_e$ of $2(3) \times 10^{-25}$ $e\,$cm. Furthermore, from the same dataset sufficiently strong limits on parameters which can induce a false $e$EDM are extracted. These are mainly the electric field \textbf{E} and the intensity of the lasers fields in the fiducial volume of the experiment. We summarize the steps required to upgrade of the experiment to reach a competitive level on $d_e$, e.g. an intense laser-cooled beam from a cryogenic buffer gas source and the light collection efficiency of fluorescence.

We investigate the evolution of the PAH population's charge state and size across key physical zones in the Orion Bar, which include the HII region, the atomic PDR (APDR), and three HI/H2 dissociation fronts (DF1, DF2, and DF3). Utilising the NASA Ames PAH Infrared Spectroscopic Database (PAHdb) and the pyPAHdb spectral modelling tool, we analysed the MIRI-MRS observations of the Orion Bar from the "PDRs4All" ERS Program. pyPAHdb modelling reveals the fractional contribution of the different PAH charge states and sizes to the total PAH emission across the Orion Bar. Cationic PAH emission peaks in the APDR region, where neutral PAHs have minimal contribution. Emission from neutral PAHs peaks in the HII region that consists of emission from a face-on PDR associated to the background OMC-1 molecular cloud, and in the molecular cloud regions past DF2. PAH anions are observed deep within the DF2 and DF3 zones. The average PAH size ranges between ~$60-74$ Nc. The modelling reveals regions of top-down PAH formation at the ionisation front, and bottom-up PAH formation within the molecular cloud region. The PAH ionisation parameter $γ$ ranges between ~$2-9 x 10^4$. Intensity ratios tracing PAH ionisation scale well with $γ$ in regions encompassing edge-on or face-on PDR emission, but their correlation weakens within the molecular cloud zone. Modelling of the $5-15$ $μ$m PAH spectrum with pyPAHdb achieves comprehensive characterization of the net contribution of neutral and cationic PAHs across different environments, whereas empirical PAH proxy intensity ratio tracers can be highly variable and unreliable outside regions dominated by PDR emission. The derived average PAH size in the different physical zones is consistent with a view of PAHs being more extensively subjected to ultraviolet processing closer to the ionisation front, and less affected within the molecular cloud.

Self-interacting dark matter (SIDM) with a sufficiently large cross section has been shown to naturally produce constant dark matter (DM) cores, as well as core-collapse, at the centers of dwarf halos on cosmic timescales, potentially reducing tensions with observation. Here, we present halos from a new dark matter only (DMO) cosmological (SIDM) simulation: Ms.Marvel DMO with a velocity-dependent self-interaction cross section with $σ/m_\text{max} = 50$ cm$^2$/g at $v_\text{max} = 35$ km/s. We compare these to the CDM suite of Storm simulations including both DMO and dark matter + hydrodynamics runs, in order to test core-formation (and core-collapse) across different dark matter models. We show that Ms.Marvel DMO can reproduce core slopes consistent with observations of isolated dwarf galaxies and more massive ($\text{M}_{vir} \gtrsim 10^{10} M_{\odot}$) CDM dwarf halos that include stellar feedback from the matched CDM run (Storm CDM+baryons). We identify nine Ms.Marvel SIDM DMO halos in the core-collapse phase of gravothermal evolution with halo masses below $2\times 10^9 M_{\odot}$. We find that using core slope to measure the core-collapse timescales of Ms.Marvel DMO halos agrees well with predicted collapse times estimated with the parametric model for SIDM halos introduced by \cite{Yang2023}. Additionally, compared to central density, core slope is less sensitive to both the radius of measurement and halo merger history. These results indicate that the slope of the inner DM density profile more cleanly differentiates core-collapsed versus core-forming halos than central density amplitude.

Theoretical models of the evolution of supermassive black hole (SMBH) pairs in post-merger remnant galaxies are necessary to motivate observational searches for dual active galactic nuclei (AGN) and gravitational wave sources. Studies have explored the dynamical evolution of SMBH pairs under the influence of dynamical friction to calculate pairing times and predict the expected population of dual-AGNs at various redshifts. We formulate a three-dimensional dynamical model of SMBH pairs in the innermost kiloparsec of a post-merger galaxy to investigate the impact of orbital inclination with respect to the galactic disk on pairing times. The SMBH pairs are evolved in 81 different galaxy configurations initialized using a Gauss-Seidel Poisson solver. The dynamics are calculated for 12 distinct initial inclinations ranging from 0 to 75 degrees in each of the galaxies to gauge the impact of inclination on pairing time. Orbits characterized by initial inclinations greater than 20 degrees frequently require longer pairing times when compared to uninclined orbits. Pairing times for orbits with inclinations $\gtrsim 45$ degrees often exceed 14 Gyr. Galaxies with higher mass SMBH pairs and faster rotating disks generally shorten pairing times relative to galaxies with less massive or slower rotating disks when the inclination is $\lesssim 45$ degrees. The model suggests that SMBH pairs that form from mergers at inclinations $\lesssim 20$ degrees are likely progenitors of dual-AGN and gravitational wave sources.

Recent JWST observations have unveiled a numerous population of low-luminosity active galactic nuclei (AGN) at $4< z<10$, with space densities roughly an order of magnitude above pre-JWST estimates, and many of these AGN have masses orders of magnitude above the local black hole mass-stellar mass ($M_{\rm BH}-M_{*}$) scaling relations. We investigate the consistency of these observations within a data-driven framework that links the galaxy stellar mass function to the supermassive black hole (SMBH) mass function and AGN luminosity functions using different $M_{\rm BH}-M_{*}$ relations and the observed Eddington-ratio distribution. By comparing our predictions against observed AGN luminosity functions at $z\sim 5.5$ we find that observations can be reproduced either by highly-elevated $M_{\rm BH}-M_{*}$ relations paired with low duty cycles, or moderate relations with higher duty cycles. Through the Soltan argument, we find that $M_{\rm BH}-M_{*}$ relations that are modestly above the local relation for AGN produce consistency between multiple tracers of the SMBH demography at $z\sim 5.5$, while more extreme normalisations would require a weakly-evolving luminosity function at $z> 5.5$. Continuity-equation modelling shows that initially high $M_{\rm BH}-M_{*}$ relations predict a strong two-phase evolutionary scenario and very steep low-mass SMBH mass functions in tension with several current estimates, while more moderate relations generate local SMBH mass functions in better agreement with present determinations and near-constant scaling relations. Our results favour a scenario where SMBHs at $z \sim 5$ on average lie modestly above local AGN scaling relations, with elevated but physically plausible duty cycles. Future wide-field clustering and demographic studies will help break the remaining degeneracies between SMBH scaling relations and AGN duty cycles at early cosmic times.

Relativistic jets in active galactic nuclei (AGNs) are known to accelerate particles to extreme energies, yet the physical origin of very-high-energy (VHE) emission from low-luminosity AGNs (LL AGNs) remains unclear. NGC 4278, a local LLAGN, has recently been identified as a VHE source following detections by LHAASO. In this study, we present a multi-wavelength and multi-messenger analysis to investigate the physical origin of this emission. Swift-XRT monitoring reveals a quasi-quiescent state characterized by a low X-ray flux. Modeling the broadband spectral energy distribution with the leptohadronic code AMES, we find that a standard one-zone synchrotron self-Compton (SSC) model underpredicts the VHE flux by $\sim$70% due to the insufficient target photon density provided by the weak X-ray emission, unless a high Doppler factor ($δ\gtrsim 5$) is invoked. Alternatively, an external inverse-Compton (EIC) scenario-scattering seed photons from a radiatively inefficient accretion flow (RIAF)-successfully reproduces the broadband spectral energy distribution with a modest jet power and Doppler factor. We further explore the neutrino production within a leptohadronic framework. The predicted muon neutrino event rate is highest in the EIC quiescent model, reaching $N_{ν_μ} \sim 0.001$ for a 15-year IceCube observation (assuming 0.1% of the Eddington luminosity is partitioned into high-energy protons). Future multi-messenger observations are essential to unveil the details of the high-energy processes of NGC 4278.

There are two well documented models of gamma ray bursts (GRBs), the "Standard' model and the "Cannonball" model. They have often been reviewed [1] and sometimes compared [2]. Here, to avoid understandable biases, I show below the results of an experiment: letting an AI compare the data and the two models. All of what follows (but two references, two footnotes and the next sentence) is the result of asking Perplexity.ai to perform this confrontational task. It should be easy for an impartial reader to reach very clear conclusions.

The material in planet-forming disks determines the composition of planets; hence, it is crucial to understand the physical and chemical processes that set the abundance and distribution of key volatiles. James Webb Space Telescope observations of disks around very-low-mass ($\sim0.1~M\odot$) stars (VLMS) have revealed their hydrocarbon-rich inner regions (e.g., $\mathrm{C_2H_2}$), with column densities significantly higher than predicted. We employ chemical kinetics models using the physical structure of the inner disk around an M~Dwarf star with an X-ray luminosity of $L_X\sim10^{29}~\mathrm{erg~s^{-1}}$. We adopt initial abundances that mimic the effects of carbon enhancement and oxygen depletion (C/O from 0.44 to 87.47) and quantify how the abundances and distributions of key volatiles respond. The column density and number of molecules ($\mathcal{N}$) of hydrocarbons and oxygen-bearing species are highly sensitive to the C/O ratio, with the largest increases in hydrocarbons occurring when carbon increases by a factor of 2, and/or oxygen decreases by a factor of 10, relative to solar. In the IR-emitting region ($T_\mathrm{gas}>200$~K), a range of C/O ratios can reproduce the observed $\mathcal{N}$ and ratios relative to $\mathrm{CO_2}$. The disk-integrated molecular ratio with respect to $\mathrm{CO_2}$ is highly sensitive to the underlying C/O ratio. However, our results apply only to a source with a single X-ray luminosity value at the middle of that observed for VLMS; hence, a degeneracy between the stellar $L_X$ and the C/O ratio cannot be discarded. Nonetheless, our findings support that an enhanced C/O is required to drive the hydrocarbon-rich chemistry observed in the inner disks around VLMS.

We present a self-supervised approach for characterizing low surface brightness tidal features in wide-field imaging data by applying the nearest-neighbor contrastive learning of visual representations (NNCLR) algorithm to a curated subset of the Dark Energy Survey Data Release 2 (DES DR2). We construct 38,334 cutouts of well-resolved galaxies in the g, r, i bands, applying a novel "tiered sigmoid scaling function" to dynamically adjust image contrast according to the object's signal-to-noise and background level. A supplemental labeled sample of 366 galaxies enables qualitative assessment of the learned embeddings. We train a convolutional neural network with image augmentations including injection of simulated background stars, and project the resulting 512-dimensional representations into two dimensions using uniform manifold approximation and projection (UMAP) and its local density preserving variant (densMAP). We find that the NNCLR latent space recovers global trends corresponding to major merger features, yet does not reliably separate stellar streams without further supervision. To interpret the network's implicit attention, we compute gradient-based saliency maps averaged over the full dataset: these reveal that the tiered sigmoid scaling effectively attenuates information from the center of the image cutouts, thereby suppressing the learning of high surface brightness features of each image cutout's central galaxy. Our study provides a blueprint for leveraging contrastive methods to mine forthcoming survey data for faint tidal substructure, and highlights key preprocessing and interpretability considerations for robust stream detection.

Presence of supermassive black holes (SMBHs) with mass $(10^{6}-10^{9}) M_{\odot}$ at $z = 10$ has been recently revealed by James Webb Space Telescope (JWST) observations. In this study we generate seeds for the above range of SMBHs in various background cosmologies. We consider cosmic timescales required for black hole growth provided by three general relativistic cosmological models ($Λ$CDM, $ω$CDM and Dynamical Dark Energy(DDE) and the braneworld cosmology. The growth of SMBHs is studied through Eddington limited and super-Eddington accretion, where the accretion starts at z=30. It is found that growth of SMBHs by z=10 within Eddington limited accretion is possible through massive seeds $(M\geq10^{4}M_{\odot})$ in all cosmologies. Super Eddington accretion onto spinning black holes with mass of few tens of solar masses can result in SMBHs by z=10 in all cosmologies. The viable cosmologies considered here are found to be unable to strongly distinguish between the seed black hole masses. The seeds generated in this work are assumed to be of primordial origin in order to satisfy the criteria of formation of high redshift massive galaxies. The fraction of primordial black holes (PBHs) contributing to dark matter ($f_{PBH}$) and their corresponding number densities for the mass range ($10^{5}-10^{8}$) $M_{\odot}$ are calculated in both seed effect and Poisson effect. In seed effect, PBHs of mass $\geq 10^{7} M_{\odot}$ contributes $\leq 10^{-2}$ to the dark matter fraction. The evolution of gas mass inside a PBH seeded dark matter halo is studied. The ratio of black hole to stellar mass is also evaluated for star formation efficiency in the range (0.1-1) and found to be ($10^{-3}-1$) for $M_{BH}=10^{8} M_{\odot}$ and ($10^{-2}-10$) for $M_{BH}=10^{9} M_{\odot}$.

We present the first detailed multi-tracer observation of a 5-pc long outer Galaxy filament, G183, and the massive young stellar object (YSO) IRAS 05480+2545 associated with it. Using the IRAM 30-m telescope at lambda = 1.4 and 3 mm, we probed the molecular gas distribution at angular resolutions of ~12"-28" (0.1-0.3 pc at d = 2.1 kpc). The velocity-resolved C18O(1-0) observations conclusively show a main filament with a skeleton of ridges. The main filament is a 5 pc long velocity-coherent structure with a continuous and quiescent velocity field along its length up to the star-forming hub that accretes mass from the filament. The internal gas kinematics of most of the G183 filament is dominated by thermal motions (sigma_NT/cs~1) and large-scale velocity gradients arising due to outflows and accretion of matter in the massive YSO. The dispersion-size relation almost up to 1 pc is consistent with Larson's law, suggesting that the origin of the filament is a turbulence cascade. The massive YSO, S1, with no corresponding radio continuum detection is characterized as a high-mass protostellar object with a mass of 156 Msun and an M/L ratio of 0.04. We identify a kinematic signature of the accretion of material from the filament onto the YSO, S1. The rates of molecular gas accretion and entrainment in S1 are estimated to be 8.6 and 2.6 (in units of 10^-4 Msun/yr), respectively. In comparison to the inner Galaxy high-mass star-forming filaments forming massive stars, G183 has a lower column density; however, the accretion and outflow rates in S1 are similar. The detection of hydrocarbons such as CH3CN and HC3N indicates the presence of hot-core chemistry in S1. These results highlight the universality of physical processes involved in massive star formation across a range of Galactic environments.

We investigate the formation of the large-scale cosmic structure in a scalar-tensor theory of gravity belonging to the class of the Brans--Dicke theories. The universe contains baryonic matter alone and neither dark matter nor dark energy. The two arbitrary functions of the scalar field characterizing the kinetic term and the self-interaction potential are set to $W(\varphi)=-1$ and $V(\varphi) = -Ξ\varphi$, respectively, with $Ξ$ a positive constant. In the weak-field limit, the theory reduces to Refracted Gravity, a non-relativistic theory whose modified Poisson equation contains the scalar field $\varphi$ that provides the gravitational boost required to describe the dynamics of galaxies and galaxy clusters without dark matter. In a flat, matter-dominated, homogeneous and isotropic universe the same scalar field $\varphi$ drives the accelerated expansion of the universe and describes the observed redshift evolution of the Hubble-Lemaître parameter $H(z)$. However, in the equation of the growth factor of the linear perturbation theory, the form of $V(\varphi)$ makes the coefficient of the source of the gravitational field proportional to $H^{-1}(z)$; therefore the gravitational field is strongly suppressed at early times and structure formation is delayed to redshift $z< 1$, in disagreement with the observation of formed galaxies at much larger redshifts. In addition, the form of $W(\varphi)$ and a linear $V(\varphi)$ imply that $\varphi$ generates twice the gravitational boost on massive particles than on photons, with possible observable consequences on the gravitational lensing phenomenon. It remains to be investigated whether different choices of $W(\varphi)$ and $V(\varphi)$, that can still make the theory reduce to Refracted Gravity in the weak-field limit, might alleviate these problems.

We present long-slit near-infrared spectroscopic observations of six nearby (z$\sim$0.1) radio-quiet type-2 quasars (QSO2s) from the Quasar Feedback (QSOFEED) sample. They have bolometric luminosities of $10^{45-46}~erg~s^{-1}$ and stellar masses of $ 10^{10.6-11.3}~M_{\odot}$. The observations were obtained with the instrument Espectrógrafo Multiobjeto Infra-Rojo (EMIR) on the 10.4 m Gran Telescopio Canarias. The nuclear K-band spectra (central $\sim$1-3 kpc of the QSO2s) reveal signatures of high-velocity outflows in either the Pa$α$ or Br$γ$ lines, depending on the redshift, and in the [Si VI] lines. The broadest kinematic components have full width at half maximum (FWHM) of $\sim$1200-2500 km $s^{-1}$. From the near-infrared hydrogen recombination lines we derived ionized outflow masses of $M_{Hion} \sim0.08-20\times 10^{6}~M_{\odot}$, mass outflow rates of $\dot{M}_{Hion}\sim0.03-6~M_{\odot}~yr^{-1}$, and kinetic powers of $\dot{E}_{Hion}\sim 10^{37.8-40.8}~erg~s^{-1}$. These ionized gas outflow masses and mass outflow rates have median values that are 5.9 and 5.8 times larger, respectively, than those derived from the [Si VI] line. Our study provides evidence, at least for these six QSO2s, that the near-infrared recombination lines and [Si VI] are tracing the same outflow (i.e., they have similar kinematics and radii), but they carry different amounts of mass. We detected warm molecular lines in the six QSO2s, from which we measured total (nuclear) gas masses from 1.1 (0.7) to 32 (13) $\times~10^3~M_{\odot}$, similar to other QSO2s with warm $H_2$ measurements reported in the literature, but we did not find any molecular outflow associated with them. Comparing with other five QSO2s with $H_2$ measurements reported in the literature, we find that the four QSO2s with detected $H_2$ outflows have total (nuclear) $H_2$ masses 2.2 (2.7) times larger, on average.

We investigate the impact of equatorial asymmetry in the magnetic field geometry on accretion dynamics around a spinning black hole using axisymmetric general relativistic magnetohydrodynamic simulations. We consider a Fishbone--Moncrief torus orbiting a Kerr black hole with spin parameter $a = 0.9375$, threaded by large-scale magnetic fields that are asymmetric about the equatorial plane. The degree of equatorial asymmetry in the magnetic field is parametrized by an angle, with values of $30^\circ$, $45^\circ$, and $60^\circ$. We examine how this equatorially asymmetric initial magnetic field configuration influences the magnetic field structure, accretion flow morphology, and angular momentum transport across a range of initial plasma beta values ($β= 0.007, 0.005, 0.001$). We find that such deformation in the magnetic field leads to noticeable changes in the inner disk structure, asymmetric outflow patterns in the poloidal plane, and time-dependent variations in accretion rates. These effects are generally more pronounced at lower beta values, where magnetic pressure dominates; in particular, the $30^\circ$ case at $β= 0.001$ exhibits strong and persistent asymmetric inflows and outflows. Our results demonstrate that equatorially asymmetric magnetic field configurations can significantly influence the structure and variability of relativistic accretion flows. These findings motivate future extensions to full three-dimensional studies, where black hole magnetosphere can be explored in a more general setting.

Small polycyclic aromatic hydrocarbons and somewhat larger cyano derivatives were detected in the cold dark cloud TMC-1 recently. Their formation from smaller hydrocarbons is not well understood, in part because abundances of many species are not known. Methylallene, CH$_3$CHCCH$_2$, may be one of the building blocks, but its rotational spectrum was characterized only to a very limited extent. We recorded rotational transitions in the 36$-$501 GHz region to extend the existing line list of methylallene and thus enable searches for the molecule in space. Quantum-chemical calculations were carried out to evaluate initial spectroscopic parameters. We obtained transition frequencies with $J \le 61$ and $K_a \le 21$ and resolved the internal rotation splitting of the CH$_3$ group at least partially. As a result, a full set of distortion parameters up to sixth order along with two octic ones were determined, as well as parameters describing the internal rotation of the methyl group. The spectroscopic parameters are accurate enough to identify methylallene up to 720 GHz, sufficient for searches even in the warm interstellar medium.

A major challenge in identifying massive quiescent galaxies at $z>3$ is distinguishing truly passive systems from dust-obscured star-forming galaxies, as both populations exhibit similar red ultraviolet (UV)-to-near-infrared (NIR) colors. In this work, we present ALMA Band 7 dust-continuum observations of five ultramassive galaxies (UMGs; $\log (M_\star / M_\odot) > 11$) spectroscopically confirmed at $z_{\rm spec} > 3$ from the MAGAZ3NE survey. Our results reveal that only one galaxy shows a faint 870 \um\ dust continuum detection, while the remaining four UMGs are undetected down to the $3σ$ depth . By incorporating ALMA constraints into the spectral energy distribution analysis, we confirm that these UV-NIR-selected systems are truly quiescent UMGs, lying more than one dex below the star-forming main sequence with $\mathrm{\log (sSFR/Gyr^{-1}) < -1}$, thereby ruling out the possibility of obscured star formation. We then estimate dust masses using both spectral energy distribution modeling and modified blackbody fitting, with consistent results between the two methods. We find that three UMGs have evolved into extremely dust-poor quiescent galaxies, with $M_{\mathrm{dust}}/M_\star \lesssim 10^{-4}$, while the ALMA-detected galaxy has a comparatively higher dust reservoir with $M_{\mathrm{dust}}/M_\star \sim 10^{-3}$. Our results present the most massive and extremely dust-poor spectroscopically confirmed quiescent galaxies known at $3 < z < 4$, providing valuable observational constraints on rapid dust removal and quenching processes in the early universe. Future molecular line observations will be essential to directly measure the gas content and verify the efficiency of the depletion process.

Powerful outflows along the accretion disk, known as disk winds, are sometimes launched in black hole X-ray binaries. These winds often manifest themselves in X-ray spectra as blueshifted, highly ionized absorption lines. Previous observations suggest that the mass loss rate from the disk due to disk winds can be comparable to or even more than the mass accretion rate onto the black hole, indicating that disk winds likely play crucial roles in shaping the accretion disk structure and affecting the surrounding environment. However, the mechanisms driving these winds, as well as how their structure changes in response to variations in the mass accretion rate, remain poorly understood. The X-ray Imaging and Spectroscopy Mission (XRISM), launched in September 2023, is equipped with Resolve, a cutting-edge X-ray micro-calorimeter that delivers unprecedented spectral resolution. Resolve is expected to significantly advance our understanding of wind launching mechanisms and their impact on accretion processes and environments. In this article, we review the progress made in the pre-XRISM era, highlight key results obtained from XRISM observations to date, and outline future prospects.

Methyl cyanide, CH$_3$CN, is present in diverse regions in space, in particular in the warm parts of star-forming regions where it is a common molecule. Rotational transitions of $^{13}$CH$_3$CN and CH$_3$$^{13}$CN in their $v_8 = 1$ lowest excited vibrational states ($E_{\rm vib} \approx 520$ K) are quite prominent in Sagittarius B2(N). In order to be able to search for transitions of the next higher vibrational state $v_8 = 2$, we recorded spectra of samples enriched in $^{13}$CH$_3$CN and CH$_3$$^{13}$CN up to $v_8 = 2$ in the 35 to 1091~GHz region and reinvestigated existing spectra of CH$_3$CN in its natural isotopic composition between 1085 and 1200 GHz. Perturbations caused by near-degeneracies in $K = 4$ of $v_8 = 2^0$ and $K = 2$ of $v_8 = 2^{-2}$ yielded accurate information on the energy spacing of 22.93 and 21.79 cm$^{-1}$ between the $l$-components of $^{13}$CH$_3$CN and CH$_3$$^{13}$CN, respectively. Fermi-type interaction between $K = 13$ and 14 of $v_8 = 1^{-1}$ and $v_8 = 2^{+2}$ probe the energy differences between the two states of both isotopomers. In addition, a $ΔK \pm2$, $Δl \mp1$ interaction between the ground vibrational state of $^{13}$CH$_3$CN and $v_8 = 1^{+1}$ provides information on their energy spacing. Furthermore, we obtained improved or extended ground state rotational transition frequencies of $^{13}$CH$_3$$^{13}$CN and extensive data for $^{13}$CH$_3$C$^{15}$N and CH$_3$$^{13}$C$^{15}$N. Finally, we report the results of our search for transitions of $^{13}$CH$_3$CN and CH$_3$$^{13}$CN in their $v_8 = 2$ states toward Sagittarius B2(N).

When solving scientific and applied problems, such as latitude monitoring, it is important to correctly exclude primary cosmic ray variations from observation data. Therefore, the purpose of this study was to develop and implement a method for correcting monitoring data, the key point of which was to obtain reception coefficients as a function of latitude. This resulted in an approximation of the rigidity dependence of the zeroth and first harmonic cosmic rays anisotropy coefficients, calculated for a ground-based cosmic ray detectors network. Analysis of the obtained results showed that the approximation was performed with high accuracy, and the results are suitable for use in latitude measurements during marine expeditions.

This article presents the observations of the extreme ground-level enhancement (GLE #77) of Solar Cycle 25 that occurred on 11 November 2025, using ground-based NaI(Tl) gamma-ray detectors deployed at Arctic and Antarctic stations, together with neutron monitor data and particle measurements from the GOES-18 satellite. The event was associated with an intense X-class solar flare and a strong solar energetic proton event. This paper reports the first ground-based detection of a GLE using gamma-ray detectors operating simultaneously in both polar regions, which are concurrent with increases in neutron monitor counts. Thus highlights the capability of polar gamma-ray detectors to complement traditional neutron monitor observations during extreme solar proton events. A detailed analysis revealed distinct prompt and delayed responses during the event evolution. Interestingly, the signature of the prompt peak of GLE#77 (at 10:38 UT) was observed up to high-rigidity neutron monitors (low latitudes). However, the delayed peak (at 13:08 UT) was not seen at the stations with rigidity > 6 GV. The timing of the prompt and delayed peaks coincided with the proton flux peaks observed by the GOES-18 satellite at energies > 150 MeV and 12-99 MeV, respectively. It is observed that the GLE amplitude has a strong dependence on geomagnetic cutoff rigidity and has a weak solar zenith angle dependence.

Context: The dynamical evolution of binary populations in embedded star clusters shapes the statistical properties of binaries observed in the Galactic field. Accurately modelling this process requires resolving both early cluster dynamics and binary interactions. Aims: We aim to characterize the early dynamical evolution of primordial binaries in embedded clusters and identify the key parameters that govern binary survival and disruption. Methods: We perform a set of direct $N$-body simulations starting from 100\% primordial binaries in a time-varying gas potential of a gas-embedded cluster. To describe the evolution of binary orbital properties, we define empirical dynamical operators for period, binding energy, and mass ratio, and calibrate them across the simulated ensemble. Results:The binding energy and orbital period evolve in a consistent, sigmoidal fashion. Their dynamical operators reveal that hard binaries heat the cluster and suppress wide binary formation, while a small residual population of soft binaries survives. The evolution of the mass-ratio distribution is less directly linked to dynamical processing and more shaped by internal processes such as stellar physics process in the pre-main-sequence phase. High-$q$ systems tend to be enhanced, while low-$q$ systems are prone to disruption. Conclusions: The binary evolution in clusters is primarily governed by binding energy and orbital period. Our model improves over previous parameterizations of the dynamical operator by allowing for the existence of wide binaries and incorporating the embedded cluster phase. For individual clusters, direct $N$-body modelling remains the only reliable approach. On Galactic scales, population synthesis methods based on the stellar dynamical operator approach developed here remain essential.

Gamma-ray bursts (GRBs) are one of the most extreme transients in the universe, but their explosion and emission mechanism remains unclear. To investigate the nature of GRB jets, here we focus on X-ray flares (XFs) and extended emissions (EEs), which are X-ray emissions that occur 100 to 1000 seconds after the main burst. They can be observed by recently developed multi-wavelength facilities. In this paper, we calculate emissions across multi-wavelengths associated with XFs and EEs under the hypothesis that XFs and EEs are optically-thin synchrotron emissions from nonthermal electrons in relativistic jets. Considering ranges of the dissipation radius $r_{\rm diss}$ and the Lorentz factor $Γ$ of the jet, we determine the parameter space in which a detectable emission can be produced at each wavelength. We found that simultaneous ultraviolet and very-high-energy gamma-ray emission associated with XFs or EEs can be detected by Swift/UVOT, SVOM/VT, and CTAO approximately every three years. The detection and non-detection rates for each detector are key to determining the uncertain yet essential values necessary for understanding the physics of GRB jets.

It is shown that Eddington-like accretion event in the Galactic center several million years ago and particle acceleration at accompanying shocks and jets could explain the observed cosmic ray spectrum at energies above 1 PeV. Cosmic ray particles are confined in extended (several hundred kiloparsec in size) Galactic halo. It is shown that the halo magnetic field could be as small as $2\times 10^{-7}$ G for the effective confinement.

We present the discovery of radio emission from the Be/X-ray binary A0538-66 with the Australian Square Kilometre Array Pathfinder (ASKAP), and results from a subsequent weekly monitoring campaign with the MeerKAT radio telescope. A0538-66, located in the Large Magellanic Cloud, hosts a neutron star with a short spin period ($P \approx 69$ ms) in a highly eccentric $\approx16.6$-day orbit. Its rare episodes of super-Eddington accretion, rapid optical and X-ray flares, and other peculiar properties make it an interesting system among high-mass X-ray binaries. Our MeerKAT data reveal that it is also one of the most radio-luminous neutron star X-ray binaries observed to date, reaching $\approx 3 \times 10^{22}~\text{erg}~\text{s}^{-1} \text{Hz}^{-1}$, with radio emission that appears to be orbitally modulated. We consider several possible mechanisms for the radio emission, and place A0538-66 in context by comparing it to similar systems.

Observations of X-ray luminous elliptical galaxies suggest that the accretion rate onto the central supermassive black hole can reach a substantial fraction of the Bondi rate. However, classical accretion theory applicable to such hot accretion flows treats spherically symmetric Bondi accretion and disc-like advection-dominated accretion flows (ADAFs) as two distinct limiting cases, lacking a unified framework for flows with finite angular momentum. In this work, we develop such a framework that continuously connects these two regimes. Our model naturally recovers the Bondi solution in the limit of vanishing angular momentum and approaches the properties of classical ADAFs at high angular momentum, while providing a physically well-defined description of the intermediate regime where neither limiting case is strictly applicable. We further demonstrate that the accretion rate is jointly regulated by the angular momentum of the ambient gas and the gas viscosity. For sufficiently large but physically reasonable viscosity, the accretion rate can remain at a significant fraction of the Bondi rate even in the presence of substantial gas rotation. These results offer a natural explanation for how such accretion rates can be sustained despite finite angular momentum in realistic galactic environments.

The standard $Λ$CDM cosmological model predicts that cosmic filaments are highly clumpy, whereas warm dark matter -- invoked to address small-scale challenges in $Λ$CDM -- produces filaments that are noticeably smoother and less structured. In this work, we investigate the potential of Lyman $α$ (Ly$α$) emission to trace cosmic filaments at redshifts $z=2.5$ and $z=4$, and assess their potential for constraining the nature of dark matter. Our analysis shows that Ly$α$ filaments provide a promising observational probe of dark matter: at $z=4$, differences in filament smoothness and surface brightness serve as distinctive signatures between models. Looking ahead, the upcoming generation of 30-meter class telescopes will be critical for enabling these measurements, offering a compelling opportunity to distinguish the nature of dark matter by mapping the structure of cosmic filaments.

Mass loss through stellar jets is closely tied to the process of accretion through the disk. Understanding phenomena such as episodic ejections and outflow asymmetries can thus shed light on the mechanism of jet launching and its connection to both mass accretion and the evolution of the protoplanetary disk. We use new VLT/MUSE Narrow Field Mode observations of the Classical T Tauri Star Th 28 to map the jet structures within 6'' of the source at an effective angular resolution of 0.''12, provided by the combination of the AO correction and image deconvolution. The emission line profiles and flux ratios are investigated and diagnostic analysis of the optical forbidden emission lines (FELs) is used to estimate the electron density, ionisation fraction, electron temperature and shock velocities in both jet lobes within 200 au of the star. The mass outflow rates in each lobe are obtained using the derived total densities and FEL luminosities and compared with the mass accretion rate. We identify several new knots in both jet lobes which have been ejected in the previous 10 years on a timescale of 3-6 years, which is significantly more frequent than previously estimated. In both lobes we find comparable mass outflow rates close to the jet base. Th 28 has undergone a significant rise in mass accretion rate between 2014 and 2023, which may be linked to the most recently ejected knot pair detected in each side of the jet. The red-shifted jet mass outflow rate shows a similar increase of a factor 2, indicating that the ratio of mass outflow to accretion remains constant. A moderately lower mass outflow rate is found in the faster blue-shifted lobe, supporting the possibility that momentum ejection is conserved on each side of the jet. The frequent knot ejections indicate that this source is a good target for further monitoring to study the accretion-ejection connection.

The size-velocity dispersion ($σ$) relation, while well established for giant HII regions, remains uncertain for their smaller counterparts (physical radii R < 20 pc). Thanks to the LAMOST MRS-N dataset's large sky coverage and high spatial/spectral resolution, we examined this relationship using 10 isolated Galactic HII regions with R < 20 pc. Our results reveal two key findings: (1) these small-size HII regions remarkably follow the same size-$σ$ relation as giant HII regions, suggesting this correlation could serve as a novel distance indicator for Galactic HII regions; and (2) we find distinct dynamical behaviors between younger and older HII regions. Specifically, in younger (< 0.5 Myr), ionization-bounded HII regions, the velocity dispersion shows no correlation with expansion velocity, indicating that turbulence is driven primarily by stellar winds and ionization processes. In contrast, in older (> 0.5 Myr), matter-bounded HII regions, a clear correlation emerges, implying that expansion-driven processes begin to play a significant role in generating turbulence. We therefore propose an evolutionary transition in the primary turbulence mechanisms, from being dominated by stellar winds and radiation to being increasingly influenced by expansion-driven dynamics, during the evolution of HII regions. Considering the small sample size used in this work, particularly the inclusion of only two young HII regions, which also have large uncertainties in their expansion velocities, further confirmation of this interpretation will require higher-resolution 2D spectroscopy to resolve blended kinematic components along the line of sight for more accurate estimation of expansion velocities, along with an expanded sample that specifically includes more young HII regions.

To deepen our understanding of star-forming properties, we studied a massive protocluster IRAS 15596-5301 using ALMA 870 um and 3 mm data. High-resolution 870 um data reveal 34 dense cores, including 3 hot molecular cores, with subsequent line surveys detecting 22 molecular species toward them. Two velocity components (I15596-red/I15596-blue) were found in the averaged H13CO+(1-0) spectrum, and two filaments were identified from velocity-resolved integrated intensity maps. A spatial overlap between the two filaments was observed, and this overlapping region exhibits a distinct bridge-shaped feature in the position-velocity diagram constructed along the entire filamentary structures. Combined with the reduced H13CO+/HCO+ ratio in the overlapping region and the three-dimensional position-position-velocity cube data, we conclude that a non-head-on collision occurs between the edges of the two filamentary structures in IRAS 15596-5301. Cluster analysis demonstrates that clusters located in the collision region host more evolved chemical rich dense cores than their counterparts in other regions. Our results thus indicate that star formation in I15596 is triggered or accelerated by a mild non-head-on collision between two filaments.

We map the extraplanar gas, with $\sim$50-200 pc resolution, in nine star-forming galaxies using Multi-Unit Spectroscopic Explorer (MUSE) observations from the GECKOS VLT Large Program targeting edge-on galaxies with similar stellar mass as the Milky Way. The narrow range in stellar mass ($\pm0.35$ dex) of the GECKOS sample makes it ideal for studying trends with star formation rate (SFR). We find strong extraplanar emission reaching $\sim$2-8 kpc from the disk midplane in all targets with $\rm{SFR}\geq$1 M$_{\odot}$ yr$^{-1}$. Targets with SFR$\,\geq\,$5 M$_{\odot}$ yr$^{-1}$ have brighter, more extended H$α$ emission compared to lower SFR targets. In high-SFR systems, the gas velocity dispersion ($σ_{\rm Hα}$) shows a biconical morphology, consistent with the expectation of outflows. This agrees with previous works suggesting high velocity dispersion in a biconical shape is a good means to identify outflows. We find mixed results using line diagnostics ([OIII]$_{5007}$/H$β$ - [NII]/H$α$ and $σ_{\rm Hα}$ - [SII]/H$α$) to spatially resolve ionisation mechanisms across the extraplanar gas. The highest [NII]/H$α$ are the extraplanar gas of the highest SFR systems, yet main-sequence galaxies have the highest [OIII]/H$β$. While the morphology of [NII]/H$α$ may be useful to identify outflows, the absolute value of the line ratio alone may not distinguish strong outflows from extraplanar gas of main-sequence galaxies. The ubiquitous extraplanar emission can be interpreted as the result of feedback, in the form of large-scale winds for starbursts or smaller-scale galactic fountains for main-sequence galaxies. Moreover, shock-heating may ionise gas at the interface of the disk and the circumgalactic medium, independent of the source of the gas.

The third data release of the Gaia mission (Gaia DR3) has enabled large-scale searches for dormant black hole and neutron star binaries with stellar companions at wide separations. A recent study has proposed thousands of dormant black hole and neutron star binary candidates using summary statistics from Gaia DR3 by simulating and fitting Gaia observables. In this Letter, we perform broadband spectral energy distribution (SED) fitting from the optical to the infrared for 1,328 candidates, incorporating GALEX ultraviolet photometry to assess the presence of hidden hot companions. We quantify ultraviolet excess by comparing observed near-ultraviolet fluxes with single-star SED predictions and further test whether excesses can be explained by non-degenerate stellar companions for sources exhibiting moderate excess. We additionally examine the Galactic kinematics of the sample to identify systems potentially affected by natal kicks during compact-object formation. By combining the ultraviolet and kinematic diagnostics, we identify 176 sources as the highest-priority candidates for follow-up observations, in which 19 are black hole candidates with previously provided masses $\geq$ 3 $M_\odot$.

A number of candidate repeating partial tidal disruption events (rpTDEs) have been reported in recent years. If these events are confirmed, the high fraction of observed rpTDEs among all tidal disruption events (TDEs) is in tension with prediction of the loss cone channel. We further point out an inequality $M_\bullet \lesssim 4\times 10^6 M_\odot (T_{\rm obt}/10\ {\rm yr})^{4/9}$ that must be satisfied for rpTDEs of solar type stars in the loss cone channel, where $M_\bullet$ is the central supermassive black hole (SMBH) mass and $T_{\rm obt}$ is the orbital period of the star. However the majority of reported rpTDE candidates potentially violate this inequality, indicating an alternative formation channel. In the commonly invoked Hills mechanism, the captured stars produced by tidal disruption of near-contact binaries can evade this inequality and may be the dominant source of rpTDEs. If the same process operates in the Galactic Center, there should exist a population of hypervelocity stars (HVSs) ejected with velocities as high as $3.6\times 10^3 (M_\bullet/10^6 M_\odot)^{1/6}\ {\rm km\ s}^{-1}$, which however have not been detected. A complete search for HVSs in the Milky Way will be critical for testing this prediction.

We present revised stellar ages for 23 pre-main sequence K- and M-type stars in the Upper Scorpius star-forming region, derived by using stellar dynamical masses to constrain isochronal ages from five pre-main sequence stellar evolutionary models. We find that mass-constrained stellar ages for all model sets are more consistent with the older, ~8-11 Myr age for Upper Sco derived using earlier-type stars. Additionally, applying the independent mass constraint to isochronal ages tends to 1) increase stellar ages for most model sets, and 2) decrease age scatter for individual sources between model sets. Models that account for global magnetic fields consistently produce the best match to our observations: they change comparatively little when the mass constraint is applied, and produce 9-10 Myr ages under both unconstrained and mass-constrained conditions. Most standard (nonmagnetic) models produce younger ages (3-5 Myr) when unconstrained, but older ages (6-9 Myr) when constrained by dynamical mass. Our results are consistent with literature findings that suggest median disk lifetimes may be >2x longer than previously thought.

XRISM has provided an unprecedented view of the emission and absorption lines in the X-ray. Notably, early results showed significant complexity to the Fe-K$α$ line profile in AGN, with clear contributions from at least three emitting structures: an inner disc, intermediary broad line region (BLR) scale material, and an outer torus. This poses a new challenge for the modelling of the emission lines, as while fast sophisticated models exist for disc line-profiles, large scale-height material is typically much more complex. In this paper we aim to address this gap, by building a fully analytic model for the emission line profiles from a wind, aimed towards BLR scale material, motivated on previous reverberation studies suggesting a wind on the inner edge of the BLR. Our approach gives a physically motivated, yet computationally fast, model for the intermediary component to the Fe-K$α$ complex seen in the XRISM data. We demonstrate our model on the XRISM observations of NGC 4151 from the performance verification phase, showing that it gives a good description of the data, with physically reasonable parameters for BLR scale material. We also show that our model naturally gives the smooth line profile seen in the data, due to the large spatial extent of a wind. Finally, we make our model code public to the community, and name it xwind.

Tensions in late-time expansion data have renewed interest in models beyond $Λ$CDM. We ask: \emph{how isotropic must dark energy be?} Working in Bianchi~I, we allow time-dependent anisotropic stress and introduce a parameterisation that enforces a vanishing line-of-sight integral of the shear, thereby satisfying the CMB ISW quadrupole bound by construction. Using Pantheon+SH0ES SNe together with DESI BAO distances, single-bin (constant) and five-bin anisotropic models improve the fit over $w$CDM by $Δ(-2\ln L_{\rm iso})=14.8$ and $26.6$ respectively, but both violate the quadrupole constraint. In contrast, a five-bin constrained model achieves $Δ(-2\ln L_{\rm iso})=15.4$ while remaining compatible with the quadrupole limit. The fit improvement arises from two sources: capturing directional structure in the Pantheon+ SNe data, and partially alleviating the tension between the SH0ES $H_0$ value and DESI BAO distances.

The physical origin of the observed cosmic neutrinos remains an open question and the subject of active research. While matter accretion onto supermassive black holes is long thought to accelerate particles to high energies, it has recently been suggested that tidal disruption events, and accretion flares in general, with prominent IR echoes can account for a fraction of the diffuse high-energy neutrino signal. Motivated by this result, we compile a sample of nearby accretion flares detected in the NEOWISE survey featuring strong IR echoes, and we cross-match it with the latest catalog of neutrino alerts, IceCat-1. We recover only a single spatial coincidence between the two catalogs, consistent with a chance coincidence. We find no temporal and spatial coincidences between the two samples, which, given the properties of our sample, appears to challenge previous conclusions. We discuss the physical implications of our results and potential future explorations.

We review a subset of the current tensions affecting the standard $Λ$CDM cosmological model, emphasizing the role of chronic systematics and significance inflation in shaping their interpretation. As a unifying framework, we consider the spherically symmetric inhomogeneous $Λ$LTB model and use it as a set of "glasses" through which to reinterpret the Hubble, dipole, and dark-energy tensions. Large-scale spatial gradients in this model introduce anisotropic expansion and position-dependent observables, allowing local estimates of $H_{0}$ to shift, dipolar signatures to arise, and an apparently evolving dark-energy equation of state to be mimicked without invoking genuinely dynamical dark energy. We discuss how these effects are constrained once the full supernova, CMB, and large-scale-structure data sets are included, and argue that it remains unclear whether any single $Λ$LTB configuration can simultaneously account for all major anomalies. More broadly, we highlight that cosmology currently lacks a widely accepted baseline model that is both theoretically well founded and capable of accommodating the Hubble and dark-energy tensions, leaving us without a true concordance framework for forecasting future surveys.

We present VLT/ERIS IFU J-band observations of MCG-03-34-64, a nearby (z = 0.0167) Seyfert galaxy hosting a candidate dual AGN system with a separation of ~100 pc between the nuclei. The observations cover, among others, the HeI1.083um, [FeII]1.257um and Pa$β$ emission lines, over a FoV of 3"x3"(~1x1 kpc$^2$). We analyse the ionised gas kinematics and identify two regions with enhanced velocity dispersion (W80~1500 km/s), suggestive of fast outflowing gas, spatially coincident with the position of the two candidate active nuclei. The spectra of the two outflows show a prominent blueshifted wing with velocities vmax ~ -1700 km/s corresponding to the highest 2-5 percentiles of samples of local AGN with similar bolometric luminosities. For the ionised phase of the two outflows, we derive comparable masses of $(4\pm1)\times 10^5$ $M_{\odot}$ and mass outflow rates of $20\pm5$ $M_{\odot}$/yr. The two distinct outflows could be associated with the two nuclei, or be generated by the interaction of the radio jet with the ISM. We also analyse the peculiar profile of the [Fe VII]6087A optical coronal line from an archival VLT/X-shooter spectrum. The comparison with the [NeV]3425A and [FeII]1.257um profiles indicates that [Fe VII] emission likely arises only from the outflow. The absence of the systemic component in [Fe VII] - unlike in [NeV], which has similar ionisation potential and critical density - suggests suppression of [Fe VII] due to iron depletion onto dust grains, while its detection in the outflow implies a lower dust content than in the host ISM. The additional information gained from the ERIS data are consistent with the scenario of a dual AGN, however further observations are required to confirm its nature.

We present an X-ray analysis of the merging galaxy cluster system PSZ2 G279.79+39.09 ($z=0.29$) using archival XMM-Newton and Chandra observations. The surface brightness image is bimodal, elongated east-west with a projected core separation of $\sim 1.35$ Mpc. We measure gas temperatures of 5.36 keV for the eastern subcluster (PSZ-E) and 5.44 keV for the western component (PSZ-W). Assuming isothermal intracluster gas, the hydrostatic masses are $\log(M_{500}/M_\odot)=14.76$ for PSZ-E and 14.54 for PSZ-W, implying a mass ratio of $\sim 1:1.7$. PSZ-E shows X-ray concentration indices of $c_{40}/c_{400}=0.124$ and $c_{100}/c_{500}=0.278$, together with a centroid shift of $w=0.016$, indicating a disturbed halo that still hosts a compact cool core; PSZ-W is comparably disturbed even in its core. Both subclusters exhibit ICM asymmetries consistent with ram-pressure stripping, and PSZ-W displays an X-ray tail extending nearly to the outskirts of PSZ-E. The orientation and length of this tail support an off-axis merger geometry. Thermodynamic maps reveal a hot ($\sim 7.3$ keV), high-pressure, high-entropy bridge between the cores. From the Rankine-Hugoniot temperature jump, we infer a Mach number $M=1.41^{+0.33}_{-0.30}$, consistent with a weak merger shock propagating at $1620^{+500}_{-420}$ km s$^{-1}$. These results indicate a merger with a non-zero impact parameter, likely observed near core passage ($\lesssim 0.5$ Gyr before or after), with the pre-pericenter scenario slightly preferred based on the projected separation and thermodynamic structure.

We present an updated gravitational lens model of the Bullet cluster (1E 0657-56) by combining JWST NIRCam imaging and NIRSpec spectroscopy. Although previous lens models relied on many multiply imaged galaxies, only six systems had spectroscopic redshifts prior to this work. Our lens model is constrained by a catalogue of 135 secure multiple images from 27 background galaxies with spectroscopic redshifts, uniformly covering both subclusters and a wide redshift range of 0.9 - 6.7. We also provide a catalogue of 199 multiple image candidates. We model the cluster with Lenstool and incorporate several large-scale haloes, cluster members, the intracluster gas, and group-scale haloes surrounding the cluster core, motivated by spectroscopic studies of cluster member kinematics. We describe the main cluster component with a complex, elongated double-peaked distribution, and the subcluster with a single large-scale halo aligning closely with the brightest cluster galaxy ($4_{-2}^{+4}$ kpc). The uncertainty of the alignment is improved threefold with the addition of JWST systems. The addition of group-scale substructures, roughly following the two axes of cluster assembly, improves the fit to the multiple image positions and provides a physically motivated alternative to constant shear. Our lens model shows the closest agreement with previous studies in aperture mass profiles at $\sim60$ kpc from the BCGs, but exhibits significant differences in the detailed mass distribution as a result of different lens-modelling strategies and adopted constraints. The differences are reflected in small but spatially coherent deviations between the new spectroscopic redshifts and redshifts predicted by earlier lens models.

We present a new determination of the primordial helium abundance based on new, high-quality Large Binocular Telescope (LBT) observations of 54 metal-poor H II regions. These regions have been observed and analyzed uniformly. We also describe a number of updates to our methodology, including updated helium emissivities. Enabled by the large, high-quality dataset, we examine our sample targets for potential systematic errors, which could bias their results. We perform a standard 95% confidence level $χ^2$ cut and find that a significantly larger fraction (47/54 = 87%) of our sample qualifies than for previous datasets. We also screen for quality and reliability, flagging targets which may introduce significant systematic errors, producing a dataset of 41 targets. In a significant breakthrough for the field, that dataset includes 15 high SNR targets with low metallicity (O/H < 4 $\times$ 10$^{-5}$). Due to this low-metallicity dataset, for the first time, a weighted average for determining the primordial helium abundance (Y$_\mathrm{p}$) is well-justified and produces a robust result. By weighted average of our 15 low-metallicity targets, we determine Y$_\mathrm{p}$ = 0.2458 $\pm$ 0.0013. This result achieves an unprecedented precision of 0.5%, and it is in good agreement with the BBN result, Y$_\mathrm{p}$ = 0.2467 $\pm$ 0.0002, based on the Planck determination of the baryon density.

Accurately determining the elemental abundances of a low metallicity nebula strongly depends on measuring the density (n$_e$) and temperature (T$_e$) of the gas. Because these two parameters are inherently degenerate when derived solely from H and He recombination lines, we rely on the density-sensitive HeI $λ$10830 line to assist in resolving this issue, especially for accurate He abundances. To facilitate this, we present near-IR (NIR) LUCI spectra of 48 low-metallicity targets from the Large Binocular Telescope (LBT) and homogeneously reduce them using Pypeit as part of the LBT $Y_{\rm p}$ Project. IR spectra require special care, and we wavelength calibrate by-hand using the bright OH emission lines, carefully apply proper telluric corrections, and co-add the spectra of LUCI1 and LUCI2 on a resampled grid to ensure accurate results. We use a Gaussian profile to fit the emission lines and measure the fluxes relative to Paschen-gamma (P$γ$), resulting in HeI $λ$10830 to P$γ$ ratios consistent with previous studies. As a result, this work significantly expands the available dataset of NIR HeI $λ$10830 fluxes in low metallicity galaxies. These high-quality measurements, where we find a median flux ratio uncertainty of $\widetildeσ = 0.08$, reduce the overall uncertainties in helium abundance estimates for individual targets. The increased size of the high-quality sample enables searching for systematic uncertainties and improves the reliability of the helium abundance determinations used to infer the primordial helium abundance ($Y_{\rm p}$).

Empirically measuring the primordial He mass fraction, $Y_{\rm p}$, requires a significant number of low-metallicity nebulae with direct constraints on He/H and O/H abundances. This technique requires high-fidelity measurements of the gas-phase physical conditions, namely the electron temperature ($T_e$) and density ($n_e$). To this end, we present deep rest-optical spectroscopy for a sample of 62 low-metallicity ($\lesssim$ 20% solar O/H) galaxies acquired using the Multi-Object Double Spectrographs (MODS) on the Large Binocular Telescope (LBT) as part of the LBT $Y_{\rm p}$ Project. We discuss new fitting methods that recover the intensity of up to 61 H and He recombination lines, of which, up to 26 will be used to determine gas-phase He abundances, and we examine the emission line properties of the LBT $Y_{\rm p}$ Project sample. We assess different scaling relations in the low-metallicity interstellar medium (ISM), finding that $n_e$[Ar IV] measured in 31 targets is systematically larger than $n_e$[S II] or $n_e$[O II]. The larger densities are insufficient to significantly bias $T_e$[O III] or the O/H abundance. $T_e$[S III] and $T_e$[O III] are strongly correlated over a range of $\sim$10$^4$ K with very low scatter, and we calibrate new $T_e$[S III]-$T_e$[O III] scaling relations for use in other low-metallicity environments. We examine different $T_e$ measured in the low-ionization gas, finding significant scatter compared to $T_e$[O III]. The precision direct O/H derived in this analysis (median uncertainty $\sim$4%) are consistent with prior literature measurements, albeit with relatively large scatter. These data provide a key component necessary to empirically measure $Y_{\rm p}$ and the abundance patterns of other elements in the ISM.

Extremely low metallicity HII regions have been observed with the goal of determining the primordial helium abundance ($Y_{\rm p}$). $Y_{\rm p}$, combined with standard big bang nucleosynthesis and the half-life of the neutron, provides a direct measurement of the number of neutrino families, but $Y_{\rm p}$ must be measured very precisely to provide meaningful constraints on physics beyond the Standard Model. Here we describe a program to combine new Large Binocular Telescope (LBT) observations with a new analysis methodology to significantly improve the determination of $Y_{\rm p}$. The LBT, with its MODS and LUCI instruments, produces spectra, which, when combined with our new analysis methodology, are capable of delivering He abundances in individual HII regions with uncertainties of approximately 2% or less. Archival LBT/MODS spectra of standard stars over a four-year period enable the determination of a wavelength-dependent uncertainty in the MODS spectral response, resulting in improved relative emission line uncertainties. An optimized sample of low-metallicity galaxies has been selected with the goal of producing a determination of $Y_{\rm p}$ with a precision of $\sim$ 0.5%, sufficient to provide an independent constraint on the effective number of neutrino families of $\sim$ 3%.

X-ray luminous cool-core (CC) galaxy clusters contain powerful cosmic ray (CR) sources. High-energy CRs powering GHz synchrotron lose energy rapidly, but long-lived (~Gyr-old) populations of 0.1-1 GeV CRs persist, propagating to ~100 kpc distances and radiating via inverse-Compton (IC) scattering of CMB photons. We explore observable consequences of such CR-IC emission. This produces remarkably thermal X-ray spectra, which could contribute significantly to emission in CC centers. These naturally connect to ultra-steep radio sources and radio mini-halos at younger ages, but become undetectable in most radio, hard-X-ray, and $γ$-ray searches (though future imaging may detect them), while reproducing apparent density, temperature, entropy, and mass deposition rates of CCs. This would provide an alternative resolution of the cooling flow problem: clusters may appear as strong CCs because of strong CR-IC, while not actually cooling so rapidly. This predicts many observed correlations between AGN/jet properties, radio galaxy and minihalo properties, cooling radii, cavity radii and apparent X-ray cooling luminosity $L_{\rm X,cool}$. Since $L_{\rm X,cool}$ is actually from CR-IC, the observed radio-X-ray ($L_{\rm radio}-L_{\rm X,cool}$), apparent cavity power ($P_{\rm cav}-L_{\rm X,cool}-L_{\rm radio}$), and strong CC-AGN correlations are predicted without free parameters. Since CR-IC leads to X-ray overestimates of thermal pressure, the ratio of SZ to X-ray pressures should drop in CC centers. CR-IC also suppresses abundances inferred from X-ray relative to optical/UV measurements in CC centers. Both of these appear to be seen in sufficiently-resolved CCs. Effects on cluster cosmology, hydrostatic mass estimation, and non-thermal pressure/turbulence estimators are small. Redshift evolution of CC surface brightness profiles could provide strong constraints or imply CR-IC at high-$z$.

We present an analysis of spectral energy distributions (SEDs), galaxy light profiles, and visual morphological classifications for 194 X-ray luminous AGN (intrinsic absorption-corrected log10 LX(0.5 to 7 keV) less than 42.5, with a maximum of 45.2 ergs per second) at redshift z less than 1.5 in the GOODS fields. We generate X-ray to far-infrared SEDs normalized at 1 micron for all AGN and sort them according to their emission slopes in the ultraviolet and infrared. We visually classify their host galaxy morphologies and compute their bulge-to-total light ratios using the software Galaxy Shapes of Light (galight). Most (94 percent) GOODS AGN exhibit obscured SEDs, defined by diminished ultraviolet and/or mid-infrared emission, while only 6 percent show unobscured, quasar-like SEDs. Secular processes appear to play a large role in stimulating AGN emission, as only around one-third of galaxies are undergoing interactions. We also describe the morphological identification of a population of suspected post-merger spheroid galaxies with obscured ultraviolet and infrared SEDs, and distinguish them from the host galaxies of AGN with less obscuration in the ultraviolet or infrared.

Polar ring galaxies (PRGs) host an outer ring of gas and stars oriented nearly perpendicular to the main stellar body. They represent extreme examples of misaligned systems and provide valuable insight into galaxy interactions, gas accretion, and peculiar gas dynamics. We compile a complete sample of kinematically confirmed PRGs and collect their H I measurements. Combining literature data with new observations from FAST, we detect H I emission in 22 sources, identify one potential H I absorption feature, and find four non-detections among 40 confirmed PRGs. Compared to galaxies in the ALFALFA and xGASS surveys, PRGs predominantly occupy the green valley or quenched regimes but exhibit higher gas fractions than typical early-type galaxies, suggesting gas accretion. The H I profile asymmetry and shape for PRGs are not consistent with that of the ALFALFA sample with p<0.05. We examine their Tully-Fisher relation (TFR) and baryonic TFR (bTFR), linking the systems' rotation velocities to their masses. The extreme outliers in TFRs for the control sample tend to display single-peaked H I profiles. PRGs do not follow a tight TFR or bTFR if the H I resides primarily in the host galaxy. But the scatter decreases significantly if we assume the gas is mainly distributed in the polar ring. Spatially resolved H I observations are essential to disentangle the gas distribution and kinematics in PRGs, which are key to understanding their formation mechanisms.

The orientation of radio jets relative to their host galaxies offers an interesting avenue for probing the connection between active galactic nuclei (AGN) and their surroundings. Several studies have also investigated the orientation of nuclear dust features. We follow up on this previous work with newer Hubble Space Telescope imaging of early-type radio galaxies, and a largely automated process for measuring position angles. We classify the dust features as lanes, disks, or rings. Lanes are irregular structures that likely form from gas-rich minor mergers, while disks and rings are more well-defined and may form from settling lanes or internal mechanisms. We find that dust lanes do not have a preferred alignment relative to their host galaxies, but are preferentially perpendicular to the jets. In contrast, dust disks and rings tend to be closely aligned with the major axes of their host galaxies, but have varying orientations relative to the jets. Our results suggest that infalling dusty material from mergers can influence the angle of the radio jet. This would allow the jet orientation to change over time, and may help explain the role of AGN feedback in maintaining quiescence in massive galaxies.

$λ$ Boötis stars are a subset of chemically peculiar A-stars that display Solar abundances in lighter elements (C, N, O, S, etc.) but a deficiency in Iron-peak elements. This difference has been attributed to the A-stars accreting pristine (metal deficient) gas from the Interstellar Medium. However, the recent discovery of $λ$ Boötis stars in clusters challenges this theory, due to the presence of ionising radiation from intermediate/massive ($>$5 M$_\odot$) stars, which could prevent accretion of pristine ISM gas. We use $N$-body simulations to track the dynamical histories of A-stars during the evolution of a star cluster. We find that some stars leave the confines of the cluster and travel beyond the tidal radius, where they may be able to accrete pristine ISM gas. These A-stars then sometimes move back into the inner regions of the cluster, but the photoionising radiation flux they receive is not high enough to prevent $λ$ Boötis abundances from occurring in these A-stars. We find that A-stars can develop $λ$ Boötis abundances and subsequently form a wide ($>100$ au) binary system, meaning that observations of binary systems that have different abundances between the component stars would not rule out the ISM accretion scenario. Whilst we have shown that $λ$ Boötis stars can reside in and around star clusters, further research is required to assess the validity of the accretion rates required to explain their abundance patterns.

The population of Active Galactic Nuclei (AGN) newly discovered by the James Webb Space Telescope (JWST) exhibits peculiar properties that distinguish it from both local type I AGN and high-redshift quasars. Most of these sources are compact, appearing as 'little dots': among them, the sub-class (10-30% of the total) characterized by significantly red optical colors has been named 'Little Red Dots' (LRDs), while here we analogously introduce the term 'Little Blue Dots' (LBDs) for the remaining, bluer sources (70-90%). We then present a comparative analysis of the prototypical representatives ('Rosetta Stones') of the two classes: GN-28074 at z=2.26, the Red Rosetta Stone, and GS-3073 at z=5.55, the Blue Rosetta Stone. In both Rosetta Stones the broad Balmer lines are better described by exponential profiles rather than single Gaussians, similarly to normal low-redshift type I AGN, indicating that exponential profiles are not unique to LRDs. They are both extremely X-ray weak, show strong auroral [OIII] 4363 emission, weak hot dust mid-IR emission, and no time variability. However, they differ in terms of excitation diagnostics: the HeII 4686 line is undetected in the Red Rosetta but strongly detected in the Blue Rosetta in both narrow and broad components, with the latter much broader than hydrogen Balmer lines. This supports BLR stratification and disfavors the cocoon electron-scattering scenario. An additional difference is the presence of prominent Balmer absorption in the Red Rosetta -- indicative of extremely dense gas along the line of sight -- but absent in the Blue Rosetta. Taken together, these results suggest that LRDs and LBDs share the same central engine as standard type I AGN, while differing in the amount and geometry of dense gas surrounding the accretion disk, and/or in their accretion properties.

The James Webb Space Telescope has unveiled an abundant population of potential active galactic nuclei (AGN) at high redshift ($z\gtrsim4$) known as little red dots (LRDs), which are likely hosted in relatively low-mass galaxies. However, previous theoretical models have highlighted the difficulty in continuously feeding massive black holes in the central regions of bursty, high-redshift galaxies because of repeated gas evacuation by stellar feedback. We analyze galaxies in high-redshift FIRE-2 simulations to understand whether they are capable of hosting the observed abundant population of high-redshift AGN. We use a gravitational torque-driven accretion (GTDA) model and a simple free-fall accretion model to derive black hole accretion rates and construct predicted AGN bolometric luminosity functions for $z=5-7$. The GTDA model and the free-fall model with black holes accreting $\lesssim 1$ percent of their central gas supply ($<100 \rm \ pc$) per free-fall time predict AGN abundances that are more than sufficient to explain the most recent LRD observations. The fiducial models, in fact, overpredict the number of low-luminosity AGN as compared with observations. We explore possible resolutions of this tension. A plausible, though likely not unique, scenario for alleviating the AGN overpredictions and which also provides a good match to the host-galaxy UV luminosity distribution suggests that LRDs are super Eddington-accreting, Eddington luminosity-limited, $M_{\rm BH}\gtrsim 2\times10^5 \ \rm M_\odot$ black holes residing in $M_\star \gtrsim 2\times10^7 \ \rm M_\odot$ galaxies.

Nuclear star clusters (NSCs) are among the densest stellar systems in the Universe and often coexist with supermassive black holes (SMBHs) at galaxy centres. While SMBH formation histories are essentially lost, NSCs preserve evolutionary imprints through their stellar populations and stellar kinematics, reflecting the cumulative effects of mergers, accretion, and internal dynamical evolution. We aim to investigate the orbital structure of the unusually large NSC in FCC 47 (NGC 1336) by decomposing its stellar orbits into dynamically distinct components. We extract stellar kinematics, and in particular the line-of-sight velocity distributions (LOSVDs), from VLT/MUSE integral-field spectroscopy using the non-parametric Bayes-LOSVD approach, and apply triaxial Schwarzschild orbit-superposition modelling with the DYNAMITE software. We decompose the orbit library into hot, warm, cold, and counter-rotating components. We detect triple-peaked LOSVDs in the nucleus, indicating a complex orbital structure. The NSC forms a counter-rotating, kinematically decoupled component. A hot pressure-supported component, a warm counter-rotating structure and a counter-rotating cold disk in the centre suggest hierarchical assembly via early star cluster accretion and later in situ star formation. Our orbital decomposition of FCC 47 supports a hybrid formation scenario for this NSC. Dynamically distinct substructures reflect the interplay of accretion and in situ star formation during galaxy evolution.

We present a statistical analysis of the extreme brightness of the fast radio burst FRB 20250316A, a luminous, apparently non-repeating event detected by CHIME/FRB. Employing a model-agnostic framework based on the Generalized Extreme Value (GEV) distribution applied to the second CHIME/FRB catalog, we quantify its rarity within the current population. Bayesian fitting of GEV models to block maxima of peak flux and fluence data reveals FRB 20250316A to be a profound statistical outlier. For the peak flux, the analysis yields a return period of $\sim$ 600 years ($1σ$ credible level), with the underlying distribution being of the heavy-tailed, unbounded Fréchet type ($ξ> 0$). The fluence analysis indicates greater complexity: while the full sample suggests a Fréchet-type distribution with a $\sim50$-year return period in $1σ$ credible level, the removal of three other notable outliers points toward a light-tailed Weibull-type distribution ($ξ< 0$) with a finite upper bound far exceeded by FRB 20250316A. This dichotomy highlights the challenge in characterizing the tail of the FRB luminosity function with limited data. Although less extreme in recurrence time than the ``Brightest Of All Time'' gamma-ray burst GRB 221009A, FRB 20250316A constitutes a similarly exceptional event (a potential FRB ``BOAT'') within the short observational history of wide-field radio surveys. Our results underscore the existence of rare, highly luminous events that may probe the upper limits or distinct sub-populations of the FRB luminosity distribution.

One of the major discoveries by the James Webb Space Telescope (JWST) is the identification of a large population of luminous galaxies at $z>10$, challenging theoretical models for early galaxy formation. The unexpectedly high number density of these systems has triggered intense debate about potential differences in the physical properties of galaxies at such extreme redshifts and those at lower redshift. However, progress has been limited by the lack of rest-frame optical diagnostics, which are critical for constraining the key properties. Here we present deep JWST/MIRI observations of a UV-luminous galaxy at $z=11.04$, CEERS2-588, only 400 Myr after the Big Bang. CEERS2-588 is detected in the MIRI F560W and F770W bands, while deep MIRI/MRS spectroscopy yields no detection of H$α$ or [OIII]$\lambda5007$ line, revealing a prominent Balmer break detected for the first time at $z>10$. Spectral energy distribution (SED) fitting indicates an extended star formation history possibly reaching $z>15$, followed by rapid quenching within the recent $\sim10$ Myr, in stark contrast to other $z>10$ galaxies. The MIRI detections also significantly improve our stellar mass estimate to $\mathrm{log}(M_*/M_\odot)=9.1^{+0.1}_{-0.1}$, making CEERS2-588 the most massive galaxy securely confirmed at $z>10$. Remarkably, the inferred gas-phase metallicity is near solar, exceeding predictions from current theoretical models. These results suggest that efficient starbursts play a key role in producing the abundant luminous galaxy population in the early universe.

Particle-in-cell (PIC) numerical simulations are currently among the most advanced tools to investigate particle acceleration at relativistic shocks. Still, they come with limitations imposed by finite computing power, whose impact is not straightforward to evaluate a priori. Observational features are hence required as verification. energy electrons accelerated at external shocks, provides a testbed for such predictions. Current numerical studies suggest that in GRB afterglows the maximum synchrotron photon energy, which corresponds to the limit of electron acceleration, may fall within the $\sim$ 0.1--10 keV X-ray energy band at late times, $t\gtrsim 10^6 - 10^7$ s. To test this prediction, we analyzed the X-ray spectra of six GRBs with \emph{Swift}/XRT detections beyond $10^7$ s: our analysis reveals no clear evidence of a spectral cutoff. Using a model that accounts for the effect of the finite opening angle of the shock on the observed maximum synchrotron photon energy, we show that these observations are incompatible with PIC simulation predictions, unless one or more physical afterglow parameters attain values at odds with those typically inferred from afterglow modeling (small radiative efficiency, low ambient density, large equipartition fraction $ε_{\rm B}$ of the magnetic field). These findings challenge existing numerical simulation results and imply a more efficient acceleration of electrons to high-energies than seen in PIC simulations, with important implications for our understanding of particle acceleration in relativistic shocks.

AGN variability offers a direct probe of accretion physics, disk structure, and black hole growth, but progress has been limited by sample size, cadence heterogeneity, and photometric systematics. The Vera C. Rubin Observatory Legacy Survey of Space and Time (LSST) will deliver multi-band light curves for millions of AGN, enabling variability studies at a true population scale. We synthesize recent results from the Zwicky Transient Facility (ZTF), which demonstrate that optical variability amplitudes and timescales are primarily regulated by accretion state, with secondary dependence on black hole mass and redshift, and establish the feasibility of survey-driven continuum reverberation mapping. ZTF measurements reveal optical continuum-emitting region sizes that often exceed standard thin disk predictions, implicating diffuse continuum emission from the broad line region as a significant contributor to observed inter-band lags. We evaluate the implications of LSST cadence and survey strategy, particularly the deep drilling fields, for continuum and emission line reverberation mapping, changing-look AGN, extreme variability quasars, and periodic variability searches. Key limitations of broadband photometric variability are identified, including variable emission line contamination, diffuse BLR continuum emission, and cadence-dependent lag recoverability. We argue that realizing LSST's full scientific potential requires community-scale, standardized variability metric pipelines, probabilistic classification integrated with alert brokers for follow-up triggering, and complementary medium-band photometric observations to isolate the accretion disk continuum. Together, these elements will enable LSST to convert photometric variability into quantitative constraints on accretion disks, BLR structure, and supermassive black hole growth across cosmic time.

We determine the galaxy luminosity function of cluster galaxies in the nearby galaxy cluster Abell 2199 (A2199), focusing on the faint-end slope down to $M_r \sim -14.5$. To achieve this, we augment the existing dataset by adding redshift data from our deep MMT/Hectospec survey and from the Dark Energy Spectroscopic Instrument (DESI), significantly improving the spectroscopic completeness down to $r_{\mathrm{petro},0} = 20.8$ within the central $30^\prime$ region. The resulting luminosity function is well described by a Schechter function with a characteristic magnitude $M^* = -21.30 \pm 0.27$ and a faint-end slope $α= -1.23 \pm 0.05$. This faint-end slope is consistent with those measured in the nearby Coma and Virgo clusters and in a cluster from the TNG50 cosmological simulation, and is slightly shallower than that of field galaxies. These findings indicate that the previously claimed steep faint-end upturn (with $α\sim -2$) in nearby galaxy clusters is not supported. Instead, they indicate that environmental processes in dense cluster cores does not seem to trigger the formation or survival of low-mass galaxies, thereby preventing a steep faint-end upturn in the luminosity function.

We report the results of a pilot study that searched for dual active galactic nuclei (AGN) in local ($z<$0.25) galaxies hosting double-peaked narrow emission lines in their optical spectra. We present high-resolution $L-$band (1.5 GHz or 18 cm) continuum images from the Very Long Baseline Array (VLBA) as well as WFC3/IR F160W images from the Hubble Space Telescope of two candidate dual AGN systems: J0948+6848 and J1223+5409. In both targets, we detected compact non-thermal radio emission that is approximately co-spatial with the near-infrared AGN. Both systems host two high brightness temperature ($>10^{8}$ K) radio sources that indicate the presence of either a parsec-scale-separation dual AGN ($d_{\text{sep}} \sim 90$ pc and $\sim 56$ pc, respectively) or a radio jet. Matched-resolution multi-band radio observations are necessary to further characterize the AGN activity in these systems.

Knowledge of the actual density distribution of matter in the local universe is needed for a variety of purposes -- for instance, as a baseline model for ultrahigh energy cosmic ray sources in the continuum limit and for predicting the diffuse Dark Matter annihilation signal. Determining the local mass density and velocity distribution is the aim of the CosmicFlows project. An alternate approach is based on catalogs of galaxies, supplemented with some scheme for filling in for unseen galaxies. Here, we compare the density field proposed by Biteau (2021) with the quasi-linear density field of CosmicFlows2 (Hoffman et al. 2018) and the mean posterior field of CosmicFlows4 (Valade 2026). We find factor-two level differences in some regions and even greater in regions toward the Galactic center zone of avoidance (ZoA) (|l| < 30°, |b| < 20°) as filled by Biteau using "cloning". Within 11 Mpc the density field is well-determined by the Local Volume catalog (Karachentsev et al. 2018) which Biteau directly incorporates; at larger distances, Biteau (2021) should not be used in the ZoA where "galaxies" are entirely fictitious but otherwise is to be preferred over CosmicFlows emphasizing the direction and integrated mass of structures; the radial distribution of mass in Biteau (2021) is less robust due to line-of-sight peculiar velocities. The angular positions of structures in CosmicFlows are sometimes not congruent with evidence in the galaxy catalog.