论文广场 - AcademicHub

01.

arXiv (quant-ph) 2026-06-11 DOI: arXiv:2606.11608

Mach's principle in atomic transitions

作者:

arXiv:2606.11608v1 Announce Type: new Abstract: We investigate the atomic transition probabilities in atom-mirror set-ups that are in circular motion. In one scenario, the atom is in circular motion inside a static cylindrical mirror. In the other scenario, the cylindrical mirror rotates around its central axis while the atom remains static. We report structural similarity in the atomic transition probabilities between these two cases – these probabilities are equivalent upon interchanging the field frequencies between the two scenarios. We interpret such an observation as a semi-classical phenomenon analogous to the classical Mach's principle.

阅读与讨论 → 访问原文 →

02.

medRxiv (Medicine) 2026-06-17 DOI: HASH:ae3169d3e881fc5dae138b195cc0fc70

Waning protection of long-acting RSV monoclonal antibodies in infants: a Bayesian analysis of clesrovimab and nirsevimab trial data

作者:

Gong ↗Flasche ↗Hodgson ↗

Clesrovimab and nirsevimab are long-acting monoclonal antibodies used to prevent respiratory syncytial virus (RSV) disease in infants, but waning protection in the first year of life is incompletely characterised. We applied a published Bayesian inference framework to clesrovimab and pooled nirsevimab trial data to estimate time-varying efficacy against medically attended RSV lower respiratory tract infection (LRTI) and RSV-associated hospitalisation, accounting for differences in placebo-arm event timing between trials. Estimated clesrovimab efficacy declined from 60.7% (95% CrI: 46.3-72.6) shortly after dosing to 38.3% (8.6-52.9) at six months against medically attended RSV LRTI, and from 87.1% (71.2-96.2) to 49.6% (10.4-70.7) against RSV-associated hospitalisation. For nirsevimab, corresponding estimates declined from 86.9% (75.4-95.0) to 53.8% (27.4-69.7) against LRTI, and from 77.5% (52.6-91.8) to 49.7% (15.7-68.3) against hospitalisation. After accounting for differences in RSV exposure timing and LRTI endpoint definitions between trials, we found no evidence of a difference in efficacy or waning between clesrovimab and nirsevimab.

阅读与讨论 → 访问原文 →

03.

arXiv (CS.CL) 2026-06-11 DOI: arXiv:2606.12250

Reassessing High-Performing LLMs on Polish Medical Exams: True Competence or Bias-Driven Performance?

作者:

Large language models (LLMs) in medicine are mainly evaluated using multiple-choice question answering (MCQA), which can overestimate real clinical ability due to guessing strategies and answer biases. To address these limitations, we introduce an expanded and more challenging benchmark based on Polish medical exams, adding over 15,000 questions, two new domains, and four structural modifications that reduce MCQA-specific artifacts and better test reasoning. We evaluate 21 LLMs and show that evaluation design strongly affects results. Under our harder setup, the best model (Qwen3.5-122B) drops by 28.4 and 31 pp on English and Polish exams, respectively. Despite low evidence of data contamination, standard MCQA scores do not reliably reflect true medical competence. To facilitate further research, we make our benchmark publicly available.

阅读与讨论 → 访问原文 →

04.

arXiv (CS.CV) 2026-06-18 DOI: arXiv:2606.18960

Mem-World: Memory-Augmented Action-Conditioned World Models for Persistent Robot Manipulation

作者:

Zirui Zheng ↗Jiaqian Yu ↗Xiongfeng Peng ↗jun shi ↗Mingyi Li ↗Chao Zhang ↗Weiming Li ↗Dong Wang ↗Huchuan Lu ↗Xu Jia ↗

Action-conditioned world models have emerged as a promising paradigm for robot learning, offering a scalable alternative to costly real-world experimentation by generating action-consistent video rollouts. However, persistent world modeling remains challenging in manipulation: frequent end-effector occlusions and rapid wrist-camera motion make the current observation insufficient for predicting future views, causing models to forget or hallucinate scene details seen in earlier frames. Existing memory retrieval strategies often fail to identify informative history in dynamic manipulation scenarios. To address this limitation, we propose Mem-World, a memory-augmented multi-view action-conditioned world model. At its core, we present W-VMem, a 4D wrist-view-centered surfel-indexed memory that anchors historical observations to temporally evolving surface elements. By explicitly modeling when and where scene elements are observed, W-VMem enables geometry-aware retrieval of relevant history frames conditioned on future actions. During generation, relevant history frames are selected via surfel-based rendering and scoring, providing informative and non-redundant context for prediction. Extensive experiments show that Mem-World generates persistent rollouts in complex manipulation scenarios, enables more reliable policy evaluation than Ctrl-World, improving the Pearson correlation with real-world performance by 14.5\%, and supports effective policy improvement through synthetic data generation, increasing success rates from 58\% to 72\% on long-horizon tasks.

阅读与讨论 → 访问原文 →

05.

arXiv (CS.CV) 2026-06-16 DOI: arXiv:2606.15880

Deep Residual Injection for Full-Spectrum Forensic Signal Perception in Multimodal Large Language Models

作者:

Kaiqing Lin ↗Zhiyuan Yan ↗Ruoxin Chen ↗Ke-Yue Zhang ↗Yue Zhou ↗Caiyong Piao ↗Bin Li ↗Taiping Yao ↗Bo Wang ↗Youchang Xiao ↗Shouhong Ding ↗

Multimodal large language models (MLLMs) have been increasingly adopted in forensics for their robust semantic understanding. As AI-generated images become realistic, semantic-level inconsistencies alone are often insufficient for reliable detection. This motivates a critical question: whether MLLMs can achieve full-spectrum forensic signal perception, i.e., capturing low-level generator artifacts without sacrificing pre-trained semantic knowledge. We further perform a layer-wise analysis of forensic signal perception in MLLMs, showing that semantic information is primarily formed in the early-to-middle layers, whereas direct fine-tuning for artifact learning disrupts these semantic representations. Based on this insight, we propose Deep Visual Residual MLLM (Deep-VRM) to preserve early semantic processing while injecting artifact-specific visual signals as a residual path into an intermediate layer, where they are fused with semantic token representations and propagated through subsequent trainable layers. This enables later layers to jointly model semantic reasoning and signal-level forensic cues, and surprisingly, the model learns to adaptively leverage different levels of forensic signals depending on the input, achieving robust and generalizable detection performance. Extensive experiments show that our method achieves state-of-the-art across most benchmarks. The code and data are available at https://github.com/KQL11/Deep-VRM.

阅读与讨论 → 访问原文 →

06.

arXiv (CS.CV) 2026-06-15 DOI: arXiv:2606.13929

Self-Evolving Visual Questioner

作者:

Yijun Liang ↗Hengguang Zhou ↗Ming Li ↗Lichen Li ↗Cho-Jui Hsieh ↗Tianyi Zhou ↗

Vision-language models (VLMs) are typically trained as passive answerers, while their ability to actively ask diverse, non-trivial, visual-centric and grounded questions remains underexplored. Existing visual questioners' performance is bottlenecked by the availability of high-quality training data or the cost of curating them. We show that a VLM can continuously improve itself as a visual questioner without any external supervision. We propose a self-evolving framework that uses a VLM itself as both a proposer and a filter to produce harder, more informative, and visual-centric questions, while maintaining their exploration diversity to avoid training collapse. These questions are then used to train the VLM in both questioner and answerer modes. To evaluate the questioner, we introduce an agentic protocol that assesses questions along perception, reasoning, and diversity dimensions. Experiments across various backbone VLMs show that our method substantially enhances the quality and substantially expands the difficulty boundary of autonomous question generation. Under the same budget, our self-supervision is more effective than training on the static source data. Moreover, the self-evolving questioner remains a competitive or even better answerer.

阅读与讨论 → 访问原文 →

07.

arXiv (CS.LG) 2026-06-17 DOI: arXiv:2511.01352

MiniFool – Physics-Constraint-Aware Minimizer-Based Adversarial Attacks in Deep Neural Networks

作者:

arXiv:2511.01352v2 Announce Type: replace Abstract: In this paper, we present a new algorithm, MiniFool, that implements physics-inspired adversarial attacks for testing neural network-based classification tasks in particle and astroparticle physics. While we initially developed the algorithm for the search for astrophysical tau neutrinos with the IceCube Neutrino Observatory, we apply it to further data from other science domains, thus demonstrating its general applicability. Here, we apply the algorithm to the well-known MNIST data set and furthermore, to Open Data data from the CMS experiment at the Large Hadron Collider. The algorithm is based on minimizing a cost function that combines a $\chi^2$ based test-statistic with the deviation from the desired target score. The test statistic quantifies the probability of the perturbations applied to the data based on the experimental uncertainties. For our studied use cases, we find that the likelihood of a flipped classification differs for both the initially correctly and incorrectly classified events. When testing changes of the classifications as a function of an attack parameter that scales the experimental uncertainties, the robustness of the network decision can be quantified. Furthermore, this allows testing the robustness of the classification of unlabeled experimental data.

阅读与讨论 → 访问原文 →

08.

arXiv (CS.CL) 2026-06-17 DOI: arXiv:2511.01650

EngTrace: A Symbolic Benchmark for Verifiable Process Supervision of Engineering Reasoning

作者:

Ayesha Gull ↗Muhammad Usman Safder ↗Rania Elbadry ↗Fan Zhang ↗Veselin Stoyanov ↗Preslav Nakov ↗Zhuohan Xie ↗

Large Language Models (LLMs) are increasingly entering specialized, safety-critical engineering workflows governed by strict quantitative standards and immutable physical laws, making rigorous evaluation of their reasoning capabilities imperative. However, existing benchmarks such as MMLU, MATH, and HumanEval assess isolated cognitive skills, failing to capture the physically grounded reasoning central to engineering, where scientific principles, quantitative modeling, and practical constraints must converge. To enable verifiable process supervision in engineering, we introduce EngTrace, a symbolic benchmark built on 90 parameterized templates, each generating unique, contamination-resistant problem instances, spanning three major engineering branches, nine core domains, and 20 distinct areas, yielding 1,350 test cases that stress-test generalization across diverse physical scenarios. Moving beyond outcome matching, we introduce a verifiable two-stage evaluation framework that uses a tiered protocol to validate intermediate reasoning traces alongside final answers through automated procedural checks and a heterogeneous AI Tribunal. Our evaluation of 27 leading LLMs reveals a distinct trade-off between numeric precision and trace fidelity, identifying a complexity cliff where abstract mathematical pre-training fails to translate into the integrative reasoning required for advanced engineering tasks.

阅读与讨论 → 访问原文 →

09.

medRxiv (Medicine) 2026-06-22 DOI: HASH:51c9bb8db4069c1d2487eeaa0d9e0ca2

AI-driven Multimodal Representation Learning for Latent Mediation Structure Discovery of Socioeconomic Disadvantage, Psychosocial Factors, and Cardiometabolic Multimorbidity

作者:

Ma ↗Cao ↗

Social disadvantage is associated with multimorbidity, but the pathways linking social conditions to disease burden remain poorly understood. We developed an AI-driven multimodal mediation framework that integrates socioeconomic, psychosocial, clinical, laboratory, behavioral, and genomic data from the All of Us Research Program. Modality-specific variational autoencoders were used to derive latent representations of each data domain, and mediation analyses were subsequently performed in latent space to evaluate indirect associations between socioeconomic disadvantage, psychosocial factors, and multimorbidity. The final analytic cohort included 20,804 participants with complete multimodal data. Across 800 exposure–mediator–outcome combinations, mediation signals were concentrated within a small number of latent dimensions. The strongest indirect association linked a socioeconomic disadvantage dimension, a psychosocial vulnerability dimension, and a cardiometabolic multimorbidity dimension (NIE = 0.002517). The psychosocial dimension was characterized by poorer mental health, greater loneliness, lower social well-being, and lower health literacy, whereas the outcome dimension was associated with hypertension, diabetes, hyperlipidemia, obesity, chronic kidney disease, and heart disease. Bootstrap analyses supported the stability of the leading pathway. These findings suggest that psychosocial vulnerability may contribute to the association between socioeconomic disadvantage and cardiometabolic multimorbidity. More broadly, the proposed framework illustrates how AI-based representation learning can be used to investigate complex relationships across high-dimensional multimodal health data.

阅读与讨论 → 访问原文 →

10.

arXiv (CS.AI) 2026-06-11 DOI: arXiv:2606.11769

When Do Data-Driven Systems Exhibit the Capability to Infer?

作者:

Maximilian Poretschkin ↗Tabea Naeven ↗

arXiv:2606.11769v1 Announce Type: new Abstract: The European AI Act is the first comprehensive regulation of artificial intelligence (AI), setting out extensive obligations, particularly for so-called high-risk and general-purpose AI systems. A key distinguishing feature of AI systems under the AI Act is the capability to infer. Since the AI Act does not clearly define what inference is, there is a gray area for certain data-driven systems. A specific example is credit scoring systems, which are listed by Annex III of the AI Act. At the same time, however, these are often implemented using statistical models for which it is unclear whether they have the capability to infer and thus fall under the AI definition of the AI Act at all. Motivated by statistical learning theory, this work develops a framework for grading different levels of the capability to infer. Based on the AI Act and the Commission Guidelines on the definition of an artificial intelligence system, we analyze which levels constitute sufficient capability to infer within the meaning of the AI Act and where further regulatory clarity is needed. We illustrate the framework by creating two realistic credit scoring workflows and show whether and where inference occurs in them. Our analysis illustrates that not only individual models but the entire data processing workflow must be considered. It also shows that the involvement of human experts during development can have significant influence on the capability to infer. Code can be found at https://github.com/fraunhofer-iais/inference-framework-creditscorecards.

阅读与讨论 → 访问原文 →

11.

arXiv (quant-ph) 2026-06-17 DOI: arXiv:2606.17587

On the entanglement induced by the deformation of phase-space

作者:

Shilpa Nandi ↗Shatarupa Maity ↗Pinaki Patra ↗

arXiv:2606.17587v1 Announce Type: new Abstract: Most quantum gravity theories propose that the fundamental concept of space-time is mostly compatible with quantum theory in noncommutative (NC) space. In the present paper, we revisit the notion of entanglement induced by NC deformations of phase space. The positive partial transpose (PPT) criterion for separability of bipartite Gaussian states is extended to a general class of Bopp's shift. In particular, we have considered both the position-position and momentum-momentum noncommutativity, with deformation parameters $\theta$ and $\eta$, respectively. It turns out that $\theta$ and $\eta$ induce the entanglement. We have directly applied the formalism for an anisotropic two-dimensional harmonic oscillator. Peres-Horodecki separability condition leads to a constraint equation for the parameter values of the oscillator in NC space. It turns out that the bipartite Gaussian state is almost always entangled in deformed space. To implement the theoretical idea, we provide an outline for a gedankenexperiment to identify the signature of phase-space noncommutativity, i.e., quantum gravity. In particular, the gedankenexperiment is devised to test the separability of supposedly separable Gaussian states in the usual commutative space, through the covariance matrix, which is constructed via measured output photocurrents after interaction of input Gaussian states and reference states. If the experiment shows that the supposedly separable states are actually entangled, then the entanglement is created through the intermediate background noncommutative space, which is a signature of the quantum nature of gravity.

阅读与讨论 → 访问原文 →

12.

arXiv (CS.AI) 2026-06-16 DOI: arXiv:2606.16329

Exploiting Search in Symbolic Numeric Planning with Patterns

作者:

Matteo Cardellini ↗Enrico Giunchiglia ↗

arXiv:2606.16329v1 Announce Type: new Abstract: In this paper, we present a procedure for numeric planning based on Symbolic Pattern Planning (SPP). Given a numeric planning problem $\Pi$, a pattern $\prec$ is a sequence of actions used to define a formula encoding the subsequences of $\prec$ executable from a starting state $S$. Cardellini, Giunchiglia, and Maratea (2024a) follow the Planning as Satisfiability approach by defining, at each step $n \ge 0$, a formula $\Pi^\prec_n$ in which $(i)$ the pattern $\prec$ is computed only for $n=0$ in the initial state $I$ of $\Pi$, and then exploited at each step $n$, $(ii)$ the starting state $S$ is set to $I$, and $(iii)$ the set $G$ of goals is required to hold in the last state that can be reached by one of the subsequences of $\prec$ concatenated $n$ times. The procedure begins with $n=0$, terminates as soon as $\Pi^\prec_n$ is satisfiable, and otherwise proceeds by incrementing $n$. In this paper, possibly at each step, $(i)$ we symbolically search for an intermediate state $P$ reachable from $I$, closer to a goal state, $(ii)$ dynamically recompute the pattern $\prec_h$ – to be used in the next step – in $P$, $(iii)$ refine the pattern $\prec_g$ used to reach $P$, and $(iv)$ start the new search from the state $S$ which can be either the initial state $I$ or the last computed intermediate state $P$, exploiting the computed patterns $\prec_g$ and $\prec_h$ to define the pattern $\prec$ to be used in the search. In particular, at each step, we define a formula $\Pi^{\prec}_{S,P}$ encoding the existence of a state $P'$ closer than $P$ to a goal state, with $P'$ reachable from the starting state $S$ when using the pattern $\prec$. We present different techniques for producing such formulas, each corresponding to a different strategy for exploring the search space. We prove their correctness and completeness, the latter under certain conditions.

阅读与讨论 → 访问原文 →

13.

arXiv (CS.CL) 2026-06-12 DOI: arXiv:2605.17770

Entropy-Gradient Inversion: Moving Toward Internal Mechanism of Large Reasoning Models

作者:

Junyao Yang ↗Chen Qian ↗Kun Wang ↗Linfeng Zhang ↗Quanshi Zhang ↗Yong Liu ↗Dongrui Liu ↗

The advancement of Large Reasoning Models (LRMs) has catalyzed a paradigm shift from reactive ``fast thinking'' text generation to systematic, step-by-step ``slow thinking'' reasoning, unlocking state-of-the-art performance in complex mathematical and logical tasks. However, the field faces the fundamental gap between token-level behavioral analysis and internal reasoning mechanisms, and the instability of reinforcement learning (RL) for reasoning optimization relying on costly external verifiers. We identify and formally define Entropy-Gradient Inversion, a robust negative correlation between token entropy and logit gradients that acts as a definitive geometric fingerprint for LRM reasoning capability. Building on this, we propose Correlation-Regularized Group Policy Optimization (CorR-PO), which embeds this inversion signature into RL reward regularization. Extensive experiments on various reasoning benchmarks across multiple model scales show CorR-PO consistently outperforms state-of-the-art baselines, confirming that stronger inversion directly correlates with superior reasoning performance.

阅读与讨论 → 访问原文 →

14.

medRxiv (Medicine) 2026-06-17 DOI: HASH:430d28f196785eb321c7c965892d44b7

Adverse Childhood Experiences Reorganise the Brain-Personality Network Across the Psychosis Spectrum

作者:

Sarti ↗Cecere ↗Dallenbach ↗H.-L. H ↗Huppi ↗R. M ↗Misra ↗A. R ↗Edkins ↗Omlor ↗Blom ↗J. M. C ↗…

Exposure to adverse childhood experiences is a pervasive risk factor for psychosis, exhibiting a linear relationship across the psychosis spectrum from subclinical schizotypal traits to schizophrenia spectrum disorders. While this association is often conceptualised within the vulnerability-stress framework, the systemic mechanisms through which childhood trauma reconfigures the brain-personality interactome remain poorly understood. We examined clinical, neuropsychological, and neuroimaging data from a sample of low- and high-schizotypy individuals, and patients with a diagnosis of schizophrenia spectrum disorder (N=120). Our aim was to map how trauma reconfigures interactions between neurobiology and schizotypal phenomenology. We adopted a mixed graphical model approach to jointly estimate conditional dependencies between childhood trauma, regional brain morphometry, and schizotypal traits across the psychosis spectrum. Our results show that childhood trauma reconfigures the brain-personality network, shifting it from a state driven by cognitive processes to one anchored in emotional (limbic) reactivity. This transition is marked by the increased influence of impulsive traits and a significant strengthening of connections within the salience network. These changes converge with a reduced thickness of the frontal executive regions, the brain's control centres, identified in our models. Collectively, our results suggest a structural phenomenological decoupling, where trauma conditioned affective circuits may bypass weakened top-down regulatory controls. These findings highlight the necessity of using integrative frameworks to capture how trauma fundamentally reshapes the relationship between the brain and schizotypal personality.

阅读与讨论 → 访问原文 →

15.

arXiv (CS.CV) 2026-06-11 DOI: arXiv:2606.11529

XPR: An Extensible Cross-Platform Point-Based Differentiable Renderer

作者:

Steve Rhyner ↗Sankeerth Durvasula ↗Aleksandr Kovalev ↗Hansel Jia ↗Adrian Zhao ↗Mrutunjayya Mrutunjayya ↗Nilesh Ahuja ↗Selvakumar Panneer ↗Christina Giannoula ↗Nandita Vijaykumar ↗

Point-based differentiable rendering underpins modern 3D reconstruction, novel-view synthesis, and learning-based graphics pipelines, but developing new rendering methods often requires extensive low-level implementation, hardware-specific kernels, and manually written backward passes. This limits rapid prototyping, reproducibility, exploration, and deployment, especially across diverse hardware platforms. This paper presents XPR, an extensible cross-platform framework for point-based differentiable rendering. XPR introduces a high-level programming interface that separates method-specific logic from the shared rendering pipeline, allowing users to implement new methods in a few lines of code. Its pipeline decomposes rendering into modular, statically shaped parallel operations that can be lowered by a cross-platform compiler to GPUs, TPUs, CPUs, and other ML accelerators. We demonstrate implementations of 3DGS, 3DGUT, and LinPrim, with only a few 100s lines of Python code, each of which can be compiled to a range of hardware platforms with the XLA compiler. These results show that XPR enables fast experimentation and portable execution for emerging point-based differentiable rendering systems.

阅读与讨论 → 访问原文 →

16.

arXiv (CS.AI) 2026-06-16 DOI: arXiv:2606.15866

STRIDE: Strategic Trajectory Reasoning via Discriminative Estimation for Verifiable Reinforcement Learning

作者:

Qinjian Zhao ↗Zhihao Dou ↗Dinggen Zhang ↗Xiangyu Li ↗Chaoda Song ↗Zhongwei Wan ↗Xinpeng Li ↗Yanyan Zhang ↗Kaijie Chen ↗Qingtao Pan ↗Chengcheng Feng ↗Zhiqiang Gao ↗…

arXiv:2606.15866v1 Announce Type: new Abstract: Reinforcement Learning with Verifiable Rewards (RLVR) has become an effective post-training paradigm for improving the reasoning abilities of large language models. However, existing RLVR methods typically rely on final-answer correctness to assign trajectory-level rewards, providing sparse supervision and treating all tokens uniformly regardless of their actual contribution to reasoning. Although recent studies introduce intermediate signals such as process rewards, high-entropy tokens, and semantic uncertainty, these signals are often not inherently verifiable and may fail to distinguish beneficial strategic patterns from harmful ones. To address this limitation, we propose STRIDE (Strategic Trajectory Reasoning with Discriminative Estimation), a fine-grained RLVR framework that derives strategic reasoning supervision from verifiable outcomes. STRIDE contrasts successful and failed trajectories within each response group to estimate the outcome-discriminative preference of each $n$-gram strategic pattern, and further combines this signal with reasoning saliency entropy to identify decision-relevant strategic patterns. These patterns are assigned differentiated advantage values during RL optimization, enabling more precise credit assignment while preserving the verifiability of RLVR. Extensive experiments demonstrate that STRIDE consistently improves reasoning performance across diverse models, tasks, and extended settings, including VLMs and agent-based systems.

阅读与讨论 → 访问原文 →

17.

arXiv (quant-ph) 2026-06-17 DOI: arXiv:2412.14280

Many-body spectral transitions through the lens of the variable-range SYK2 model

作者:

Andrea Legramandi ↗Soumik Bandyopadhyay ↗Philipp Hauke ↗

arXiv:2412.14280v2 Announce Type: replace-cross Abstract: The Sachdev-Ye-Kitaev (SYK) model is a cornerstone in the study of quantum chaos and holographic quantum matter. Real-world implementations, however, deviate from the idealized all-to-all connectivity, raising questions about the robustness of its chaotic properties. In this work, we investigate a quadratic SYK model with distance-dependent interactions governed by a power-law decay. By analytically and numerically studying the spectral form factor (SFF), we uncover how transitions present in the single-particle limit carry over to the many-body system. Non-trivial cancellations in the one-loop contributions lead to a robustness of the SFF under a considerable reduction of the interaction range. Further suppression leads to a breakdown of perturbation theory around the infinite-range path-integral saddle and the appearance of new spectral regimes, marked by a higher dip and the emergence of a secondary plateau. Our results highlight the interplay between single-particle criticality and many-body dynamics, offering new insights into the quantum chaos-to-localization transition and its reflection in spectral statistics.

阅读与讨论 → 访问原文 →

18.

arXiv (CS.CL) 2026-06-16 DOI: arXiv:2603.07079

Entropy-Aware On-Policy Distillation of Language Models

作者:

Woogyeol Jin ↗Taywon Min ↗Yongjin Yang ↗Dennis Wei ↗Yi Zhou ↗Swanand Ravindra Kadhe ↗Nathalie Baracaldo ↗Kimin Lee ↗

On-policy distillation is a promising approach for transferring knowledge between language models, where a student learns from dense token-level signals along its own trajectories. This framework typically uses reverse KL divergence, encouraging the student to match the teacher's high-confidence predictions. However, we show that the mode-seeking property of reverse KL reduces generation diversity and yields unstable learning signals when the teacher distribution has high entropy. To address this, we introduce Entropy-Aware On-Policy Distillation. Our key idea is augmenting the standard reverse KL objective with forward KL when teacher entropy is high, capturing the full range of plausible outputs while retaining precise imitation elsewhere. It balances mode-seeking precision with mode-covering robustness without sacrificing on-policy training efficiency. Experiments show that our method maintains generation diversity (sustained token-level entropy) and improves student-teacher alignment (lower forward KL on high-entropy tokens). Across six math reasoning benchmarks, this yields Pass@8 accuracy gains of +1.37 for Qwen3-0.6B-Base, +2.39 for Qwen3-1.7B-Base, and +5.05 for Qwen3-4B-Base compared to baseline on-policy distillation methods. These results demonstrate that accounting for teacher uncertainty is essential for maintaining diversity and achieving effective knowledge transfer.

阅读与讨论 → 访问原文 →

19.

arXiv (CS.CL) 2026-06-18 DOI: arXiv:2606.19236

STARE: Surprisal-Guided Token-Level Advantage Reweighting for Policy Entropy Stability

作者:

Haipeng Luo ↗Qingfeng Sun ↗Songli Wu ↗Can Xu ↗Wenfeng Deng ↗Han Hu ↗Yansong Tang ↗

Reinforcement Learning with Verifiable Rewards algorithms like GRPO have emerged as the dominant post-training paradigm for complex reasoning in LLMs, yet commonly suffer from policy entropy collapse during training. We conduct a first-order gradient analysis of token-level entropy dynamics under GRPO and identify a token-level credit assignment mismatch: the per-token entropy variation decomposes into the product of the trajectory-level advantage and an entropy sensitivity function over the next-token distribution, yielding an advantage-surprisal four-quadrant structure and a near-criticality property. Motivated by it, we propose STARE (Surprisal-guided Token-level Advantage Reweighting for policy Entropy stability), which identifies entropy-critical token subsets via batch-internal surprisal quantiles, selectively reweights their effective advantages, and incorporates a target-entropy closed-loop gate for stable entropy regulation. Across model scales from 1.5B to 32B and three task families (Short CoT, Long CoT, and Multi-Turn Tool Use), STARE sustains stable RL training over thousands of steps while maintaining policy entropy within the target band. On AIME24 and AIME25, STARE outperforms DAPO and other competitive baselines by 4%-8% in average accuracy, with reflection tokens and response length growing in tandem, indicating sustained exploration-exploitation balance that further unlocks RL training potential.Code is available at https://github.com/hp-luo/STARE.

阅读与讨论 → 访问原文 →

20.

arXiv (CS.AI) 2026-06-18 DOI: arXiv:2606.18746

What Must Generalist Agents Remember?

作者:

Khurram Yamin ↗Namrata Deka ↗Maitreyi Swaroop ↗Albert Ting ↗Jeff Schneider ↗Bryan Wilder ↗

arXiv:2606.18746v1 Announce Type: new Abstract: This paper develops a formal account of what generalist agents must store in memory in order to act near-optimally across multiple environments and goals. It shows that when two domains share an observational bottleneck but require incompatible optimal actions, any uniformly near-optimal policy must induce distinct memory distributions at that bottleneck. The result yields a separation theorem: sufficiently successful agents cannot rely only on current state observations, but must preserve domain-relevant information in memory. The paper further shows that if an agent's memory contains enough information to estimate values for related goals, then that memory can be used to approximately reconstruct the agent's local transition dynamics. Together, these results characterize memory as the substrate that supports domain disambiguation, transition-model reconstruction, and planning for generalist agents.

阅读与讨论 → 访问原文 →

21.

arXiv (CS.AI) 2026-06-17 DOI: arXiv:2512.25065

Vulcan: Instance-specialized, Verifiable Systems Heuristics Through LLM-driven Search

作者:

Rohit Dwivedula ↗Divyanshu Saxena ↗Sujay Yadalam ↗Eric Hayden Campbell ↗Daehyeok Kim ↗Aditya Akella ↗

arXiv:2512.25065v2 Announce Type: replace-cross Abstract: Systems resource management tasks rely primarily on hand-designed heuristics. However, growing hardware heterogeneity and workload diversity require heuristics specialized to particular deployment instances, making manual design expensive and difficult to scale. In this paper, we explore how to synthesize systems heuristics using LLMs. The main challenge is ensuring that generated heuristics execute safely, integrate correctly with the surrounding system, and still achieve strong performance. We propose Vulcan, a framework that identifies LLM-friendly interfaces that isolate core decision logic from the rest of the implementation. With Vulcan, LLM-generated code is restricted to simple stateless decision functions, while trusted runtime abstractions provide rich derived statistics for meaningful policy exploration without system-integration bugs. To ensure execution safety, LLMs synthesize heuristics in a restricted language, Anvil, that guarantees important properties by construction. We evaluate Vulcan across three well-studied domains and demonstrate up to 4.9x higher savings for spot-VM scheduling, up to 2x lower miss ratios for cache eviction, and up to 10% higher application performance for tiered-memory systems, while ensuring execution safety throughout.

阅读与讨论 → 访问原文 →

22.

arXiv (CS.CL) 2026-06-16 DOI: arXiv:2606.14832

PhoneHarness: Harnessing Phone-Use Agents through Mixed GUI, CLI, and Tool Actions

作者:

Chenxin Li ↗Zhengyao Fang ↗Zhengyang Tang ↗Pengyuan Lyu ↗Xingran Zhou ↗Xin Lai ↗Fei Tang ↗Liang Wu ↗Yiduo Guo ↗Weinong Wang ↗Junyi Li ↗Yi Zhang ↗…

Phone agents are increasingly expected to complete real mobile workflows rather than merely predict the next screen action. However, much of the current mobile-agent literature still evaluates agents primarily as GUI controllers that observe a screen, emit taps and swipes, and are scored by target app state. Real phone-use tasks are broader: they require deciding when to use app GUIs, device-side commands, or structured tools, while leaving evidence that the intended side effect actually occurred. We introduce PhoneHarness, a mixed-action benchmark and execution harness for studying phone-use agents on verifiable mobile workflows. PhoneHarness runs a device-side agent loop over GUI, CLI, and host-side tool actions, combining deterministic action routing with bounded GUI delegation and auditable execution traces. Its benchmark, PhoneHarness Bench, evaluates whether agents complete tasks with observable side effects, not only whether they produce plausible final answers. On the annotated evaluation split, PhoneHarness reaches a 75.0% pass rate, outperforming the strongest non-PhoneHarness settings by 12.9 percentage points. PhoneHarness and PhoneHarness Bench therefore play distinct but mutually dependent roles: the harness makes mixed phone workflows executable, while the benchmark measures whether agents can use that harness reliably and safely. Our findings suggest that reliable phone automation depends on action-surface routing and verifiable execution, not only visual GUI control.

阅读与讨论 → 访问原文 →

23.

arXiv (CS.CL) 2026-06-17 DOI: arXiv:2606.17819

A Framework for Evaluating Agentic Skills at Scale

作者:

Maksim Shaposhnikov ↗Nicolas Fortuin ↗Simon Stipcich ↗Maria I. Gorinova ↗Amy Heineike ↗Rob Willoughby ↗

Agent skills – structured, reusable knowledge artifacts that augment LLM agent capabilities – have been rapidly adopted in industry, yet their cross-domain impact and use across commercial and open-source models remain under-studied, and no reusable methodology exists for evaluating an individual skill. In this work, we present an evaluation framework that lets a skill author construct realistic tasks to rigorously assess the aspects of a skill that matter most to them, and that estimates skill utility by solving those tasks. Further, we apply our evaluation approach at scale to 500 real-world skills, generating 1,000 tasks derived from the skills' content, along with instruction-following and goal-completion scoring rubrics. Using these metrics, we evaluate how 19 agent-model configurations, both proprietary and open-source, perform on the tasks. Our results show that models vary widely in how closely they adhere to the instructions encoded in skills, leading to substantial differences in their performance gains. Furthermore, we show that access to a skill significantly changes model behavior compared to the no-skill setup, providing an essential mechanism for encoding opinionated workflows into LLM agents. We release our evaluation dataset to support future work on agent skills.

阅读与讨论 → 访问原文 →

24.

arXiv (CS.AI) 2026-06-16 DOI: arXiv:2606.16222

Latent Thought Flow: Efficient Latent Reasoning in Large Language Models

作者:

Xiandong Zou ↗Jing Huang ↗Jianshu Li ↗Pan Zhou ↗

arXiv:2606.16222v1 Announce Type: new Abstract: Large Language Models (LLMs) increasingly rely on intermediate reasoning, yet explicit Chain-of-Thought (CoT) suffers from a linguistic space bottleneck: each thought must be decoded into tokens, causing high inference overhead. Latent reasoning moves deliberation into continuous space, but existing methods mostly learn deterministic or reward-maximizing paths, lacking a principled way to allocate probability across trajectories with different correctness and costs. We propose Latent Thought Flow (LTF), which models reasoning as variable-length continuous trajectories and trains a sampler to match a reward-induced posterior over answer quality and computation cost. We instantiate this with a continuous GFlowNet using stochastic latent transitions. To handle sparse answer supervision, we introduce an Entropy-Weighted Subtrajectory Balance objective for intermediate rewards and a reference-prior regularizer to anchor exploration. Experiments under finetuning and transfer learning settings show that LTF outperforms explicit CoT and latent reasoning baselines, improving accuracy by 9.5% while reducing reasoning length by 27.2% on average compared with strong latent reasoning baselines.

阅读与讨论 → 访问原文 →

25.

arXiv (CS.CV) 2026-06-16 DOI: arXiv:2606.16996

ActiveSAM: Image-Conditional Class Pruning for Fast and Accurate Open-Vocabulary Segmentation

作者:

Tran Dinh Tien ↗Zhiqiang Shen ↗

Segment Anything Model 3 (SAM 3) provides a strong frozen backbone for concept-prompted segmentation, but applying it directly to open-vocabulary semantic segmentation (OVSS) is inefficient: full-resolution decoding is typically run over the entire dataset vocabulary, whereas each image contains only a small active subset of classes. We introduce ActiveSAM, a training-free, zero-shot inference framework that turns SAM 3 into an active-vocabulary segmenter. ActiveSAM first canonicalizes and expands class prompts, then estimates an image-conditioned active set from a low-resolution presence preview. Only the retained classes are decoded at full resolution, using bucketed prompt multiplexing with the frozen SAM 3 decoder. The preview stage uses only class-presence evidence and skips unnecessary segmentation-head computation, while the final stage applies margin-aware background calibration to suppress low-confidence pixels. ActiveSAM requires no target-dataset training, no weight updates, and no oracle class-presence labels. Across eight OVSS benchmarks, ActiveSAM improves the speed-accuracy tradeoff of training-free open-vocabulary semantic segmentation, outperforming the current state-of-the-art SegEarth-OV3 by approximately +1.4 mIoU on average while running up to 5.5x faster on large-vocabulary datasets. ActiveSAM also demonstrates the strongest robustness under image corruption that simulates real-world distribution shift, making it well-suited for deployment in noisy-input domains such as autonomous driving and embodied AI. Code is available at https://github.com/VILA-Lab/ActiveSAM.

阅读与讨论 → 访问原文 →

探索全球前沿学术脉络