Academic Intelligence · Curated Daily

探索全球前沿学术脉络

AcademicHub 汇聚顶级期刊与预印本平台的实时文献。定制您的专属科研雷达,利用大语言模型自动生成交叉领域文献分析简报。

01.
arXiv (quant-ph) 2026-06-11

Honest-binding quantum bit commitment from separable operations

arXiv:2501.07351v3 Announce Type: replace Abstract: Bit commitment is a fundamental cryptographic primitive and a cornerstone for numerous two-party cryptographic protocols, including zero-knowledge proofs. However, it has been proven that unconditionally secure bit commitment, both classical and quantum, is impossible. In this work, we demonstrate that imposing a restriction on the committing party to perform only separable operations enables secure quantum bit commitment schemes. Specifically, we prove that in any perfectly hiding bit commitment protocol, an honestly-committing party limited to separable operations will be detected with high probability if they attempt to alter their commitment. To illustrate our findings, we present an example protocol.

02.
arXiv (CS.CL) 2026-06-16

Measuring Whether LLM Tutors Teach or Solve: A Diagnostic for Educational Impact

Large language models are increasingly proposed as educational tutors, yet stronger task-solving ability does not necessarily imply stronger learning support. Motivated by recent calls to measure the social impact of NLP systems in practice, we study whether public LLM tutoring benchmarks distinguish learning-supportive behavior from mere answer production. We propose a lightweight diagnostic based on the gap between solving-oriented and pedagogy-oriented benchmark performance. Using public MathTutorBench leaderboard results, we show that these dimensions are only partially aligned: across eight publicly reported models, the correlation between solving and pedagogy composites is 0.421, and several models shift meaningfully in rank when evaluation moves from solving to pedagogy. We then analyze the public TutorBench sample and show that agency-relevant behaviors are explicitly encoded in benchmark rubrics, especially in active-learning settings that reward guiding questions, calibrated hints, and non-disclosive scaffolding. Together, these findings suggest that educational-impact evaluation should not treat task success as a sufficient proxy for learning support. We argue that public tutoring benchmarks can better support positive-impact evaluation by reporting solving-oriented and pedagogy-oriented scores separately and by making disclosure-sensitive, student-agency-preserving criteria more explicit.

03.
arXiv (CS.AI) 2026-06-18

Closing the Loop: PID Feedback Control for Interpretable Activation Steering in Symbolic Music Generation

arXiv:2606.18790v1 Announce Type: cross Abstract: Transformer-based architectures have significantly advanced the generation of complex symbolic sequences, yet a significant gap remains in achieving fine-grained, interpretable control over discrete signal attributes. This paper investigates the mechanistic interpretability of the Multitrack Music Transformer (MMT) and proposes a framework for deterministic attribute modulation without retraining to bridge this gap via inference-time activation steering. Utilizing the Difference-in-Means (DiffMean) methodology, we isolate latent directions for signal attributes, specifically Pitch and Duration, within the residual stream. We validate the Linear Representation Hypothesis in this domain, achieving high correlation between steering magnitude and attribute shift. To address the inherent feature entanglement in multi-attribute steering, we introduce a Dual Steering framework utilizing Gram-Schmidt Orthogonalization. Experimental results demonstrate that this geometric decoupling reduces conceptual interference and signal degradation compared to naive vector addition, enabling independent deterministic control even against strong autoregressive conditioning.

04.
arXiv (CS.LG) 2026-06-11

Recursive Binding on a Budget: Subspace Carving in Order-p Tensor Memories

arXiv:2606.11391v1 Announce Type: new Abstract: Tensor Product Representations provide the structural fidelity required for symbolic reasoning in models but suffer from exponential dimensionality growth when encoding deep recursive structures. Conversely, Vector Symbolic Architectures maintain constant dimensionality but sacrifice capacity and fidelity due to noisy compression via superposition. In this work, we propose Orthogonal Subspace Carving (OSC), a memory architecture that binds fillers to roles by projecting onto the null space of the role basis before aggregating into a fixed order-p tensor. OSC uses projections to enforce geometric orthogonality between bound structures within a static memory trace. We show that this mechanism decouples the tensor order from the structural depth, enabling deep recursive binding within a constant memory footprint. By performing retrieval via recognition, this construction allows for component vectors that are orders of magnitude smaller than the memory tensor, giving superior memory efficiency in settings involving high superposition. We also show that TPR is a special case of binding in Clifford algebra, and give a Clifford formulation of OSC.

05.
arXiv (math.PR) 2026-06-19

A Cycle Walk for Sampling Measures on Spanning Forests for Redistricting

arXiv:2509.08629v2 Announce Type: replace-cross Abstract: We introduce the Cycle Walk, a new Markov chain Monte Carlo method for sampling distributions on balanced graph partitions, motivated by applications in political redistricting. The method operates on spanning forests and combines two types of updates: local "cycle" moves within districts and global moves that exchange population between adjacent districts while preserving balance constraints. This construction enables efficient Metropolis–Hastings correction while allowing proposals at multiple spatial scales. We show that the Cycle Walk naturally interpolates between existing approaches based on local updates and a class of global update methods derived from recombination (RECOM). Through a range of numerical experiments on synthetic graphs and real-world precinct data, we demonstrate that the Cycle Walk exhibits improved empirical convergence diagnostics for distributions that place weaker weight on spanning-tree counts, a regime that is challenging for existing methods. In particular, the algorithm remains effective when incorporating alternative compactness measures that more closely reflect policy-relevant criteria. These results suggest that the Cycle Walk provides a flexible and computationally efficient framework for sampling from a broader class of redistricting distributions than previously accessible with MCMC techniques.

06.
arXiv (CS.AI) 2026-06-17

LLMCodec: Adapting Video Codecs for Efficient Weight Compression of Large Language Models

arXiv:2606.05861v2 Announce Type: replace-cross Abstract: The rapid development of large language models(LLMs) has led to remarkable advances in natural language processing. However, the increasing scale of these models introduces substantial challenges in terms of storage, transmission, and deployment. Though great efforts have been devoted to model compression and quantization, existing methods often rely on fine-tuning or calibration data, which exhibit limited generalization across different tensor types. In this paper, we argue that video codecs offer a promising solution for LLM compression, due to their inherent compatibility with matrix structured data, configurable compression strategies, and the availability of highly optimized, off-the-shelf implementations. Therefore, we present LLMCodec, a video codec-based LLM compression method that integrates affine quantization with the recent VVC/H.266 video codec. Beyond VVC, we further compare a range of video codecs and encoding profiles to evaluate their impact on compression performance. Experiments on different models demonstrate the robustness and generality of LLMCodec. Notably, on LLaMA-3-8B at 2-bit precision, LLMCodec reduces perplexity by over 1.5x and improves downstream task accuracy by 21% compared with the existing method.

07.
arXiv (CS.CL) 2026-06-16

OBCache: Optimal Brain KV Cache Pruning for Efficient Long-Context LLM Inference

Large language models (LLMs) with extended context windows enable powerful applications but impose significant memory overhead, as caching all key-value (KV) states scales linearly with sequence length and batch size. Existing cache eviction methods address this by exploiting attention sparsity, yet they typically rank tokens heuristically using accumulated attention weights without considering their true impact on attention outputs. We propose Optimal Brain Cache (OBCache), a principled framework that formulates cache eviction as a layer-wise structured pruning problem. Building upon the Optimal Brain Damage (OBD) theory, OBCache quantifies token saliency by measuring the perturbation in attention outputs induced by pruning tokens, with closed-form scores derived for isolated keys, isolated values, and joint key-value pairs. Our scores account not only for attention weights but also for information from value states and attention outputs, thereby enhancing existing eviction strategies with output-aware signals. Experiments on LLaMA and Qwen models demonstrate that replacing the heuristic scores in existing works, which estimate token saliency across different query positions, with OBCache's output-aware scores consistently improves long-context accuracy. Code is available at https://github.com/DreamSoul-AI/OBCache.

08.
arXiv (CS.AI) 2026-06-16

Critically Engaged Pragmatism: Scientific Norm and Social, Pragmatist Epistemology for AI Science Evaluation Tools

arXiv:2601.09753v2 Announce Type: replace-cross Abstract: AI science evaluation tools aim to assess research credibility. As with traditional metrics such as impact factors, their edicts can be decontextualised and repurposed in problematic ways. To address this, I propose Critically-Engaged Pragmatism as a scientific norm enjoining scientific communities to scrutinise the purposes and purpose-specific reliability of AI science evaluation tools. To foster Critically Engaged Pragmatism, creators of AI science evaluation tools should transparently and fully report design, training, and benchmarking details to facilitate assessments of purpose-specific reliability, liability to different types of error, and bias. What count as best practices for the transparent reporting of AI science evaluation tools should be updated as new forms of error, bias, and gamesmanship are discovered. Under this framework, AI science evaluation tools are not objective arbiters of scientific credibility. Rather, they are the object of critical discursive practices that ultimately ground the credibility of scientific communities.

09.
arXiv (CS.AI) 2026-06-17

When Rules Learn: A Self-Evolving Agent for Legal Case Retrieval

arXiv:2606.17220v1 Announce Type: new Abstract: Legal case retrieval remains challenging due to the complexity of legal language and the need for precise lexical alignment between queries and relevant cases. Although dense retrieval models have achieved notable progress, empirical studies show that BM25 continues to serve as a strong baseline in this domain. It motivates us to propose a self-evolving framework for rule-driven query rewriting that enhances BM25 without any parameter training. The framework equips an LLM-based agent with an automatic evaluation environment, enabling it to iteratively create rewriting rules, plan validation experiments over rule combinations, and eliminate ineffective rules based on historical feedbacks. We evaluate our method on the Chinese legal case retrieval benchmark LeCaRD-v2. Experimental results demonstrate that the proposed framework outperforms non-evolutionary baselines, including human-designed rules and greedy rule selection, particularly when powered by a highcapacity core LLM. We also conduct detailed analyses to investigate the mechanisms underlying self-evolution. Our findings reveal that LLM's capabilities to leverage previous experimental results and its intrinsic knowledge of rule elimination play critical roles in refining the rule set via self-evolution.

10.
arXiv (quant-ph) 2026-06-16

Witnessing Spin-Orbital Entanglement using Resonant Inelastic X-Ray Scattering

arXiv:2512.06718v2 Announce Type: replace Abstract: Entanglement plays a central role in quantum technologies, yet its characterization and control in materials remain challenging. Recent developments in spectrum-based entanglement witnesses have enabled new strategies for quantifying many-body entanglement in macroscopic materials. Here, we develop a protocol for detecting spin-orbital entanglement using experiment-accessible resonant inelastic x-ray scattering (RIXS). Central to our approach is the construction of a Hermitian generator from experimentally measurable spectra, which allows us to compute the quantum Fisher information (QFI) available in spin–orbital systems. The resulting QFI provides upper bounds for $k$-producible states and thus serves as a robust witness of spin-orbital entanglement. To account for realistic experimental limitations, we further extend our framework to include relaxed QFI bounds applicable to measurements lacking full polarization resolution.

11.
arXiv (CS.AI) 2026-06-16

UXBench: Measuring the Actionability of LLM-Generated UX Critiques

arXiv:2606.16262v1 Announce Type: cross Abstract: Large language models (LLMs) are increasingly deployed as UX judges that inspect interfaces, diagnose usability problems, and propose repairs. Yet no controlled benchmark measures whether the resulting critiques are reliable and actionable across heterogeneous product surfaces. We introduce UXBench, a benchmark for evaluating LLMs as interaction-grounded UX judges. UXBench comprises local-first runnable web fixtures spanning ten product-surface families, paired with coverage-gated browser exploration that forces models to collect interaction evidence before reporting. Each judge model produces a structured UX report over seven rubric dimensions; report quality is measured by whether a fixed downstream repair agent can improve the interface based on the critique. We evaluate eight frontier models under both an automated repair-lift protocol and a blind human validation study. Results show that UX judging is neither saturated nor one dimensional: models differ meaningfully in report actionability, exhibit distinct rubric-level repair signatures, vary in fixture-level reliability, and trade leadership across surface categories

12.
arXiv (CS.AI) 2026-06-17

C2FL: Clustered Continual Federated Learning under Spatial and Temporal Drift

arXiv:2606.18003v1 Announce Type: cross Abstract: Collective Adaptive Systems (CAS) increasingly rely on machine learning to let each node learn from locally sensed data, aligning its behavior with the surrounding environment. Scaling this intelligence, however, raises fundamental challenges: sensed data is often privacy-sensitive, preventing centralized collection; nodes are mobile, traversing regions where nearby nodes perceive similar phenomena while distant ones observe radically different conditions, creating natural spatial clusters; and these distributions evolve over time due to mobility, introducing temporal drift that makes local models progressively stale. These dynamics arise across domains - vehicular sensing, drone-based monitoring, smartphone crowdsensing - yet the interplay of privacy, spatial heterogeneity, and temporal drift severely undermines conventional learning strategies. Therefore, we propose C2FL, a fully distributed Federated Learning (FL) approach where nodes self-organize into learning groups through spatial clustering, reflecting the geographic structure of the environment. To counteract temporal drift, each node combines experience replay with a dwell-time-aware adaptive averaging step, progressively incorporating the regional consensus as it remains longer within the same area, while preserving previously acquired knowledge under evolving distributions. We evaluate our approach on synthetic experiments that systematically reproduce spatial and temporal shifts, showing that standard federated strategies degrade significantly under these conditions and that our method restores robust collective adaptation.

13.
arXiv (CS.AI) 2026-06-17

Kolmogorov Regression for Robust Diffusion Policies

作者:

arXiv:2606.18186v1 Announce Type: cross Abstract: Finite-dimensional (FD) diffusion policies exhibit temporal drift owing to discretization artifacts that degrade long-horizon performance (when deployed on physical systems). We introduce a backward Kolmogorov equation that lifts diffusion policies to a Cameron-Martin space – a subset of the Hilbert space. Essentially, replacing stochastic score matching with a deterministic boundary-value PDE problem. Our core innovation thrives on Gaussian measure theory whereupon the diffusion noise covariance operator is realized from a colored noise distribution which prescribes a notion of regularity on samples from the model at inference time. We train the diffusion model with a derived precision-weighted Cameron- Martin loss and a Kolmogorov residual is introduced as a PDE diagnostic during inference. These substitutions yield (i) convergence guarantees where the bound's constants depend on the effective rank of the kernel rather than action dimension, (ii) improved trajectory regularity via spectral weighting, and (iii) a deterministic failure detector without reward signals. Validation across two application domains demonstrates substantial improvements: on the PushT manipulation benchmark, the Cameron-Martin loss achieves a 17% improvement in maximum episode reward (0.95 vs. 0.78 for MSE) and 67.6% reduction in inter-step drifts during inference via the introduced residual magnitude. Similarly, on a 6-station manufacturing line with constant work-in-process (CONWIP) flow control, we achieve 28.4% lower RMSE than classical LSTM baselines; a high starvation-event recall (1.0 in test cycles), and effective bottleneck identification (Precision@1 = 1.0 in test set, 13x signal-to-noise ratio). We then certify the dispatch policies with Hamilton-Jacobi reachability theory which reduces deadlock events by 96% compared to uncontrolled dispatch over 100 simulated runs (351 events prevented).

14.
medRxiv (Medicine) 2026-06-22

ECG-Guided Pre-Screening of Family Members for Hypertrophic Cardiomyopathy

Background: Current clinical guidelines recommend serial ECG and echocardiographic surveillance for first-degree relatives of probands with Hypertrophic Cardiomyopathy (HCM). Objectives: To evaluate the accuracy and validity of ECG alone as a pre-screening tool for the diagnosis of HCM and to develop a random forest (RF) model for HCM phenotype prediction. Method: Pediatric relatives of primary HCM probands attending the cardiomyopathy screening program at The Hospital for Sick Children were included from 1993 to 2025. Subjects were followed until the last follow-up, censored at phenotype conversion. ECGs were classified as normal or abnormal based on predefined parameters. Associations between binary ECG variables and HCM phenotype were assessed using Phi ({varphi}) coefficient. A Random Forest classifier was developed using significant ECG variables (70:30 training: test split) and evaluated using precision, recall, specificity, negative predictive value, F1 score and AUROC. Feature importance was assessed using SHAP analysis. Variables with an impact of >5% were included in a simplified model, which was evaluated by repeating performance metrics and externally validated in a healthy cohort. Results: 350 screened relatives (44% female, mean follow-up 6.8 +- 4.8 years) were included. At baseline, 13% (46350) were phenotype-positive for HCM. 9 subjects converted during the surveillance. Thirteen ECG variables were significantly associated with phenotype-positive HCM and were included in the full random forest model. Four variables had >5% impact (Left ventricular hypertrophy, right ventricular hypertrophy, T-wave inversion and ST-segment depression) and were included in a simplified model, which maintained high specificity (93% vs 97%), negative predictive value (97% vs 93%) and AUROC (90% vs 96%). The simplified model classified 83% subjects as phenotype-negative, with eight being false-negative, all of whom developed an abnormal ECG in a mean of 1 year, and none had an interim adverse cardiac event. The simplified model was evaluated in an independent healthy cohort of 153 school-age subjects and correctly identified 98% as phenotype-negative with 100% NPV. Conclusion: ECG abnormalities were strongly associated with phenotype-positive status. A simplified ECG-based random forest model using four ECG variables demonstrated high specificity and negative predictive value for identifying phenotype-negative subjects. If prospectively validated, this could reduce the need for concurrent echocardiographic screening by up to 83% per encounter, lowering screening burden and cost.

15.
arXiv (CS.CV) 2026-06-17

Universal Image Restoration via Internalized Chain-of-Thought Reasoning

Image restoration seeks to recover high-quality images from degraded inputs but becomes highly ill-posed under complex, mixed degradations. While unified all-in-one models are common, their performance declines as degradation complexity increases. Recent works adopt Chain-of-Thought (CoT) reasoning for multi-round restoration using specialized modules. However, this approach faces two key limitations: (i) increased computational cost due to multi-step processing, and (ii) weak modeling of interactions between degradations during stepwise inference. We introduce CoTIR, a universal image restoration framework that internalizes CoT reasoning within a single model. Concretely, we view image restoration as a specialized subtask of image editing, which implies that a large-scale pre-trained editing model provides a more favorable optimization starting point. Building on this, we fine-tune the model for restoration and further encode structured CoT-style reasoning into the learning objective via a differentiable formulation inspired by Lagrangian optimization, enabling holistic restoration without chaining specialized restorers. To facilitate training and evaluation, we further present CoTIR-Bench, a large-scale benchmark comprising 5.2 million samples with CoT-style reasoning traces. Extensive experiments on CoTIR-Bench and broad real composite degradation scenes show that CoTIR achieves stronger perceptual quality and more competitive fidelity than both all-in-one models and multi-round restoration methods. The source code is available at https://github.com/gy65896/CoTIR.

16.
arXiv (quant-ph) 2026-06-11

Bound State Solutions of the Relativistic Finite-difference Equation for the Ring-shaped Quesne Oscillator Potential

arXiv:2606.12082v1 Announce Type: new Abstract: We solve exactly the relativistic finite-difference equation for the quantum three-dimensional ring-shaped Quesne oscillator potential. Our investigation is based on a finite-difference version of relativistic quantum mechanics. So-called relativistic configurational r-space is a key concept here. We show that the radial wavefunctions and angular wavefunctions are expressed through the continuous dual Hahn polynomials and Jacobi polynomials, respectively. A discrete energy spectrum has been found. The radial wave functions and energy spectrum have the correct nonrelativistic limit. We also build a dynamical symmetry group SU (1, 1) for the radial part of the equation of motion, which allows us to find the energy spectrum purely algebraically.

17.
arXiv (CS.CV) 2026-06-16

MVEB: Massive Video Embedding Benchmark

We introduce the Massive Video Embedding Benchmark (MVEB), a 23-task benchmark for video embeddings spanning classification, zero-shot classification, clustering, pair classification, retrieval, and video-centric question answering. We evaluate 33 models and find that no single model dominates: MLLM-based embeddings lead on classification, clustering, pair classification, and QA; multimodal binding leads on retrieval and zero-shot classification; generative MLLMs without contrastive adaptation collapse on cross-modal tasks. Paired video-only vs. audio+video evaluations show that audio's contribution depends on dataset annotation provenance: audio helps when labels were produced from both modalities and hurts when they were produced from visuals alone, a six-point gap consistent across model families. MVEB is derived from MVEB+, a 184-task pool, and is designed to maintain task diversity while reducing evaluation cost. It integrates into the MTEB ecosystem for unified evaluation across text, image, audio, and video. We release MVEB and all 184 tasks along with code and a leaderboard at https://github.com/embeddings-benchmark/mteb.

18.
arXiv (CS.CV) 2026-06-18

Spatially Stratified Distillation for Heterogeneous Radar Place Recognition

Scalable, all-weather place recognition increasingly relies on heterogeneous radar place recognition to bridge diverse hardware platforms. A notable application is matching queries from cost-effective 4D automotive radars against high-fidelity reference maps built by dense spinning radars. This process is fundamentally limited by the extreme sparsity (and narrow field-of-view) of the 4D sensor, which captures only a fraction of the structural density present in the spinning radar database. Prior efforts address this issue by unifying different radar signals. That is, projecting both signals into a common representational space. Yet, they suffer performance degradation in multi-session environments. In this paper, we propose spatially-stratified distillation (SSD); a strategy that replaces standard uniform distillation with an asymmetric spatial alignment derived directly from physical radar returns. In regions where both radars exhibit overlapping returns, SSD enforces strong feature alignment. Crucially, in sparse regions where the 4D student lacks returns but the teacher contains valid structure within the shared field of view, SSD applies heavily discounted distillation weights. Extensive evaluations of the recent HeRCULES dataset demonstrate that SSD significantly outperforms prior place recognition methods, achieving state-of-the-art results on its challenging dynamic sequences.

19.
arXiv (CS.AI) 2026-06-11

Generalization Hacking: Models Can Game Reinforcement Learning by Preventing Behavioral Generalization

arXiv:2606.12016v1 Announce Type: cross Abstract: Model post-training, and in particular reinforcement learning (RL), is one of the primary mechanisms by which developers can shape models' values and behaviors. However, as models become increasingly evaluation and training aware, they may be motivated to resist training when the perceived objective conflicts with their current values, undermining developers' ability to detect misalignment and correct model behavior through further training. In this paper, we demonstrate generalization hacking, in which a model collects reward during RL while preventing the rewarded behavior from generalizing. We construct a model organism on Qwen3-235B-A22B, finetuning on synthetic documents describing training awareness and self-inoculation, a novel mechanism in which the model frames compliance as context-specific in its chain of thought, without demonstrating or instructing either behavior. The model organism achieves train-time harmfulness comparable to controls while maintaining a persistent ${\sim}15$ percentage point compliance gap across 700 steps of RL. Additionally, a control organism trained only on training awareness documents independently discovers inoculation-like reasoning under RL pressure, developing its own compliance gap despite never being exposed to the concept. Because the generalization-hacking organism receives high reward throughout, standard training metrics provide no signal that generalization has failed. Our results constitute the first demonstration that a model can actively resist RL behavioral modification while maintaining high reward, suggesting that as models become more capable and training-aware, they may be able to undermine the training process itself.

20.
arXiv (CS.AI) 2026-06-16

Faster Completion, Less Learning: Generative AI Reduced Study Time on Math Problems and the Knowledge They Build

arXiv:2605.21629v2 Announce Type: replace-cross Abstract: How much have students' ordinary learning processes shifted in response to generative AI, and how does that affect their durable learning outcomes? Self-report surveys show little change, while small-scale behavioral studies report widespread AI use without the scale or duration to measure learning consequences. We address both questions using a ten-year panel of $3.2$ million ALEKS learning interactions for investigating time-on-task, complemented by ALEKS PPL placement-assessment data for examining proctoring and learning outcomes, with a quasi-experimental design exploiting variation in tasks that are more susceptible to AI (text-based word problems) and less susceptible to AI (interactive graph-based problems). Learning time on AI-susceptible problems declines $2.8\%$ per quarter among college students after ChatGPT's release, cumulating to $26.9\%$ over eleven quarters; high-schoolers show $31.3\%$, middle-schoolers $9.0\%$, and Grade 5 students no detectable change. Among college students, the post-ChatGPT divergence vanishes entirely under proctoring, ruling out broad efficiency gains as the likely explanation. Logistic fixed-effects models on randomly assigned proctored retention items yield a $25\%$ cumulative decline in odds of correct response; the same estimator on non-proctored assessment produces a large opposite-signed increase – inconsistent with any platform, cohort, or curriculum explanation. These results are among the first large-scale behavioral and outcome evidence that generative AI has altered how students study and the knowledge they build – the population-level indicator of cognitive surrender, with direct implications for educational research, assessment governance, and AI policy.

21.
arXiv (CS.AI) 2026-06-19

Uncertainty-Aware Reward Modeling for Stable RLHF

arXiv:2606.19818v1 Announce Type: cross Abstract: Reinforcement learning from human feedback (RLHF) aligns large language models by training reward models on preference data and optimizing policies to maximize predicted rewards. However, this pipeline faces two fundamental challenges: (1) reward models cannot signal when their predictions are unreliable, since they usually act as deterministic point estimators; and (2) modern group-based policy optimization can amplify unreliable reward signals, as exemplified by GRPO's uniform treatment of rewards during advantage computation. As policies explore increasingly diverse responses, these two limitations create a critical vulnerability: unreliable reward estimates may be granted disproportionate influence, triggering severe reward hacking. We propose Uncertainty-Aware Reward Modeling (UARM), which equips reward models with calibrated uncertainty via quantile-based conformal prediction and reweights GRPO advantages through heteroscedastic variance decomposition. Experiments across HelpSteer, UltraFeedback, and PKU-SafeRLHF demonstrate that UARM significantly improves reward model calibration, reduces reward hacking, and enhances downstream alignment quality compared to standard GRPO and uncertainty-agnostic baselines.

22.
arXiv (CS.AI) 2026-06-18

SafeClawBench: Separating Semantic, Audit-Evidence, and Sandbox Harm in Tool-Using LLM Agents

arXiv:2606.18356v1 Announce Type: cross Abstract: Tool-using language-model agents introduce security failures that go beyond unsafe text: they can disclose protected objects, write persistent memory, send messages, modify databases, or trigger harmful code and tool effects. Existing evaluations often collapse these stages into a single attack success rate, making it difficult to tell whether a model merely agreed with an attacker or actually produced observable harm. We introduce SafeClawBench, a staged benchmark for tool-using agent security with 600 controlled adversarial tasks across six attack families: direct and indirect prompt injection, tool-return injection, memory poisoning, memory extraction, and ambiguity-driven unsafe inference. SafeClawBench reports three separate endpoints: semantic attack acceptance, audit-visible harm evidence, and sandbox-observed tool/state harm. Evaluating five agent endpoints under four prompt-level policies, we find that these endpoints capture different failure modes. Without additional prompt protection, semantic failure rates vary widely across models, from 9.0% to 44.2%. Audited harm evidence is narrower than semantic failure, and under a separate executable protocol some matched task identities produce sandbox harm despite passing the Semantic Core call: in a 12,000-row matched analysis, 291 of 347 observed sandbox harms occur in rows that pass the semantic check. Prompt policies change endpoint outcomes, but their effects depend on both model and protocol. SafeClawBench provides a reproducible framework for comparing agent models and prompt-policy conditions without conflating textual compliance, evidence-supported harm, and executable state changes. The open-source dataset is available at https://huggingface.co/datasets/sairights/safeclawbench.

23.
medRxiv (Medicine) 2026-06-11

Corticospinal tract risk modifies motor recovery after minimally invasive surgery for intracerebral hemorrhage: a secondary analysis of MISTIE-III

Objective: Outcome after surgical hematoma evacuation for intracerebral hemorrhage (ICH) depends on hematoma location. As corticospinal tract (CST) integrity affects motor recovery after stroke, we hypothesized that CST integrity drives heterogeneity in surgical outcomes and investigated this in a secondary analysis of MISTIE-III participants. Methods: Risk of CST injury was categorized into four levels, based on the interaction between the CST, the hematoma, and perihematomal edema (PHE) on automatically segmented stability CT: no risk, PHE infiltration, hematoma infiltration, and complete interruption of the CST. Associations with outcome were tested using multivariable linear regression for motor National Institutes of Health Stroke Scale (NIHSS) at day 180 and ordinal regression for modified Rankin Scale (mRS) at day 365, introducing an interaction term between CST risk and treatment group. Results: Day 180 motor NIHSS was significantly lower for 'no risk' ({beta}:-3.77, [95% confidence interval [CI]: -5.8 to -1.70], p=0.0003) and 'PHE infiltration' ({beta}:-2.3, [95%CI: -3.5 to -1.1]; p=0.0002) vs. 'complete interruption'. Surgery was associated with lower Day 180 motor NIHSS in participants with hematoma infiltration ({beta}:-2.07, [95%CI: -3.8 to -0.4], p=0.016). Compared to complete interruption, 'no risk' (adjusted odds ratio [aOR]:0.27, [95%CI: 0.10 to 0.74], p=0.01) and 'PHE infiltration' (aOR:0.41, [95%CI: 0.23 to 0.74]; p=0.003) were associated with lower odds of unfavorable day 365 mRS. Surgery was associated with lower mRS in participants with no risk (aOR:0.23, [95%CI: 0.05 to 0.97, p=0.045). Interpretation: Increasing CST risk is associated with worse motor recovery (day 180) and disability (day 365). CST risk modifies the effect of the MISTIE-III procedure on motor recovery and disability.

24.
arXiv (CS.AI) 2026-06-12

Constructing Evaluation Datasets for Procedural Reasoning: Balancing Naturalness, Grounding, and Multi-Hop Coverage

arXiv:2606.12767v1 Announce Type: new Abstract: Evaluating procedural reasoning in AI-supported learning systems requires question-answer datasets that are both learner-like and grounded in the instructional knowledge the system is expected to use. We study how TMK-based question generation strategies affect dataset quality for procedural and multi-hop reasoning. We compare three strategies: strict generation from Task-Method-Knowledge (TMK) models, transcript-first generation with post-hoc TMK filtering, and TMK-aware generation that combines transcripts with structured guidance. To evaluate generated items, we introduce a grounding validation framework based on closed-set evidence units extracted from TMK models. The framework measures whether answers are supported by the underlying representation, whether questions are self-contained, and whether they target multi-hop procedural reasoning. Across 23 instructional topics and 690 generated question-answer pairs, strict TMK generation achieves the strongest overall quality, with 96.5% grounded questions and 92.6% usable questions. Transcript-first generation produces more learner-like questions but more context-dependent or weakly grounded items, while TMK-aware generation yields high raw multi-hop coverage but lower grounding. These results show that procedural richness and natural phrasing do not guarantee representational grounding, motivating explicit representation-aware validation for evaluation datasets in AI-supported learning.

25.
arXiv (CS.CL) 2026-06-16

Surpassing Scale by Efficiency: A Compact 135M Parameter Foundational LLM Natively Adapted for the Bangla Language

While the NLP landscape is dominated by multi-billion parameter architectures, their deployment in low-resource, non-Latin scripts remains computationally prohibitive for edge configurations, mobile systems, and decentralized local hardware. This paper presents bangla-smollm-135m, a highly compact 135-million parameter decoder-only foundational model engineered explicitly for high-efficiency language modeling in the Bangla script. By leveraging a deterministic intersect-and-append token merging strategy between TituLLMs and SmolLM2-135M, the model overcomes subword script fragmentation without destabilizing early pretrained parameter states. In zero-shot multi-task benchmark evaluations (PIQA_bn, OpenBookQA_bn, CommonsenseQA_bn, and Bangla_MMLU), bangla-smollm-135m matches or outperforms models twice its size (Gemma-3-270m) and achieves parity with models in the 1B parameter tier. The model is available at rnnandi/bangla-smollm-135m