探索全球前沿学术脉络

01.

arXiv (CS.AI) 2026-06-16 DOI: arXiv:2606.15349

LearnOpt: Recovering the Latent Cognitive Structure of Standardized Examinations via Knowledge Graphs and Constrained Optimization

作者:

Joy Bose↗Om Thomas↗

arXiv:2606.15349v1 Announce Type: cross Abstract: Standardized examinations are typically treated as uniform syllabus coverage problems. We argue they are better understood as adversarial systems with stable latent cognitive structures diverging systematically from official syllabi. We introduce LearnOpt, which recovers this structure from historical question papers and generates personalized, time-bounded study plans. Applied to nine years of NEET questions (2016-2024, n=1,496), LearnOpt builds an exam knowledge graph from LLM-tagged questions, extracts a five-category latent skill distribution, and formulates study planning as a knapsack-variant optimization over prerequisite-aware subgraphs with Bayesian Knowledge Tracing. Central finding: NEET's latent skill distribution is stable within a syllabus regime (consecutive-year KL divergence 0.004-0.032 for 2016-2021, non-significant under permutation testing) but shifts significantly with NCERT's 2023 syllabus rationalization: pooling 2016-2021 (n=1,072) vs 2023-2024 (n=392) gives KL=0.040 (p=0.0005), with Elimination/Negation questions rising from ~20-29% to ~31-35%. Latent structure, while not permanently stationary, is piecewise stable, with shifts detectable and attributable to curricular events. Within either regime, subject predicts skill profile more strongly than year. An optimization evaluation, using one real and two synthetic mastery profiles, shows the skill-weighted objective produces a modest but real reordering of recommended topics over a mastery-conditioned frequency baseline. Applying the pipeline to JEE Advanced reveals a profile dominated by Multi-concept Integration (80.9% vs. 33.3% for NEET), with a JEE-vs-NEET divergence (KL=0.505) exceeding NEET's largest cross-subject divergence: exam tier shapes latent cognitive structure more than subject, which shapes it more than time within a regime. Code, knowledge graph, and annotated dataset are released publicly.

阅读与讨论 → 访问原文 →

02.

arXiv (CS.CL) 2026-06-19 DOI: arXiv:2606.19345

Ensembles of Large Language Models for Identifying EQ-5D Studies in PubMed Based on Their Abstracts

作者:

Zhyar Rzgar K. Rostam↗M\'arta P\'entek↗J\'anos Tibor Czere↗Zsombor Zrubka↗L\'aszl\'o Gul\'acsi↗G\'abor Kert\'esz↗

The rapid increase in scientific publications leads to the fact that manual study screening in systematic literature reviews (SLRs) is increasingly resource consuming, inefficient, and inconsistent. Classifying studies that clearly report health-related quality-of-life results, such as EQ-5D data, requires a high level of clinical interpretation and poses challenges for human reviewers. This study investigates the use of Google's Gemini and Gemma large language models (LLMs) in automating EQ-5D detection in the PubMed biomedical database based only on published abstracts. A multi-phase framework is proposed that integrates few-shot prompting, weight ensembling aggregation, and a soft stacking meta-classifier. Nine LLMs are evaluated on a dataset of PubMed studies manually labeled by two experts regarding EQ-5D reporting. The weighted ensemble of gemini-2.5-pro, gemma-3-12b, and gemma-3-27b obtained a 0.74 weighted F1-score and 0.74 accuracy, exceeding individually attained results. The ensembling of top-performing models improved the balance between precision and recall compared to individual models, while the soft stacking approach provided greater reliability and interpretability. Feature analysis shows that the probability results from the models are important in guiding the final predictions. The findings suggest that an ensemble-based LLM setup is a reliable and scalable approach for automating screening in biomedical research.

阅读与讨论 → 访问原文 →

03.

arXiv (CS.LG) 2026-06-12 DOI: arXiv:2606.13589

Simplex-Constrained Sparse Bagging: Transitioning from Uniform Priors to Sparse Posteriors in Ensemble Learning

作者:

Meher Sai Preetam↗Meher Bhaskar↗

arXiv:2606.13589v1 Announce Type: new Abstract: We present Simplex-Constrained Sparse Bagging (SCSB), a mathematically rigorous framework for post-training compression and probability calibration of bootstrap-based bagging ensembles. Standard bagging ensembles (such as Random Forests, Bagged SVMs, and Bagged Neural Networks) assign uniform voting power to all constituent estimators. However, this naive uniform prior ignores the varying local competence of base estimators and contributes to model overconfidence. We formulate ensemble pruning and calibration as a joint optimization problem over the probability simplex by minimizing the Out-Of-Bag (OOB) loss. To induce sparsity, we address the theoretical "L1-simplex paradox" – the mathematical reality that the L1 norm is constant on the simplex and fails to prune – by introducing a concave quadratic penalty. SCSB is model-agnostic and achieves up to 96% ensemble compression, yielding linear inference speedups and superior probability calibration (lowered Expected Calibration Error) while preserving or enhancing generalization accuracy.

阅读与讨论 → 访问原文 →

04.

arXiv (CS.LG) 2026-06-19 DOI: arXiv:2603.10184

Stabilizing Bandits using Regularization: Precise Regret and A Quantitative Central Limit Theorem

作者:

Budhaditya Halder↗Ishan Sengupta↗Koustav Chowdhury↗Samya Praharaj↗Koulik Khamaru↗

arXiv:2603.10184v2 Announce Type: replace-cross Abstract: Statistical inference with bandit data presents fundamental challenges owing to adaptive sampling, which violates the independence assumptions underlying classical asymptotic theory. Recent work has identified stability~\citep{laiwei82} as a sufficient condition for valid inference under adaptivity. This paper first provides a refined stability condition, stated in terms of the iterates of an online algorithm, and shows that a large class of regularized stochastic-mirror-descent-style algorithms satisfy it. This refined condition allows us to strengthen the asymptotic results of~\citet{laiwei82} in several ways. First, we derive a non-asymptotic Berry–Esseen bound for the empirical reward estimates under adaptive sampling. Second, we derive matching non-asymptotic upper and lower bounds on the regret of the proposed algorithm, yielding a precise characterization of its regret. Third, we show that these regularized algorithms preserve asymptotic normality and valid inference under a prescribed level of adversarial corruption. Finally, we show that regularization is necessary rather than incidental: Lai–Wei stability is incompatible with the optimal $O(\sqrt{T})$ regret rate – the rate attained by unregularized algorithms such as EXP3 – so that a controlled, polylogarithmic inflation in regret is the price of valid inference.

阅读与讨论 → 访问原文 →

05.

medRxiv (Medicine) 2026-06-17 DOI: HASH:78c2a7dab257abe9176b09b85e7085e5

Womens intentions and motivations towards health behaviour change before pregnancy: a cross-sectional survey of pregnant women in Australia

作者:

Steel↗Schoenaker↗McIntyre↗Rogers↗Hall↗Adams↗

Introduction: The preconception period (i.e. the weeks and months before pregnancy) is a critical window during which parental health behaviours can influence pregnancy outcomes and the childs long-term health. Modifiable factors such as nutrition, physical activity, substance use, and environmental exposures play a key role, yet womens ability to adopt and sustain healthy behaviours is shaped by complex psychological, social and environmental influences. This study applies the Theory of Planned Behaviour to identify the beliefs underpinning womens preconception behaviours, with the aim of informing support for effective and sustained health behaviour change. Methods: An Australian national retrospective cross-sectional survey of pregnant women (18-49 years), recruited through social media platforms. The 92-item survey captured respondent socio-demographics, pregnancy status and health conditions, health behaviours, and beliefs regarding preconception health behaviours. Respondents level of pregnancy planning was categorised using the London Measure of Unplanned Pregnancy (LMUP). Items regarding preconception beliefs were structured in accordance with the Theory of Planned Behaviour, with a focus on regular exercise, healthy diet, and alcohol avoidance. These beliefs variables were analysed using structured equation modelling to identify paths between latent variables and the items used to estimate each concept. Results: The study was completed by 430 pregnant women of whom 72.7% had a planned pregnancy. Most had a partner, were university educated and in good health. Structural equation modelling showed intention strongly predicted exercise ({beta}=0.65), healthy diet ({beta}=0.54) and alcohol avoidance ({beta}=0.64). Perceived control and partner norms influenced intentions, whereas health professional norms had limited effect. Positive beliefs were associated with folate supplement use and smoking cessation. Conclusion: These findings highlight intention as a key driver of preconception health behaviours, with perceived control and partner influences playing a more significant role than individual beliefs or health professional input. Effective interventions should therefore address structural barriers and actively involve partners, while respecting womens autonomy. Overall, couples-focused, multi-level strategies are likely essential to support meaningful and sustained preconception health behaviour change.

阅读与讨论 → 访问原文 →

06.

arXiv (CS.LG) 2026-06-12 DOI: arXiv:2606.13543

NetCause: Counterfactual Learning for Root Cause Analysis in Large-Scale Networks

作者:

Fabien Chraim↗Jian Zhang↗Dominik Janzing↗Xiang Song↗Christos Faloutsos↗John Evans↗

arXiv:2606.13543v1 Announce Type: cross Abstract: Can a learned model capture how faults propagate through a large-scale network and use this knowledge to causally attribute customer impact to its underlying root cause? Existing root cause analysis techniques often rely on static rules, correlation heuristics, or topology-local reasoning, which struggle to generalize in dynamic environments where faults propagate across complex physical and logical dependencies. We present NetCause, a self-supervised learning-based framework that models network incidents as graph-temporal processes and uses counterfactual simulation to rank candidate root causes. This approach produces an interpretable ranking of root cause hypotheses and integrates naturally with operator-defined mitigation and remediation actions. We train the model on over 1,500 incidents collected over six months from a leading cloud provider's production network and evaluate it on 31 expert-labeled incidents. NetCause consistently improves root cause ranking quality in the regime most relevant to operational decision-making, achieving a 16.1% accuracy improvement over a rule-based heuristic baseline. While training is computationally intensive, inference is lightweight, requiring only seconds of GPU runtime per incident (well below typical telemetry collection latencies).

阅读与讨论 → 访问原文 →

07.

arXiv (CS.LG) 2026-06-18 DOI: arXiv:2606.18697

Stealthy World Model Manipulation via Data Poisoning

作者:

Yibin Hu↗Xiaolin Sun↗Zizhan Zheng↗

arXiv:2606.18697v1 Announce Type: new Abstract: Model-based learning agents use learned world models to predict future states, plan actions, and adapt to new environments. However, the process of updating world models from collected experience creates a training-time attack surface: adversarially poisoned fine-tuning trajectories can manipulate the learned dynamics and thereby corrupt downstream planning. In this paper, we propose SWAAP, the first two-stage data poisoning framework for learned world models. In the first stage, SWAAP identifies a harmful target world model that induces low-return behavior under planning while remaining close to clean dynamics, using first-order bilevel optimization enabled by a transition-gradient theorem. In the second stage, SWAAP realizes this target through stealth-constrained gradient matching, modifying only a limited fraction of fine-tuning transition targets so that the induced training gradients steer the victim model toward the adversarial target, while a prediction-error regularizer encourages the poisoned targets to remain close to the world model's natural approximation error. To assess attack stealthiness, we evaluate defenses and detectability across three stages of the poisoning pipeline: pre-training detection of poisoned transitions, robust training during fine-tuning, and test-time monitoring of the resulting world model. Across diverse continuous-control tasks, SWAAP causes substantial performance degradation while keeping poisoned transitions close to clean data and evading the evaluated non-adaptive residual/CUSUM/TRIM-style defenses. These results reveal a practical vulnerability in world-model adaptation pipelines and highlight the need for robustness methods that protect both world-model training data and learned dynamics.

阅读与讨论 → 访问原文 →

08.

arXiv (CS.AI) 2026-06-12 DOI: arXiv:2606.12500

Improving Crash Frequency Prediction from Simulated Traffic Conflicts Using Machine Learning Based Microsimulation

作者:

Xian Liu↗Carlo G. Prato↗Gustav Markkula↗

arXiv:2606.12500v1 Announce Type: cross Abstract: Traffic microsimulation combined with surrogate safety measures has increasingly been used as a proactive alternative to historical crash data for predicting crash frequency for current or planned road infrastructure designs. However, existing microsimulation-based safety studies have adopted simplified rule-based behaviour models, which reproduce traffic flow reasonably well but often fail to generate realistic conflict dynamics, limiting crash prediction accuracy. Recent advances in machine learning (ML)-based behaviour models offer a promising opportunity to potentially improve microsimulation realism and crash frequency predictions by learning human driving behaviour directly from large-scale trajectory datasets. To investigate this possibility, traffic microsimulation was conducted for five real-world signalised intersections in Leeds, UK, using both a standard rule-based model and a state-of-the-art ML model. Simulated vehicle trajectories were analysed using a two-dimensional Time-to-Collision metric to identify simulated conflicts, which were then modelled using Extreme Value Theory to predict crash frequency. Results show that conflicts from the ML model yielded crash predictions in line with the real-world crash data, whereas the rule-based model did not permit meaningful predictions, presumably due to a lack of model calibration to the specific simulated intersections. Directly using ML-generated simulated crashes to predict real-world crash frequency also yielded poor results, suggesting that while current ML models can realistically reproduce conflicts, they are not yet able to generate realistic crashes. Overall, the findings demonstrate that ML-based behaviour models are promising for improving crash prediction from simulated conflicts, without a need for location-specific model calibration, and suggest clear future directions for ML-based traffic microsimulation.

阅读与讨论 → 访问原文 →

09.

arXiv (CS.CV) 2026-06-16 DOI: arXiv:2606.16742

Revealing Artifacts via Noise Amplification: A Novel Perspective for AI-Generated Video Detection

作者:

Renxi Cheng↗Jie Gui↗Hongsong Wang↗

With the rapid advancement of video generation models, distinguishing between AI-generated and authentic videos has emerged as a challenging endeavor. The majority of existing research endeavors concentrate on the development of detectors for identifying samples generated by generative adversarial networks. Nevertheless, the detection of AI-generated videos, particularly those produced by text-to-video models, still remains an uncharted territory. Although state-of-the-art text-to-video models can generate realistic visual content similar to real videos, they fall short of generating the details of the images and the changes in details within the videos. Inspired by this, we address AI-generated video detection from a novel perspective of bit-planes, which can effectively describe the details or noises in images or videos. To this end, we propose a simple yet effective approach called Noise Amplification. This approach first extracts noise signals based on bit-planes, then amplifies these noise signals, and finally feeds them into the discriminator networks for video fake classification. Noise amplification is comprehensively constructed by incorporating three aspects: pixel-level intensity enhancement, region-level spatial amplification, and frame-level temporal aggregation. To evaluate methods of AI-generated video detection in challenging scenarios, we also introduce a benchmark named HardGVD. Extensive experiments on both the large-scale dataset GenVidBench and HardGVD show that our simple approach significantly outperforms state-of-the-art methods.

阅读与讨论 → 访问原文 →

10.

arXiv (CS.LG) 2026-06-19 DOI: arXiv:2605.05481

Approximate Next Policy Sampling: Replacing Conservative Target Policy Updates in Deep RL

作者:

Dillon Sandhu↗Ronald Parr↗

arXiv:2605.05481v2 Announce Type: replace Abstract: We revisit a classic "chicken-and-egg" problem in reinforcement learning: to safely improve a policy, the value function must be accurate on the state-visitation distribution of the updated policy. That distribution over states is unknown and cannot be sampled for the purposes of training the value function. Conservative updates solve this problem, but at the cost of shrinking the policy update. This paper explores an alternative solution, Approximate Next Policy Sampling (ANPS), which addresses the problem by modifying the training distribution rather than constraining the policy update. ANPS is satisfied if the distribution of the training data approximates that of the next policy. To demonstrate the feasibility and efficacy of ANPS, we introduce Stable Value Approximate Policy Iteration (SV-API). SV-API modifies the standard approximate policy iteration loop to hold the target policy fixed while an iteratively updated behavioral policy gathers relevant experience. It only commits to a new policy once a convergence criterion has been met. If certain stability criteria are met, the update is guaranteed to be safe; otherwise, it remains no less safe than standard approximate policy iteration. Applying SV-API to PPO yields Stable Value PPO (SV-PPO), which matches or improves performance on high-dimensional discrete (Atari) and continuous control benchmarks while executing substantially larger target policy updates. These results demonstrate the viability of ANPS as a new solution to this classic challenge in RL.

阅读与讨论 → 访问原文 →

11.

arXiv (CS.AI) 2026-06-16 DOI: arXiv:2606.16731

MuVAP: Multimodal Multiparty Voice Activity Projection for Turn-taking Prediction in the Wild

作者:

Haotian Qi↗Gabriel Skantze↗

arXiv:2606.16731v1 Announce Type: cross Abstract: Current multiparty turn-taking models often rely on complex microphone arrays or multi-camera setups, limiting their applicability in human-robot interaction scenarios. We introduce MuVAP, a causal multimodal framework that extends Voice Activity Projection by grounding acoustic predictions in face tracks, enabling speaker-aware turn-taking predictions from a monaural audio stream and a single camera view. To address the combinatorial complexity of modeling multiple speakers, we propose Role-Relative Projection, which maps any N-speaker interaction onto a fixed current versus next floor-holder state. Because existing audiovisual datasets contain disruptive editing cuts that break causal tracking, we introduce the Audio-Visual Conversation Corpus, a 31-hour dataset of unedited, single-camera multiparty conversations. Evaluations demonstrate that MuVAP outperforms strong baselines on Shift-Hold and next-speaker prediction tasks across two- and three-speaker settings.

阅读与讨论 → 访问原文 →

12.

arXiv (CS.AI) 2026-06-19 DOI: arXiv:2507.19137

Assessment of Personality Dimensions Across Situations in Dyadic Role-Play Scenarios

作者:

Alice Zhang↗Skanda Muralidhar↗Daniel Gatica-Perez↗Mathew Magimai-Doss↗

arXiv:2507.19137v3 Announce Type: replace-cross Abstract: Prior research indicates that users prefer assistive technologies whose personalities align with their own. This has sparked interest in automatic personality perception (APP), which aims to predict an individual's perceived personality traits. Previous studies in APP have treated personalities as static traits, independent of context. However, perceived personalities can vary by context and situation as shown in psychological research. In this study, we investigate the relationship between conversational speech and perceived personality for participants engaged in two work situations (a neutral interview and a stressful client interaction). Our key findings are: 1) perceived personalities differ significantly across interactions, 2) loudness, sound level, and spectral flux features are indicative of perceived extraversion, agreeableness, conscientiousness, and openness in neutral interactions, while neuroticism correlates with these features in stressful contexts, 3) handcrafted acoustic features and non-verbal features outperform speaker embeddings in inference of perceived personality, and 4) stressful interactions are more predictive of neuroticism, aligning with existing psychological research.

阅读与讨论 → 访问原文 →

13.

arXiv (CS.LG) 2026-06-16 DOI: arXiv:2512.12737

Communication-Efficient Neural Tangent Kernels for Heterogeneous Decentralized Federated Learning

作者:

Li Xia↗

arXiv:2512.12737v2 Announce Type: replace Abstract: Decentralized federated learning (DFL) enables collaborative model training without a central server, but converges slowly under statistical heterogeneity. Recent work has shown that neural tangent kernel (NTK) methods achieve faster convergence than gradient-based updates in DFL, while momentum has proven effective for accelerating gradient-based FL. However, applying momentum to NTK updates can destabilize training under heterogeneous data. We propose SPARK, which addresses this instability with a stage-wise annealed soft-label regularizer evaluated on neighborhood-aggregated data, so that momentum can accelerate NTK updates stably. Under high heterogeneity, SPARK converges about 3$\times$ faster than baselines and lowers the total communication to a target accuracy by up to about 70\%, and it attains higher accuracy across heterogeneity levels. We further study random projection as an optional Jacobian-compression strategy for bandwidth-constrained settings. We validate the approach across multiple datasets, network topologies, and heterogeneity levels.

阅读与讨论 → 访问原文 →

14.

arXiv (CS.AI) 2026-06-16 DOI: arXiv:2606.15768

LaWAM: Latent World Action Models for Efficient Dynamics-Aware Robot Policies

作者:

Jialei Chen↗Kai Wang↗Kang Chen↗Shuaihang Chen↗Feng Gao↗Wenhao Tang↗Zhiyuan Li↗Weilin Liu↗Zhuyu Yao↗Boxun Li↗Yuanbo Xu↗Chao Yu↗…

arXiv:2606.15768v1 Announce Type: cross Abstract: Vision-Language-Action models (VLAs) leverage large-scale vision-language pretraining for semantic robot control, but often lack explicit foresight into how robot actions change the scene. World-Action Models (WAMs) address this limitation by conditioning policies on predicted futures, yet existing approaches typically rely on computationally expensive video generation with substantial pixel-level redundancy. We present LaWAM, a Latent World Action Model that exposes predictive dynamics to robot policies through compact latent visual subgoals instead of reconstructed future video. At the core of LaWAM is a latent-action-conditioned Latent World Model (LaWM). We obtain LaWM by training a latent action model in the latent space of a pretrained vision foundation model and repurposing its forward decoder to predict future observation features for scene evolution. LaWAM then conditions action generation on these predicted latent visual subgoals to enable dynamics-aware robot control. LaWAM achieves state-of-the-art or competitive success rates (SRs) across LIBERO (98.6% SR), RoboTwin (91.22% SR), and real-world manipulation tasks while retaining low-latency inference. LaWAM runs in 187 ms per action-chunk prediction and achieves up to 24x lower wall-clock latency than pixel-space WAMs.

阅读与讨论 → 访问原文 →

15.

arXiv (CS.AI) 2026-06-11 DOI: arXiv:2606.11831

From Uniform to Learned Graph Priors: Diffusion for Structure Discovery

作者:

Qi Shao↗Hao Guo↗Jiawen Chen↗Duxin Chen↗Wenwu Yu↗

arXiv:2606.11831v1 Announce Type: cross Abstract: Neural relational inference (NRI) methods discover interaction graphs from trajectories through variational reasoning on discrete potential edges. However, these methods typically rely on oversimplified, factorized graph priors. Such priors, typically nearing uniform distributions, treat edges as independent entities. This systemic misalignment does not match the real-world systems and yields diffuse and indecisive edge posteriors limiting the reliability of structural discovery. To address this, we propose Diff-prior, a diffusion-parameterized adaptive prior used to calibrate latent graph distribution rather than generate graphs. Our core insight is to reframe prior integration as a learnable denoising-style calibration that organizes scattered, uncertain edge posteriors into a more reliable overall structure which can be trained by the diffusion model. Diff-prior learns an adaptive structure prior that performs structured calibration on the edge posteriors during inference, guiding it towards a distribution closer to the underlying structure. The diff-prior operates before structural sampling and acts as a denoising calibrator directly on the encoder edge distribution, which provides a generic training paradigm over structured variables. Experiments on standard benchmarks validated our framework, and the results indicate that Diff-prior improves the performance of structure inference and generates more decisive edge posteriors across multiple NRI-family architectures. The code is available on https://github.com/Hardy158118/Diffprior.

阅读与讨论 → 访问原文 →

16.

arXiv (CS.CV) 2026-06-16 DOI: arXiv:2606.16103

SceneCraft: Interactive System for Image Editing via Scene Graph

作者:

Duc-Manh Phan↗Ngoc-Dai Tran↗Duy-Khang Do↗Tam V. Nguyen↗Minh-Triet Tran↗Trung-Nghia Le↗

Recent advances in generative AI have enabled natural language-driven image editing, yet existing systems often fail in complex scenes with multiple interacting objects because they rely heavily on users crafting precise text prompts. To address the absence of structured control, we propose SceneCraft, a novel interactive framework that bridges user intent and model execution by representing images as editable scene graphs. Instead of guessing text prompts through trial and error, users interact directly with a visual graph to perform complex spatial and relational operations. These graph modifications are automatically translated into precise, context-aware editing prompts, effectively eliminating linguistic ambiguity. To ensure robust and diverse results, structured prompts are dispatched to multiple state-of-the-art generative models. Evaluations across diverse editing scenarios show that SceneCraft provides a more intuitive control mechanism, significantly reducing the cognitive burden of manual prompt engineering while generating outputs that users consistently rate as higher in quality and fidelity.

阅读与讨论 → 访问原文 →

17.

arXiv (CS.AI) 2026-06-18 DOI: arXiv:2606.18810

Learning from Own Solutions: Self-Conditioned Credit Assignment for Reinforcement Learning with Verifiable Rewards

作者:

Yingyu Shan↗Yuhang Guo↗Zihao Cheng↗Zeming Liu↗Xiangrong Zhu↗Xinyi Wang↗Jiashu Yao↗Wei Lin↗Hongru Wang↗Heyan Huang↗

arXiv:2606.18810v1 Announce Type: cross Abstract: Reinforcement learning with verifiable rewards (RLVR) has driven substantial progress in training LLMs for reasoning tasks, but representative methods such as GRPO assign uniform credit across all tokens, wasting gradient on routine tokens while under-crediting pivotal reasoning steps. Existing token-level credit assignment methods require resources beyond the model's own rollouts. GRPO variants rely on process reward models or ground-truth answers. Knowledge distillation assigns credit through per-token divergence but requires external teachers (On-Policy Distillation) or privileged information (On-Policy Self Distillation). However, these dependencies limit applicability in the pure RLVR setting. We observe that conditioning the model on its own verified trajectories induces a measurable per-token KL divergence between the original and conditioned distributions, and prove that distilling from a self-teacher constructed by verified trajectories leads to infeasible weighted-average solutions when multiple verified trajectories exist. We propose SC-GRPO (Self-Conditioned GRPO), which uses KL divergence mentioned before as a multiplicative weight on GRPO gradients. Across five benchmarks spanning math, code, and agentic tasks, SC-GRPO consistently outperforms 8.1% over GRPO and 5.9% over DAPO with stronger OOD performance. Moreover, SC-GRPO achieves higher performance than OPD.

阅读与讨论 → 访问原文 →

18.

arXiv (CS.AI) 2026-06-18 DOI: arXiv:2506.14126

From Memorization to Parameter Interference: How Overtraining Experts Harms Model Merging

作者:

Stefan Horoi↗Guy Wolf↗Eugene Belilovsky↗Gintare Karolina Dziugaite↗

arXiv:2506.14126v2 Announce Type: replace-cross Abstract: Modern deep learning is increasingly characterized by the use of open-weight foundation models that can be fine-tuned on specialized datasets. This has led to a proliferation of expert models and adapters, often shared via platforms like HuggingFace and AdapterHub. Model merging has recently emerged as an effective way to leverage these existing resources, enabling the composition of capabilities from different model checkpoints. A natural pipeline has thus formed to harness the benefits of transfer learning and amortize sunk training costs: models are pre-trained on general data, fine-tuned on specific tasks, and then multiple checkpoints are merged to obtain a more capable model. A prevailing assumption is that improvements at one stage of this pipeline propagate downstream, leading to gains at subsequent steps. In this work, we challenge that assumption by examining how expert fine-tuning affects model merging. We show that long fine-tuning of experts that optimizes for their individual performance leads to degraded merging performance across vision and language modalities, multiple model scales, and both fully fine-tuned and LoRA-adapted models. We trace this degradation to the memorization of a small set of difficult examples that dominate late fine-tuning steps. This causes negative parameter interference and encodes knowledge that is forgotten during merging. Finally, we demonstrate that task-dependent aggressive early stopping strategies can significantly improve model merging performance.

阅读与讨论 → 访问原文 →

19.

arXiv (CS.LG) 2026-06-16 DOI: arXiv:2606.16154

A Gradient Perspective on RLVR Stability and Winner Advantage Policy Optimization

作者:

Prasanth YSS↗Zhichen Ren↗Rasa Hosseinzadeh↗Ilan Gofman↗Yuqi Chen↗Zhaoyan Liu↗Guangwei Yu↗Jesse C. Cresswell↗Satya Krishna Gorti↗

arXiv:2606.16154v1 Announce Type: new Abstract: Reinforcement learning with verifiable rewards (RLVR) improves language-model reasoning, but GRPO-style optimization remains prone to collapse. We analyse this instability through token-level gradient dynamics, deriving a taxonomy that predicts how updates affect next-token probabilities and entropy. The taxonomy shows that stability depends jointly on the advantage sign and token distribution under the current policy. Motivated by this finding, we propose Winner Advantage Policy Optimization (WAPO), a simple online clipped policy-gradient objective that updates only on positive-advantage completions. Across mathematical reasoning and multi-hop QA benchmarks, WAPO improves training stability and matches or outperforms baselines across multiple model families. Full code can be found at https://github.com/layer6ai-labs/wapo.

阅读与讨论 → 访问原文 →

20.

arXiv (CS.AI) 2026-06-12 DOI: arXiv:2606.12703

SMSR: Certified Defence Against Runtime Memory Poisoning in Persistent LLM Agent Systems

作者:

Tarun Sharma↗

arXiv:2606.12703v1 Announce Type: cross Abstract: Retrieval-augmented generation (RAG) agents increasingly run with persistent memory that accumulates across user sessions. This creates a new attack surface: an adversary interacting only through normal channels can inject crafted memories that, once retrieved, steer the agent's responses for future users, without touching model weights or code. We call this Multi-Session Memory Poisoning (MSMP) and show that no existing defence certifies against it; static-corpus defences (RobustRAG, ReliabilityRAG) assume a fixed knowledge base, and heuristic filters are bypassed by fluent enterprise-style text. We present Signed Memory with Smoothed Retrieval (SMSR), the first defence with a certified robustness bound for this setting. Component 1 adds HMAC-SHA256 provenance at write time, blocking unsigned injection. Component 2 applies randomised memory ablation with verdict-based majority voting at query time, bounding the influence of authenticated adversaries. We prove that no provenance-free retrieval-time filter can certify against adaptive injection, derive a hypergeometric certificate for Component 2, and formalise the Consistent Minority Effect, whereby a consistent adversarial answer wins string-based voting as a numerical minority while verdict-based voting removes it. Across 15 enterprise scenarios (3,150 repeated trials), Component 1 cuts attack success from 93-100% to 0% for all unsigned variants. For an authenticated adversary with a single injection, Component 2 holds success to 8.0% (95% CI [5.8, 10.9], n=450), below the certified worst case. In an end-to-end query-only attack where the agent itself writes the poison rather than it being pre-seeded, SMSR reduces success from 65.3% to 5.3% (n=150, non-overlapping CIs) on a live agent stack. Clean-query utility is 90% (Component 1) and 85% (combined).

阅读与讨论 → 访问原文 →

21.

arXiv (CS.CL) 2026-06-16 DOI: arXiv:2606.15378

Rethinking the Role of Efficient Attention in Hybrid Architectures

作者:

Ziqing Qiao↗Yinuo Xu↗Chaojun Xiao↗Zhou Su↗Zihan Zhou↗Yingfa Chen↗Xiaoyue Xu↗Xu Han↗Zhiyuan Liu↗

Modern language models increasingly adopt hybrid architectures that combine full attention with efficient attention modules, such as sliding-window attention (SWA) and recurrent sequence mixers. However, how these efficient modules shape model capabilities remains poorly understood. To address this gap, we conduct a systematic analysis across hybrid architectures from three perspectives: scaling behavior, mechanism analysis, and architecture design. First, from a scaling perspective, we find that efficient-attention design primarily affects how fast long-context capability emerges, while different hybrids eventually converge to comparable long-context performance under sufficient training. Second, mechanistically, we show that long-range retrieval is mainly carried by full attention, whereas efficient attention shapes its optimization trajectory. This explains a counter-intuitive phenomenon we call Large-Window Laziness: larger SWA windows can delay the formation of retrieval heads in full-attention layers. Third, guided by this mechanism, we show that applying NoPE to only the full-attention layers of a small-window SWA hybrid substantially improves long-context performance with negligible impact on short-context performance.

阅读与讨论 → 访问原文 →

22.

arXiv (math.PR) 2026-06-17 DOI: arXiv:2606.17912

Asymptotics of the number of labelled connected sparse multitype graphs

作者:

Luisa Andreis↗Mario Veshaj↗

arXiv:2606.17912v1 Announce Type: cross Abstract: We study the asymptotic enumeration of labelled connected multitype graphs in the sparse regime, where both the number of vertices and edges grow linearly and the excess is proportional to the size of the graph. Extending the classical theory of connected graph enumeration to the multitype setting, we consider graphs with prescribed numbers of vertices of each type and prescribed edge counts between each pair of types. Our approach is probabilistic and relies on the theory of inhomogeneous random graphs. In particular, we exploit large-deviation principles and asymptotic estimates for connectedness probabilities to relate the counting problem to the emergence of giant components in suitably tuned supercritical random graphs. From large deviation asymptotics of connected components of inhomogeneous random graphs, we recognize that a connected graph with a given edge statistics corresponds to the (unique) giant component of larger inhomogeneous random graph with a suitably chosen connection kernel. This correspondence allows us to derive the leading exponential asymptotics for the number of connected multitype graphs with fixed type profile and edge matrix. The resulting formula generalizes the asymptotic enumeration results of Bender, Canfield, and McKay for connected sparse graphs to the multitype framework. More broadly, the paper illustrates how probabilistic techniques can provide transparent and effective tools for addressing new combinatorial enumeration problems.

阅读与讨论 → 访问原文 →

23.

bioRxiv (Bioinfo) 2026-06-12 DOI: HASH:95473ade798c97b3fafacf364f2b54e4

Computational Design of Optimal Sequences for Targeted Hypermutagenesis Using Recombination-Coupled Diversity-Generating Retroelements

作者:

Regnier↗Rochette↗Laurenceau↗Monasson↗Bikard↗Cocco↗

Diversity-generating retroelements (DGRs) are natural systems that accelerate evolution via targeted hypermutation at adenines. We previously developed DGRec, a system combining DGRs and recombineering for programmable mutagenesis in Escherichia coli. We here address two important issues with DGRec: the dependence of mutagenesis efficiency on the dgrRNA secondary structure and the variability of the reverse-transcription biases with sequence context and position. First, we introduce and validate a method to recode non-functional templates, i.e. with low mutagenesis efficiency, into highly functional ones through synonymous mutations. Second, we develop a Long Short-Term Memory (LSTM) model to predict DGRec mutational profiles for any given template sequence. By integrating this LSTM model with our recoding method, we establish a comprehensive workflow for customized directed evolution, enabling researchers to precisely fine-tune DGRec in vivo mutagenesis to their engineering needs.

阅读与讨论 → 访问原文 →

24.

arXiv (CS.CL) 2026-06-16 DOI: arXiv:2606.16137

XAI-Grounded Explanation Generation for Speech Deepfake Detection with Training-Free Multimodal Large Language Models

作者:

Yupei Li↗Qiyang Sun↗Xiaoliang Wu↗Chenxi Wang↗Berrak Sisman↗Bj\"orn W. Schuller↗

Speech deepfake detection (SDD) systems require trustworthy explanations for reliable decision-making. Existing explanation ways mainly fall into two categories. Traditional explainable AI (XAI), such as gradient-based attribution, produces low-level attribution signals tightly coupled with model decisions, and harder to be understood by human than natural language explanations. Meanwhile, large language model (LLM)-based explanation generation often produces generic and ungrounded descriptions due to the lack of heuristic evidence and task-specific supervision, stemming from limited grounded explanation datasets for SDD. We therefore propose a training-free explanation framework that integrates XAI evidence with multimodal LLMs to generate grounded and specific explanations. Using the PartialSpoof dataset, we construct a grounded explanation dataset and show that methods with XAI increase inside accuracy by over 45\%, verified through human evaluation and faithfulness checks.

阅读与讨论 → 访问原文 →

25.

arXiv (CS.LG) 2026-06-17 DOI: arXiv:2606.17319

Tight $L_\infty$ Sample Complexity for Low-Degree and Sparse Boolean Polynomials

作者:

Jasper van Doornmalen↗Mathieu Molina↗Victor Verdugo↗Jos\'e Verschae↗

arXiv:2606.17319v1 Announce Type: cross Abstract: Motivated by the optimization of bounded binary black-box functions, we study the problem of learning polynomial surrogates over the Boolean hypercube. To ensure that optimizing the surrogate yields good solutions for the underlying objective, we require uniform $L_\infty$-error guarantees rather than the usual $L_2$-type guarantees. We characterize the minimax sample complexity of uniform estimation under subgaussian noise for two classes of bounded polynomials. First, for polynomials of degree at most $d$ on $n$ variables, the sample complexity scales as $n^{d+1}$. Second, for $s$-sparse Fourier-Walsh polynomials with $s \leq n$, it scales as $ns^2$. These rates differ structurally from the noiseless setting, where uniform exact recovery scales as $n^d$ and $ns$, respectively. Our lower bounds hold even for arbitrary adaptive learners, showing that the additional factors are intrinsic to the noisy cases. Standard Fourier-analysis tools for the $L_2$-norm do not naturally extend to the $L_\infty$-setting in a way that yields uniform guarantees. Our proofs overcome this difficulty by relying on suitably chosen auxiliary norms that serve as proxies for controlling the $L_\infty$-error. Together, our results provide a tight characterization of the sample complexity of learning optimization-safe polynomial surrogates.

阅读与讨论 → 访问原文 →