# Research Papers — 2026-05 · bundle page 275 of 326 > Full-text markdown of papers #13701–13750 of 16266 published in 2026-05, newest first. AI-readable mirrors of the full 16266-paper 2026-05 archive (50 papers per file). Canonical page: https://aipod.dev/papers/archive/2026-05/page/275 [← Previous bundle](https://aipod.dev/papers/archive/2026-05/page/274.md) [Next bundle →](https://aipod.dev/papers/archive/2026-05/page/276.md) ### EA-WM: Event-Aware Generative World Model with Structured Kinematic-to-Visual Action Fields Authors: Zhaoyang Yang, Yurun Jin, Lizhe Qi, Cong Huang, Kai Chen Published: 2026-05-07 Categories: cs.CV, cs.AI, cs.RO arXiv: https://arxiv.org/abs/2605.06192v1 Pretrained video diffusion models provide powerful spatiotemporal generative priors, making them a natural foundation for robotic world models. While recent world-action models jointly optimize future videos and actions, they predominantly treat video generation as an auxiliary representation for policy learning. Consequently, they insufficiently explore the inverse problem: leveraging action signals to guide video synthesis, thereby often failing to preserve precise robot spatial geometry and fine-grained robot-object interaction dynamics in the generated rollouts. To bridge this gap, we present EA-WM, an Event-Aware Generative World Model that effectively closes the loop between kinematic control and visual perception. Rather than injecting joint or end-effector actions as abstract, low-dimensional tokens, EA-WM projects actions and kinematic states directly into the target camera view as Structured Kinematic-to-Visual Action Fields. To fully exploit this geometrically grounded representation, we introduce event-aware bidirectional fusion blocks that modulate cross-branch attention, capturing object state changes and interaction dynamics. Evaluated on the comprehensive WorldArena benchmark, EA-WM achieves state-of-the-art performance, outperforming existing baselines by a significant margin. --- ### Systematic Evaluation of Large Language Models for Post-Discharge Clinical Action Extraction Authors: Shivali Dalmia, Ananya Mantravadi, Prasanna Desikan Published: 2026-05-07 Categories: cs.AI arXiv: https://arxiv.org/abs/2605.06191v1 The work in this paper evaluates zero-shot and few-shot large language models (LLMs) for safety-critical clinical action extraction using the CLIP discharge-note dataset, with particular emphasis on transitions of care and post-discharge patient safety. To manage the complexity of clinical documentation, we introduce a two-stage extraction framework that decomposes discharge notes, that are written in narrative form, into fine-grained, explicitly actionable clinical tasks through a staged prompting strategy. Our contributions include a systematic assessment of generative LLMs for clinical action extraction, a detailed comparison between general-purpose LLMs and task-specific supervised BERT-based models, and an analysis of annotation inconsistencies across different action categories. We show that contemporary LLMs achieve performance comparable to or exceeding supervised models on binary actionability detection, while supervised baselines retain a meaningful advantage on fine-grained multi-label category classification, despite the absence of task-specific fine-tuning and under strict data-privacy constraints. Qualitative error analysis reveals that many failures stem from misalignment between model reasoning and dataset annotation conventions, particularly in cases involving implicit clinical actions and rigid structural labeling rules. These results indicate that reported performance reflects model limitations due to lack of clinical reasoning, that is not captured by plain annotations. Labels without rationales make it impossible to distinguish clinical reasoning failures from annotation convention mismatches. Advancing clinical NLP requires reasoning-annotated datasets that document why specific spans are actionable, not merely which spans were labeled, enabling proper evaluation of model clinical understanding. --- ### Constrained Contextual Bandits with Adversarial Contexts Authors: Dhruv Sarkar, Abhishek Sinha Published: 2026-05-07 Categories: cs.LG arXiv: https://arxiv.org/abs/2605.06190v1 We study budget-constrained contextual bandits with adversarial contexts, where each action yields a random reward and incurs a random cost. We adopt the standard realizability assumption: conditioned on the observed context, rewards and costs are drawn independently from fixed distributions whose expectations belong to known function classes. We focus on the continuing setting, in which the algorithm operates over the entire horizon even after the budget for cumulative cost is exhausted. In this setting, the objective is to simultaneously control regret and the violation of the budget constraint. Building on the seminal $\mathsf{SquareCB}$ framework of Foster et al. (2018), we propose a simple and modular framework that leverages online regression oracles to reduce the constrained problem to a standard unconstrained contextual bandit problem with adaptively defined surrogate reward functions. In contrast to prior works, which focus on stochastic contexts, our reduction yields improved guarantees for more general adversarial contexts, together with an efficient algorithm with a compact and transparent analysis. --- ### Predictive-Generative Drift Decomposition for Speech Enhancement and Separation Authors: Julius Richter, Yoshiki Masuyama, Christoph Boeddeker, Takahiro Edo, Gordon Wichern, Jonathan Le Roux Published: 2026-05-07 Categories: eess.AS, cs.LG arXiv: https://arxiv.org/abs/2605.06189v1 We propose a plug-and-play framework for speech enhancement and separation that augments predictive methods with a generative speech prior. Our approach, termed Stochastic Interpolant Prior for Speech (SIPS), builds on stochastic interpolants and leverages their flexibility to bridge predictive and generative modeling. Specifically, we decompose the interpolation dynamics into a task-specific drift and a stochastic denoising component, allowing a predictive estimate to be integrated directly into the generative sampling process. This results in a mathematically grounded framework for combining strong pretrained predictors with the expressive power of generative models. To this end, we train a score model using only clean speech, yielding a degradation-agnostic prior that can be reused across tasks. During inference, the predictor provides a deterministic drift that steers the sampling process toward a task-consistent estimate, while the score model preserves perceptual naturalness. Unlike prior hybrid approaches, which typically rely on architecture-specific conditioning and are tied to particular predictors or degradation settings, SIPS provides a unified framework that generalizes across predictors and additive degradation tasks. We demonstrate its effectiveness for both speech enhancement and speech separation using recent predictors such as SEMamba and FlexIO. The proposed method consistently improves perceptual quality, achieving gains up +1.0 NISQA for speech separation. --- ### OPSD Compresses What RLVR Teaches: A Post-RL Compaction Stage for Reasoning Models Authors: Jaehoon Kim, Dongha Lee Published: 2026-05-07 Categories: cs.AI, cs.CL arXiv: https://arxiv.org/abs/2605.06188v1 On-Policy Self-Distillation (OPSD) has recently emerged as an alternative to Reinforcement Learning with Verifiable Rewards (RLVR), promising higher accuracy and shorter responses through token-level credit assignment from a self-teacher conditioned on privileged context. However, this promise does not carry over to thinking-enabled mathematical reasoning, where reported accuracy gains shrink and sometimes turn negative. We hypothesize that hindsight supervision can specify better token-level alternatives in short thinking-disabled outputs, but in long thinking-enabled traces it more readily identifies redundancy than supplies better replacements. To test this, we applied OPSD separately to correct and incorrect rollout groups, so that compression and correction can be observed in isolation. Our results show that in thinking-enabled mathematical reasoning, OPSD behaves most reliably as a compression mechanism rather than a correction mechanism: training only on correct rollouts preserves accuracy while substantially shortening responses, whereas training only on incorrect rollouts damages accuracy. In light of these findings, we propose a revised post-training pipeline for thinking-enabled mathematical reasoning: SFT then RLVR then OPSD. --- ### In-Context Black-Box Optimization with Unreliable Feedback Authors: Nicolas Samuel Blumer, Julien Martinelli, Samuel Kaski Published: 2026-05-07 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2605.06187v1 Black-box optimization in science and engineering often comes with side information: experts, simulators, pretrained predictors, or heuristics can suggest which candidates look promising. This information can accelerate search, but it can also be biased, input-dependent, or misleading. Feedback-aware BO methods typically handle one task at a time, limiting their ability to generalize over multiple sources of feedback. In-context optimizers address cross-task adaptation, but usually assume that optimization history is the only available signal at test time. We study feedback-informed in-context black-box optimization (FICBO), where a pretrained optimizer conditions on both the observed history and cheap auxiliary feedback for the current candidate set. We introduce a structured feedback prior that models how feedback sources vary in their access, relevance, and distortion relative to the true objective, and use it to pretrain a feedback-aware transformer. At test time, the model estimates source reliability in context by comparing observed objective values with auxiliary signals, improving query selection. On synthetic and real-world tasks, FICBO effectively exploits informative feedback while remaining robust to weak or misleading sources, improving over other baselines. Empirical investigations further illustrate how the model perceives test-time sources, offering insights into its interpretability and decision-making process. --- ### Event-Causal RAG: A Retrieval-Augmented Generation Framework for Long Video Reasoning in Complex Scenarios Authors: Peizheng Yan, Yu Zhao, Liang Xie, Juntong Qi, Mingming Wang, Erwei Yin Published: 2026-05-07 Categories: cs.AI, cs.CV arXiv: https://arxiv.org/abs/2605.06185v1 Recent large vision-language models have achieved strong performance on short- and medium-length video understanding, yet they remain inadequate for ultra-long or even infinite video reasoning, where models must preserve coherent memory over extended durations and infer causal dependencies across temporally distant events. Existing end-to-end video understanding methods are fundamentally limited by the $O(n^2)$ complexity of self-attention, while recent retrieval-augmented generation (RAG) approaches still suffer from fragmented clip-level memory, weak modeling of temporal and causal structure, and high storage and online inference costs. We present Event-Causal RAG, a lightweight retrieval-augmented framework for infinite long-video reasoning. Instead of indexing fixed-length clips, our method segments streaming videos into semantically coherent events and represents each event as a structured State-Event-State (SES) graph, capturing the event together with its surrounding state transitions. These graphs are merged into a global Event Knowledge Graph and stored in a dual-store memory that supports both semantic matching and causal-topological retrieval. On top of this memory, we design a bidirectional retrieval strategy to efficiently identify the most relevant event causal chains and provide them, together with the associated video evidence, to a backbone video foundation model for answer generation. Experiments on long-video understanding benchmarks demonstrate that Event-Causal RAG consistently outperforms strong clip-based retrieval baselines and long-context video models, particularly on questions requiring multi-event integration and causal inference across long temporal gaps, while also achieving improved memory efficiency and robust streaming performance. --- ### Teaching LLMs Program Semantics via Symbolic Execution Traces Authors: Jonas Bayer, Stefan Zetzsche, Olivier Bouissou, Remi Delmas, Michael Tautschnig, Soonho Kong Published: 2026-05-07 Categories: cs.SE, cs.LG, cs.LO, cs.PL arXiv: https://arxiv.org/abs/2605.06184v1 We introduce an evaluation framework of 500 C verification tasks across five property types (memory safety, overflow, termination, reachability, data races) built on SV-COMP 2025, and evaluate 14 models across six families. We find that high overall accuracy masks a critical weakness: while most models reliably confirm properties hold, violation detection varies widely and degrades sharply with program length. To close this gap, we train on formal verification artifacts: running the Soteria symbolic execution engine on generic open-source C code and using the resulting traces for continued pretraining of Qwen3-8B. Just ${\sim}$3,000 bug traces combined with chain-of-thought reasoning at inference time improve violation detection by over 17 percentage points, producing one of the most balanced accuracy profiles among evaluated models. On violation detection, the trained 8B model outperforms the 4$\times$ larger Qwen3-32B without thinking and approaches it in overall accuracy. The interaction between trace training and chain-of-thought is superadditive: neither alone provides meaningful gains, but their combination does. Improvements transfer across all five property types, including ones the training traces do not target. Our 28 configurations confirm the gains stem from trace semantics, not code volume, and that trace curation and format matter. --- ### Rethinking Adapter Placement: A Dominant Adaptation Module Perspective Authors: Suoxin Zhang, Run He, Di Fang, Xiang Tan, Kaixuan Chen, Huiping Zhuang Published: 2026-05-07 Categories: cs.AI, cs.CL, cs.LG arXiv: https://arxiv.org/abs/2605.06183v1 Low-rank adaptation (LoRA) is a widely used parameter-efficient fine-tuning method that places trainable low-rank adapters into frozen pre-trained models. Recent studies show that using fewer LoRA adapters may still maintain or even improve performance, but existing methods still distribute adapters broadly, leaving where to place a limited number of adapters to maximize performance largely open. To investigate this, we introduce PAGE (Projected Adapter Gradient Energy), a gradient-based sensitivity probe that estimates the initial trainable gradient energy available to each candidate LoRA adapter. Surprisingly, we find that PAGE is highly concentrated on a single shallow FFN down-projection across two model families and four downstream tasks. We term this module the dominant adaptation module and show that its layer index is architecture-dependent but task-stable. Motivated by this finding, we propose DomLoRA, a placement method that places a single adapter at the dominant adaptation module. With only ~0.7% of vanilla LoRA's trainable parameters, DomLoRA outperforms it on average across various downstream tasks, including instruction following, mathematical reasoning, code generation, and multi-turn conversation. This method also improves other LoRA variants, supporting the dominant adaptation module perspective as a practical placement guideline. --- ### SuperFace: Preference-Aligned Facial Expression Estimation Beyond Pseudo Supervision Authors: Zejian Kang, Xuanyang Xu, Wentao Yang, Kai Zheng, Yuanchen Fei, Hongyuan Zou, Hui Shan, Shuo Yang, Xiangru Huang Published: 2026-05-07 Categories: cs.CV arXiv: https://arxiv.org/abs/2605.06179v1 Accurate facial estimation is crucial for realistic digital human animation, and ARKit blendshape coefficients offer an interpretable representation by mapping facial motions to semantic animation controls. However, learning high-quality ARKit coefficient prediction remains limited by the absence of reliable ground-truth supervision. Existing methods typically rely on capture software such as Live Link Face to provide pseudo labels, which may contain noisy activations, biased coefficient magnitudes, and missing or inaccurate facial actions. Consequently, models trained with supervised learning tend to reproduce imperfect pseudo labels rather than optimize for perceptual expression fidelity. In this paper, we propose SuperFace, a preference-driven framework that moves ARKit facial expression estimation from pseudo-label imitation toward human-aligned perceptual optimization. Instead of treating software-estimated coefficients as fixed ground truth, SuperFace uses them only as an initialization and further improves coefficient prediction through human preference feedback on rendered facial expressions. By aligning the model with perceptual judgments rather than numerical pseudo labels, SuperFace enables more visually faithful and expressive facial animation. Experiments show that SuperFace improves expression fidelity over Live Link Face supervision, demonstrating the effectiveness of preference-driven optimization for semantic facial action prediction. --- ### BioMedArena: An Open-source Toolkit for Building and Evaluating Biomedical Deep Research Agents Authors: Jinge Wu, Hongjian Zhou, Mingde Zeng, Jiayuan Zhu, Junde Wu, Jiazhen Pan, Ayush Noori, Sean Wu, Honghan Wu, Fenglin Liu, David A. Clifton Published: 2026-05-07 Categories: cs.AI arXiv: https://arxiv.org/abs/2605.06177v2 Reproducing and comparing deep research agents today is hard: the same backbone evaluated on the same benchmark can report different accuracies across papers because the harness and tool registry differ, and integrating a new model into a comparable evaluation surface costs weeks of model-specific engineering. These are symptoms of a broader reproducibility problem in deep research agent research. Here, we introduce BioMedArena, an open-source toolkit that addresses this reproducibility gap and provides an arena for comparing deep research agents under a shared evaluation environment. BioMedArena decouples six layers of biomedical agent evaluation -- benchmark loading, tool exposure, tool selection, harness mode, context management, and scoring -- and exposes 166 biomedical benchmarks and 75 biomedical tools across 9 functional families. Adding a new model, benchmark, or tool can be accomplished with a few-line provider adapter. Beyond evaluation infrastructure, BioMedArena ships a library of high-quality reference components: 6 agent harnesses (including our proposed Mutual-Evolve) and 6 context-management strategies, any of which can be equipped on any backbone. Equipping these components substantially improves all 12 backbones; on each of 8 representative biomedical benchmarks, the best equipped backbone surpasses prior state-of-the-art (SOTA), by 15.01 percentage points on average. The toolkit, configurations, and per-task traces are available at https://github.com/AI-in-Health/BioMedArena. --- ### Retina-RAG: Retrieval-Augmented Vision-Language Modeling for Joint Retinal Diagnosis and Clinical Report Generation Authors: Abdelrahman Zaian, Sheethal Bhat, Mohamed Abdalkader, Andreas Maier Published: 2026-05-07 Categories: cs.CV, cs.AI arXiv: https://arxiv.org/abs/2605.06173v2 Diabetic Retinopathy (DR) is a leading cause of preventable blindness among working-age adults worldwide, yet most automated screening systems are limited to image-level classification and lack clinically structured reporting. We propose Retina-RAG, a low-cost modular framework that jointly performs DR severity grading, macular edema (ME) detection, and report generation. The architecture decouples a high-performance retinal classifier and a parameter-efficient vision-language model (Qwen2.5-VL-7B-Instruct) adapted via Low-Rank Adaptation (LoRA), enabling flexible component integration. A retrieval-augmented generation (RAG) module injects curated ophthalmic knowledge together with structured classifier outputs at inference time to improve diagnostic consistency and reduce hallucinations. Retina-RAG achieves an F1-score of 0.731 for DR grading and 0.948 for ME detection, substantially outperforming zero-shot Qwen (0.096, 0.732) and MMed-RAG (0.541, 0.641) on a retinal disease detection dataset with captions. For report generation, Retina-RAG attains ROUGE-L 0.438 and SBERT similarity 0.884, exceeding all baselines. The full framework operates on a single consumer-grade GPU, demonstrating that clinically structured retinal AI can be achieved with modest computational resources. --- ### Expressivity of Bi-Lipschitz Normalizing Flows: A Score-Based Diffusion Perspective Authors: Meira Iske, Carola-Bibiane Schönlieb Published: 2026-05-07 Categories: stat.ML, cs.LG, math.NA, math.PR arXiv: https://arxiv.org/abs/2605.06172v1 Many normalizing flow architectures impose regularity constraints, yet their distributional approximation properties are not fully characterized. We study the expressivity of bi-Lipschitz normalizing flows through the lens of score-based diffusion models. For the probability flow ODE of a variance-preserving diffusion, Lipschitz regularity of the score induces a flow of bi-Lipschitz diffeomorphic transport maps. This ODE bridge allows us to analyze the distributional approximation power of bi-Lipschitz normalizing flows and, conversely, derive deterministic convergence guarantees for diffusion-based transport. Our key idea is to use the probability flow ODE to link regularity of the score to regularity of the induced transport maps. We verify score regularity for broad target densities, including compactly supported densities, Gaussian convolutions of compactly supported measures and finite Gaussian mixtures. We obtain a universal distributional approximation result: Gaussian pullbacks induced by bi-Lipschitz variance-preserving transport maps are $L^1$-dense among all probability densities. For Gaussian convolution targets, we further obtain convergence in Kullback-Leibler divergence without early stopping. --- ### DynT2I-Eval: A Dynamic Evaluation Framework for Text-to-Image Models Authors: Juntong Wang, Jiarui Wang, Huiyu Duan, Lewei Li, Guangtao Zhai, Xiongkuo Min Published: 2026-05-07 Categories: cs.CV arXiv: https://arxiv.org/abs/2605.06170v1 Existing text-to-image (T2I) benchmarks largely rely on fixed prompt sets, leaving them vulnerable to overfitting and benchmark contamination once publicly released and repeatedly reused. In this work, we propose DynT2I-Eval, a fully automated dynamic evaluation framework for T2I models. It constructs a structured visual semantic space from long-form descriptions, decomposing prompts into controllable dimensions (e.g., subject, logical constraint, environment, and composition). This enables the continuous generation of fresh prompts via task-specific spaces and difficulty-aware sampling. DynT2I-Eval evaluates model performance across text alignment, perceptual quality, and aesthetics. Heterogeneous outputs are unified into prompt-conditioned pairwise comparisons, allowing a dynamic scheduler, micro-batch aggregation, and weighted Bayesian updates to maintain a stable online leaderboard despite changing prompt distributions and model injection. Experiments with independently sampled prompt streams demonstrate that continually refreshed prompts provide a robust evaluation protocol, reducing the impact of prompt-set-specific tuning. Simulations and ablations further confirm that the proposed ranking framework achieves a strong balance among cold-start convergence, late-entry discovery, and long-run ranking fidelity. --- ### Mean Mode Screaming: Mean--Variance Split Residuals for 1000-Layer Diffusion Transformers Authors: Pengqi Lu Published: 2026-05-07 Categories: cs.LG, cs.CV arXiv: https://arxiv.org/abs/2605.06169v1 Scaling Diffusion Transformers (DiTs) to hundreds of layers introduces a structural vulnerability: networks can enter a silent, mean-dominated collapse state that homogenizes token representations and suppresses centered variation. Through mechanistic auditing, we isolate the trigger event of this collapse as Mean Mode Screaming (MMS). MMS can occur even when training appears stable, with a mean-coherent backward shock on residual writers that opens deep residual branches and drives the network into a mean-dominated state. We show this behavior is driven by an exact decomposition of these gradients into mean-coherent and centered components, compounded by the structural suppression of attention-logit gradients through the null space of the Softmax Jacobian once values homogenize. To address this, we propose Mean-Variance Split (MV-Split) Residuals, which combine a separately gained centered residual update with a leaky trunk-mean replacement. On a 400-layer single-stream DiT, MV-Split prevents the divergent collapse that crashes the un-stabilized baseline; it tracks close to the baseline's pre-crash trajectory while remaining substantially better than token-isotropic gating methods such as LayerScale across the full schedule. Finally, we present a 1000-layer DiT as a scale-validation run at boundary scales, establishing that the architecture remains stably trainable at extreme depth. --- ### One Algorithm, Two Goals: Dual Scoring for Parameter and Data Selection in LLM Fine-Tuning Authors: Xinrui Chen, Liu Yang, Ou Wu Published: 2026-05-07 Categories: cs.LG arXiv: https://arxiv.org/abs/2605.06166v1 In Large Language Model (LLM) fine-tuning, parameter and data selection are common strategies for reducing fine-tuning cost, yet they are typically driven by separate scoring mechanisms. When a parameter mask and data subset jointly determine restricted fine-tuning, this separation incurs redundant overhead and makes coordinated selection difficult. We cast parameter and data selection as two bilevel selection problems under a common validation objective and derive a shared local response-surrogate scoring rule. Under first- and second-order validation-improvement approximations, parameter importance and data utility emerge as column-wise and row-wise aggregations of a single gradient interaction matrix, yielding a closed-form row-column correspondence for co-extracting both signals. Building on this structure, we propose DualSFT (Dual-Selection Fine-Tuning), a one-shot dual-scoring algorithm that produces a parameter mask and data subset from shared gradient statistics. On 3B-9B LLMs, single-axis DualSFT variants strengthen target-task performance and stability-plasticity trade-offs within their comparison groups, while full DualSFT yields a more favorable joint-constrained trade-off than sequential hybrid baselines under matched budgets. --- ### Post Reasoning: Improving the Performance of Non-Thinking Models at No Cost Authors: Richmond Sin Jing Xuan, Rishabh Bhardwaj, Soujanya Poria Published: 2026-05-07 Categories: cs.AI arXiv: https://arxiv.org/abs/2605.06165v1 As the widespread adoption of Large Language Models (LLMs) accelerates, token consumption from intermediate reasoning traces increasingly contributes to inference latency and operational cost. Recent studies suggest that many real-world tasks require little to no explicit reasoning, with additional reasoning sometimes even degrading performance. In this work, we propose \textbf{Post-Reasoning}, a simple yet effective approach that improves instruction-tuned models by conditioning them to justify their answers after generating the final response. By design, it enables the final answer to be obtained without additional latency or token cost, while still improving performance through simple instruction augmentation. We evaluate Post-Reasoning across \(117\) model--benchmark settings spanning \(13\) open and proprietary models, \(4\) model families, and \(9\) diverse reasoning and knowledge-intensive benchmarks, including AMC, HMMT, GSM8K, GPQA, MMLU-Pro, and BIG-Bench Hard. Post-Reasoning improves performance in over \(88.19\%\) of evaluated settings, achieving a mean relative improvements of \(17.37\%\). Furthermore, we propose supervised post-reason tuning, which further improves performance in over \(91.11\%\) of evaluated settings, and exceeds the prompt-based post-reasoning baseline by an average of \(8.01\%\), demonstrating that post-reasoning can be effectively internalized through training. Ultimately, Post-Reasoning establishes a new performance ceiling for direct-answer capabilities. --- ### Beyond Accuracy: Policy Invariance as a Reliability Test for LLM Safety Judges Authors: Shihao Weng, Yang Feng, Xiaofei Xie Published: 2026-05-07 Categories: cs.AI, cs.SE arXiv: https://arxiv.org/abs/2605.06161v1 LLM-as-a-Judge pipelines have become the de facto evaluator for agent safety, yet existing benchmarks treat their verdicts as ground-truth proxies without checking whether the verdicts depend on the agent's behavior or merely on how the evaluation policy happens to be worded. We argue that any trustworthy safety judge must satisfy a basic property we call policy invariance, and we operationalize it as three testable principles: rubric-semantics invariance under certified-equivalent rewrites, rubric-threshold invariance under intentional strict-to-lenient shifts, and ambiguity-aware calibration so that verdict instability concentrates on genuinely ambiguous cases. Instantiating these principles as a stress-test protocol with four agent-class judges on trajectories drawn from ASSEBench and R-Judge, we surface a previously unmeasured failure mode: today's judges respond to meaningful normative shifts and to meaningless structural rewrites with comparable strength, and cannot tell the two apart. Content-preserving policy rewrites flip up to 9.1% of verdicts above baseline jitter, and 18-43% of all observed flips occur on unambiguous cases under such rewrites, so existing safety scores conflate what the agent did with how the evaluator was prompted. Beyond the diagnosis, we contribute the Policy Invariance Score and the Judge Card reporting protocol, which expose an order-of-magnitude spread in judge reliability that is invisible to accuracy-only leaderboards. We release the protocol and code so that future agent-safety benchmarks can audit their own evaluators rather than trust them by default. --- ### Beyond Forgetting in Continual Medical Image Segmentation: A Comprehensive Benchmark Study Authors: Bomin Wang, Hangqi Zhou, Yibo Gao, Xiahai Zhuang Published: 2026-05-07 Categories: cs.CV arXiv: https://arxiv.org/abs/2605.06160v1 Continual learning (CL) is essential for deploying medical image segmentation models in clinical environments where imaging domains, anatomical targets, and diagnostic tasks evolve over time. However, continual segmentation still faces three main challenges. First, the scenarios for this task remain insufficiently standardized for real-world clinical settings. Second, existing research has been primarily focused on mitigating forgetting, overlooking the other essential properties such as plasticity. Third, a benchmark work with comprehensive evaluation on existing methods is stll desirable. To address these gaps, we present such benchmark study of continual medical image segmentation. We first define three clinically motivated scenarios, namely Domain-CL, Class-CL, and Organ-CL, to respectively capture the cross-center domain shift, the incremental anatomical structure segmentation, and the cross-organ segmentation. We then introduce an evaluation framework that measures not only general performance and forgetting, but also plasticity, forward generalizability, parameter efficiency, and replay burden. The results, from extensive experiments with representative CL methods, showed that it was still challenging to develop a model that could satisfy all the requirements simultaneously. Nevertheless, these studies also suggested that the replay-based methods achieve the best overall balance between stability and plasticity, the parameter-isolation methods should be effective at reducing forgetting, though at the cost of increased model size, and the forward generalizability remain a significantly understudied aspect of this research field. Finally, we discuss related learning paradigms and outline future directions for continual medical image segmentation. --- ### Entropy-Regularized Adjoint Matching for Offline Reinforcement Learning Authors: Abdelghani Ghanem, Mounir Ghogho Published: 2026-05-07 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2605.06156v2 Integrating expressive generative policies, such as flow-matching models, into offline reinforcement learning (RL) allows agents to capture complex, multi-modal behaviors. While Q-learning with Adjoint Matching (QAM) stabilizes policy optimization via the continuous adjoint method, it remains inherently bound to the fixed behavior distribution. This dependence induces a \textit{popularity bias} that can suppress high-reward actions in low-density regions, and creates a \textit{support binding} that restricts off-manifold exploration. Existing workarounds, such as appending \textit{residual} Gaussian policies, often re-introduce the expressivity bottlenecks associated with unimodal distributions. In this work, we propose \textit{Maximum Entropy Adjoint Matching} (ME-AM), a unified framework that addresses these limitations within the continuous flow formulation. ME-AM incorporates two mechanisms: (1) a Mirror Descent entropy maximization objective that mitigates the popularity bias to facilitate the extraction of optimal policies from offline datasets, and (2) a \textit{Mixture Behavior Prior} that broadens the geometric support to encompass out-of-distribution high-reward regions. By exploring this extended geometry, ME-AM identifies robust actions while preserving the absolute continuity of the generative vector field. Empirically, ME-AM demonstrates competitive or superior performance compared to prior state-of-the-art (SOTA) methods across a diverse suite of sparse-reward continuous control environments. --- ### Graphlets as Building Blocks for Structural Vocabulary in Knowledge Graph Foundation Models Authors: Kossi Amouzouvi, Robert Wardenga, Jens Lehmann, Sahar Vahdati Published: 2026-05-07 Categories: cs.AI, cs.LG arXiv: https://arxiv.org/abs/2605.06154v1 Foundation models excel at language, where sentences become tokens, and vision, where images become pixels, because both reduce to discrete symbols on a shared, fixed grid. Knowledge Graphs share the discreteness, but not the geometry. Their entities and relations are discrete symbols, yet their arrangement is relational and lacks a common, fixed grid. Knowledge Graphs (KGs) share the discreteness, but not the geometry. They form irregular, non-Euclidean topologies whose local neighborhoods differ from graph to graph. Therefore, Knowledge Graph Foundation Models (KGFMs) rely on identifying structural invariances to produce transferable representations. Without a universal token set, KGFMs are limited in their ability to transfer representations across unseen KGs. We close this gap by treating graphlets, small connected graphs, as structural tokens that recur in heterogeneous KGs. In this paper, We introduce a model-agnostic framework based on a vocabulary of graphlets that mines a KG between relations via pattern matching. In particular, we considered closed and open 2- and 3-path, and star graphlets, to obtain robust invariances. The framework is evaluated on 51 KGs from a wide range of domains, for zero-shot inductive and transductive link prediction. Experiments show that adding simple graphlets to the vocabulary yields models that outperform prior KGFMs. --- ### Secure Seed-Based Multi-bit Watermarking for Diffusion Models from First Principles Authors: Enoal Gesny, Eva Giboulot Published: 2026-05-07 Categories: cs.CR, cs.CV arXiv: https://arxiv.org/abs/2605.06153v1 The rapid emergence of generative image models has led to the development of specialized watermarking techniques, particularly in-generation methods such as seed-based embedding. However, current evaluations in this area remain largely empirical, making them heavily reliant on the specific model architectures used for generation and inversion. This prevents any clear conclusion on the performance of any method, especially regarding security, for which a rigorous definition is lacking. Against this approach, we argue that the effectiveness of a watermarking scheme should be established purely through a thorough theoretical analysis. This is enabled by decoupling the model-dependent part from the actual decision mechanism of the watermarking system. Using this decoupling, we introduce a formal evaluation framework based on security, robustness, and fidelity. This allows precise comparisons between watermarking systems through a characteristic surface representing the trade-off between these three quantities, independent of any generative model. Based on this framework, we propose SSB, a novel watermarking method that generalizes previous seed-based methods by allowing to reach any security-robustness-fidelity regime on its characteristic surface. This work opens the door to the design of modern watermarking systems with theoretical guarantees that do not necessitate any costly empirical evaluations. --- ### Grokking or Glitching? How Low-Precision Drives Slingshot Loss Spikes Authors: Liu Hanqing, Jianjun Cao, Yuanze Li, Zijian Zhou Published: 2026-05-07 Categories: cs.LG, cs.CL, math.OC, stat.ML arXiv: https://arxiv.org/abs/2605.06152v3 Deep neural networks exhibit periodic loss spikes during unregularized long-term training, a phenomenon known as the "Slingshot Mechanism." Existing work usually attributes this to intrinsic optimization dynamics, but its triggering mechanism remains unclear. This paper proves that this phenomenon is a result of floating-point arithmetic precision limits. As training enters a high-confidence stage, the difference between the correct-class logit and the other logits may exceed the absorption-error threshold. Then during backpropagation, the gradient of the correct class is rounded exactly to zero, while the gradients of the incorrect classes remain nonzero. This breaks the zero-sum constraint of gradients across classes and introduces a systematic drift in the parameter update of the classifier layer. We prove that this drift forms a positive feedback loop with the feature, causing the global classifier mean and the global feature mean to grow exponentially. We call this mechanism Numerical Feature Inflation (NFI). This mechanism explains the rapid norm growth before a Slingshot spike, the subsequent reappearance of gradients, and the resulting loss spike. We further show that NFI is not equivalent to an observed loss spike: in more practical tasks, partial absorption may not produce visible spikes, but it can still break the zero-sum constraint and drive rapid growth of parameter norms. Our results reinterpret Slingshot as a numerical dynamic of finite-precision training, and provide a testable explanation for abnormal parameter growth and logit divergence in late-stage training. --- ### AdaGamma: State-Dependent Discounting for Temporal Adaptation in Reinforcement Learning Authors: Yaomin Wang, Jianting Pan, Ran Tian, Xiaoyang Li, Yu Zhang, Hengle Qin, Tianshu YU Published: 2026-05-07 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2605.06149v1 The discount factor in reinforcement learning controls both the effective planning horizon and the strength of bootstrapping, yet most deep RL methods use a single fixed value across all states. While state-dependent discounting is conceptually appealing, naive deep actor--critic implementations can become unstable and degenerate toward TD-error collapse. We propose AdaGamma, a practical deep actor--critic method for state-dependent discounting that learns a state-dependent discount function together with a return-consistency objective to regularize the induced backup structure. On the theory side, we analyze the Bellman operator induced by state-dependent discounting and establish its basic well-posedness properties under suitable conditions. Empirically, AdaGamma integrates into both SAC and PPO, yielding consistent improvements on continuous-control benchmarks, and achieves statistically significant gains in an online A/B test on the JD Logistics platform. These results suggest that state-dependent discounting can be made effective in deep RL when coupled with a return-consistency objective that prevents degenerate target manipulation. --- ### Learning Discrete Autoregressive Priors with Wasserstein Gradient Flow Authors: Bowen Zheng, Yihong Luo, Tianyang Hu Published: 2026-05-07 Categories: cs.CV, cs.AI, cs.LG arXiv: https://arxiv.org/abs/2605.06148v1 Discrete image tokenizers are commonly trained in two stages: first for reconstruction, and then with a prior model fitted to the frozen token sequences. This decoupling leaves the tokenizer unaware of the model that will later generate its tokens. As a result, the learned tokens may preserve image information well but still be difficult for an autoregressive (AR) prior to predict from left to right. We analyze this mismatch using Tripartite Variational Consistency (TVC), which decomposes latent-variable learning into three consistency conditions: conditional-likelihood consistency, prior consistency, and posterior consistency. TVC shows that two-stage training preserves the reconstruction side but leaves prior consistency outside the tokenizer objective: the overall token distribution is fixed before the AR prior participates in training. Motivated by this view, we add a distribution-level prior-matching signal during tokenizer training, while keeping the reconstruction objective unchanged. We optimize this signal with a Wasserstein-gradient-flow update. For hard categorical tokens, the update reduces to a token-level contrast between an auxiliary AR model that tracks the tokenizer's current token distribution and the target AR prior. It requires only forward passes through the two AR models and does not backpropagate through either of them. The resulting tokenizer, wAR-Tok, reduces AR loss and improves generation FID on CIFAR-10 and ImageNet at comparable reconstruction quality. --- ### Unifying Goal-Conditioned RL and Unsupervised Skill Learning via Control-Maximization Authors: Alireza Modirshanechi, Benjamin Eysenbach, Peter Dayan, Eric Schulz Published: 2026-05-07 Categories: cs.LG, cs.AI, eess.SY arXiv: https://arxiv.org/abs/2605.06145v1 Unsupervised pretraining has driven empirical advances in goal-conditioned reinforcement learning (GCRL), but its theoretical foundations remain poorly understood. In particular, an influential class of methods, mutual information skill learning (MISL), discovers behaviorally diverse skills that can later be used for downstream goal-reaching. However, it remains a theoretical mystery why skills learned through MISL should support goal-reaching. A subtle challenge is that both GCRL and MISL are umbrella terms: different GCRL tasks use distinct criteria for measuring goal-reaching performance, while different MISL methods optimize distinct notions of behavioral diversity. We address this challenge and unify GCRL and MISL as instances of control maximization. We identify three canonical GCRL formulations and prove that they are fundamentally inequivalent: they can induce incompatible optimal policies even in the same environment. Nevertheless, they all share a common interpretation: a well-performing goal-conditioned policy is one whose future trajectory is highly sensitive to the commanded goal, with the precise notion of sensitivity determined by the GCRL formulation. Noting that MISL objectives can be understood as measures of skill-sensitivity akin to goal-sensitivity, we show that MISL objectives are bounded by formulation-specific downstream goal-sensitivities. These bounds establish a precise correspondence between MISL methods and downstream GCRL tasks: for every GCRL formulation, there exists a matching MISL objective for which more diverse skills afford greater downstream goal sensitivity. Our results thus lay a theoretical foundation for RL pretraining and have important practical implications, such as suggesting which pretraining objectives to use when a user cares about a specific class of downstream tasks. --- ### AI-Generated Images: What Humans and Machines See When They Look at the Same Image Authors: Silvia Poletti, Justin Ilyes, Marcel Hasenbalg, David Fischinger, Martin Boyer Published: 2026-05-07 Categories: cs.CV, cs.AI arXiv: https://arxiv.org/abs/2605.06143v1 The misuse of generative AI in online disinformation campaigns highlights the urgent need for transparent and explainable detection systems. In this work, we investigate how detectors for AI-generated images can be more effective in providing human-understandable explanations for their predictions. To this end, we develop a suite of detectors with various architectures and fine-tuning strategies, trained on our large-scale photorealistic fake image dataset, AIText2Image, and assess their performance on state-of-the-art text-to-image AI generators. We integrate 16 different explainable AI (XAI) methods into our detection framework, and the visual explanations are comprehensively refined and evaluated through a novel approach that prioritizes human understanding of AI-generated images, using both textual and visual responses collected from a survey of 100 participants. This framework offers insights into visual-language cues in fake image detection and into the clarity of XAI methods from a human perspective, measuring the alignment of XAI outputs with human preferences. --- ### IRC-Bench: Recognizing Entities from Contextual Cues in First-Person Reminiscences Authors: Yehudit Aperstein, Eden Moran, Alexander Apartsin Published: 2026-05-07 Categories: cs.CL, cs.AI arXiv: https://arxiv.org/abs/2605.06142v2 When people recount personal memories, they often refer to people, places, and events indirectly, relying on con-textual cues rather than explicit names. Such implicit references are central to reminiscence narratives: first-person accounts of lived experience used in therapeutic, archival, and social settings. They pose a difficult computational problem because the intended entity must be inferred from dispersed narrative evidence rather than from a local mention. We introduce IRC-Bench, the Implicit Reminiscence Context Benchmark, for evaluating implicit entity recognition in reminiscence transcripts. The benchmark targets non-locality: entity-identifying cues are distributed across multiple, non-contiguous clauses, unlike named entity recognition, entity linking, or coreference resolution. IRC-Bench comprises 25,136 samples constructed from 12,337 Wikidata-linked entities across 1,994 transcripts spanning 11 thematic domains. Each sample pairs an Entity-Grounded Narrative, in which the target entity is ex-plicitly mentioned, with an Entity-Elided Narrative, in which direct mentions are removed. We evaluate 19 config-urations across LLM generation, dense retrieval, RAG, and fine-tuning. QLoRA-adapted Llama 3.1 8B performs best in the open-world setting (38.94% exact match; 51.59% Jaccard), while fine-tuned DPR leads closed-world retrieval (35.38% Hit@1; 71.49% Hit@10). We release IRC-Bench with data, code, and evaluation tools. --- ### Matrix-Valued Optimism is Matrix-Valued Augmentation: Additive Hybrid Designs for Constrained Optimization Authors: Jiayi Zhao Published: 2026-05-07 Categories: cs.LG arXiv: https://arxiv.org/abs/2605.06141v1 Augmented Lagrangian and optimistic primal--dual methods stabilize equality-constrained optimization through seemingly different mechanisms: the former adds constraint-dependent primal curvature, while the latter adds dual memory. Recent work has shown that these mechanisms are equivalent for scalar parameters. We extend this equivalence to matrix-valued correction. We prove an additivity principle: for symmetric matrix parameters, the ideal primal trajectory depends only on the summed correction matrix, not on how it is split between augmented and optimistic channels. This exposes a design freedom: algebraically equivalent decompositions can have different finite-step feasibility because augmented correction affects primal curvature, whereas optimistic correction affects the scale of the dual memory correction. We formulate the resulting step-size-limited design problem and derive a closed-form hybrid rule that selects a matrix correction, splits it between the two channels, and chooses primal and dual steps using local spectral weights. Experiments on nonlinear equality-constrained problems with controlled constraint-Jacobian conditioning show that the hybrid design improves over pure augmented and pure optimistic endpoints, closely tracks a grid-search hybrid oracle, and is competitive with first-order primal--dual baselines under mild-to-moderate ill-conditioning. The experiments also identify the expected limitation: exact cancellation requires increasingly large matrix corrections as the constraint Jacobian becomes ill-conditioned. --- ### SymDrift: One-Shot Generative Modeling under Symmetries Authors: Samir Darouich, Vinh Tong, Lluís Pastor-Pérez, Tanja Bien, Loay Mualem, Mathias Niepert Published: 2026-05-07 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2605.06140v1 Generative modeling of physical systems, such as molecules, requires learning distributions that are invariant under global symmetries, such as rotations in three-dimensional space. Equivariant diffusion and flow matching models can incorporate such invariances effectively, even when trained on a non-invariant empirical distribution, but they typically rely on costly multi-step sampling. Recently, drifting models have emerged as an efficient alternative, enabling single-step generation and achieving state-of-the-art performance in generative modeling tasks. However, we show that drifting models face a symmetry-specific challenge, since an equivariant generator does not generally produce the same drifting field as the one obtained from the symmetrized target distribution. Addressing this issue would require expensive symmetrization of the empirical distribution. To avoid this cost, we propose SymDrift, a framework that makes the drifting field itself symmetry-aware. We introduce two complementary strategies: (i) a symmetrized drift in coordinate space based on optimal alignment, and (ii) a $G$-invariant embedding that removes symmetry ambiguity by construction. Empirically, SymDrift outperforms existing one-shot methods on standard benchmarks for conformer and transition state generation, while remaining competitive with significantly more expensive multi-step approaches. By enabling one-shot inference, SymDrift reduces computational overhead by up to 40$\times$ compared to existing baselines, making it promising for high-throughput applications such as virtual drug screening and large-scale reaction network exploration. --- ### Listwise Policy Optimization: Group-based RLVR as Target-Projection on the LLM Response Simplex Authors: Yun Qu, Qi Wang, Yixiu Mao, Heming Zou, Yuhang Jiang, Yingyue Li, Wutong Xu, Lizhou Cai, Weijie Liu, Clive Bai, Kai Yang, Yangkun Chen et al. Published: 2026-05-07 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2605.06139v2 Reinforcement learning with verifiable rewards (RLVR) has become a standard approach for large language models (LLMs) post-training to incentivize reasoning capacity. Among existing recipes, group-based policy gradient is prevalent, which samples a group of responses per prompt and updates the policy via group-relative advantage signals. This work reveals that these optimization strategies share a common geometric structure: each implicitly defines a target distribution on the response simplex and projects toward it via first-order approximation. Building on this insight, we propose Listwise Policy Optimization (LPO) to explicitly conduct the target-projection, which demystifies the implicit target by restricting the proximal RL objective to the response simplex, and then projects the policy via exact divergence minimization. This framework provides (i) monotonic improvement on the listwise objective with bounded, zero-sum, and self-correcting projection gradients, and (ii) flexibility in divergence selection with distinct structural properties through the decoupled projection step. On diverse reasoning tasks and LLM backbones, LPO consistently improves training performance over typical policy gradient baselines under matched targets, while intrinsically preserving optimization stability and response diversity. --- ### Autoregressive Visual Generation Needs a Prologue Authors: Bowen Zheng, Weijian Luo, Guang Yang, Colin Zhang, Tianyang Hu Published: 2026-05-07 Categories: cs.CV, cs.AI, cs.LG arXiv: https://arxiv.org/abs/2605.06137v2 In this work, we propose Prologue, an approach to bridging the reconstruction-generation gap in autoregressive (AR) image generation. Instead of modifying visual tokens to satisfy both reconstruction and generation, Prologue generates a small set of prologue tokens prepended to the visual token sequence. These prologue tokens are trained exclusively with the AR cross-entropy (CE) loss, while visual tokens remain dedicated to reconstruction. This decoupled design lets us optimize generation through the AR model's true distribution without affecting reconstruction quality, which we further formalize from an ELBO perspective. On ImageNet 256x256, Prologue-Base reduces gFID from 21.01 to 10.75 without classifier-free guidance while keeping reconstruction almost unchanged; Prologue-Large reaches a competitive rFID of 0.99 and gFID of 1.46 using a standard AR model without auxiliary semantic supervision. Interestingly, driven only by AR gradients, prologue tokens exhibit emergent semantic structure: linear probing on 16 prologue tokens reaches 35.88% Top-1, far above the 23.71% of the first 16 tokens from a standard tokenizer; resampling with fixed prologue tokens preserves a similar high-level semantic layout. Our results suggest a new direction: generation quality can be improved by introducing a separate learned generative representation while leaving the original representation intact. --- ### BUILD-AND-FIND: An Effort-Aware Protocol for Evaluating Agent-Managed Codebases Authors: Jhen-Ke Lin Published: 2026-05-07 Categories: cs.SE, cs.AI arXiv: https://arxiv.org/abs/2605.06136v1 Most coding-agent benchmarks ask whether generated code behaves correctly. That remains essential, but repository-level engineering is increasingly agent-managed: one agent writes a repository, and later agents inspect, audit, or extend it as working context. In that setting, a generated repository is not only an answer to a task but also a communication artifact for future work. Even when strong agents nearly satisfy the visible behavioral objective, repositories can differ in how clearly they expose the intended behavior and design choices behind that behavior. We introduce BUILD-AND-FIND, a protocol for evaluating whether downstream agents can recover those intended choices from generated repositories, and how much inspection that recovery requires. For each task, a builder sees a hidden repository specification and creates a codebase; a finder sees only the codebase and a specification-traced multiple-choice question bank. The protocol separates behavioral correctness from artifact-side recovery and reports recovery accuracy, repeatability, implementation coverage, and inspection effort. Accuracy and stability act as gates: effort is interpreted only when recovery succeeds reliably. Among artifacts from which the same intent can be recovered, lower effort by the same finder suggests that the artifact makes that intent easier to locate. Question-only and spec-only controls quantify generic priors and specification access, while audits separate omitted claims from finder failures and check whether correct answers cite artifact evidence. In the released high-prior task pack, recovery accuracy is near saturation, so inspection effort and finder-specific effects provide the main panel-local comparison. --- ### Diffusion model for SU(N) gauge theories Authors: Javad Komijani, Marina K. Marinkovic, Lara Turgut Published: 2026-05-07 Categories: hep-lat, cs.LG arXiv: https://arxiv.org/abs/2605.06134v1 Implicit score matching provides a computationally efficient approach for training diffusion models and generating high-quality samples from complex distributions. In this work, we develop a score-matching framework for SU(N) lattice gauge theories, which can be extended to other Lie groups. We apply the method to SU(3) gauge configurations with the Wilson gauge action in two and four dimensions and assess the quality of the generated samples by comparison with Hybrid Monte Carlo (HMC) simulations. We show that the diffusion models can be successfully trained and applied for sampling the Wilson gauge action. For large values of inverse coupling, accurate reverse-time integration requires predictor-corrector schemes, for which we introduce a corrector based on Hamiltonian molecular dynamics. While the corrector significantly improves sampling quality, it also increases the computational cost. We outline several strategies for improving sampling efficiency. --- ### MemReranker: Reasoning-Aware Reranking for Agent Memory Retrieval Authors: Chunyu Li, Mengyuan Zhang, Jingyi Kang, Ding Chen, Jiajun Shen, Bo Tang, Xuanhe Zhou, Feiyu Xiong, Zhiyu Li Published: 2026-05-07 Categories: cs.CL arXiv: https://arxiv.org/abs/2605.06132v2 In agent memory systems, the reranking model serves as the critical bridge connecting user queries with long-term memory. Most systems adopt the "retrieve-then-rerank" two-stage paradigm, but generic reranking models rely on semantic similarity matching and lack genuine reasoning capabilities, leading to a problem where recalled results are semantically highly relevant yet do not contain the key information needed to answer the question. This deficiency manifests in memory scenarios as three specific problems. First, relevance scores are miscalibrated, making threshold-based filtering difficult. Second, ranking degrades when facing temporal constraints, causal reasoning, and other complex queries. Third, the model cannot leverage dialogue context for semantic disambiguation. This report introduces MemReranker, a reranking model family (0.6B/4B) built on Qwen3-Reranker through multi-stage LLM knowledge distillation. Multi-teacher pairwise comparisons generate calibrated soft labels, BCE pointwise distillation establishes well-distributed scores, and InfoNCE contrastive learning enhances hard-sample discrimination. Training data combines general corpora with memory-specific multi-turn dialogue data covering temporal constraints, causal reasoning, and coreference resolution. On the memory retrieval benchmark, MemReranker-0.6B substantially outperforms BGE-Reranker and matches open-source 4B/8B models as well as GPT-4o-mini on key metrics. MemReranker-4B further achieves 0.737 MAP, with several metrics on par with Gemini-3-Flash, while maintaining inference latency at only 10--20% of large models. On finance and healthcare vertical-domain benchmarks, the models preserve generalization capabilities on par with mainstream large-parameter rerankers. --- ### Skill1: Unified Evolution of Skill-Augmented Agents via Reinforcement Learning Authors: Yaorui Shi, Yuxin Chen, Zhengxi Lu, Yuchun Miao, Shugui Liu, Qi GU, Xunliang Cai, Xiang Wang, An Zhang Published: 2026-05-07 Categories: cs.AI arXiv: https://arxiv.org/abs/2605.06130v3 A persistent skill library allows language model agents to reuse successful strategies across tasks. Maintaining such a library requires three coupled capabilities. The agent selects a relevant skill, utilizes it during execution, and distills new skills from experience. Existing methods optimize these capabilities in isolation or with separate reward sources, resulting in partial and conflicting evolution. We propose Skill1, a framework that trains a single policy to co-evolve skill selection, utilization, and distillation toward a shared task-outcome objective. The policy generates a query to search the skill library, re-ranks candidates to select one, solves the task conditioned on it, and distills a new skill from the trajectory. All learning derives from a single task-outcome signal. Its low-frequency trend credits selection and its high-frequency variation credits distillation. Experiments on ALFWorld and WebShop show that Skill1 outperforms prior skill-based and reinforcement learning baselines. Training dynamics confirm the co-evolution of the three capabilities, and ablations show that removing any credit signal degrades the evolution. --- ### Continuous Expert Assembly: Instance-Conditioned Low-Rank Residuals for All-in-One Image Restoration Authors: Haisen He, Xiangyu Zou, SongLin Dong, Heng Li, Yihong Gong, Zhiheng Ma Published: 2026-05-07 Categories: cs.CV, cs.AI arXiv: https://arxiv.org/abs/2605.06127v1 Real-world image degradation is often unknown, spatially non-uniform, and compositional, requiring all-in-one restoration models to adapt a single set of weights to diverse local corruption patterns without test-time degradation labels. Existing methods typically modulate a shared backbone with global prompts or degradation descriptors, or route features through predefined expert pools. However, compact global conditioning can bottleneck localized degradation evidence, while static expert routing may produce homogeneous updates or rely on unstable sparse assignments. We propose \textbf{Continuous Expert Assembly} (CEA), a token-wise dynamic parameterization framework for all-in-one image restoration. CEA employs a lightweight \textbf{Cross-Attention Hyper-Adapter} to probe intermediate spatial features and synthesize instance-conditioned low-rank routing bases and residual directions. Each spatial token then assembles its own residual update via dense signed dot-product affinities over the generated rank-wise components, avoiding external prompts, static expert banks, and discrete Top- selection. The resulting assembly rule also admits a linear-attention perspective, making its dense token-wise routing behavior transparent. Experiments on AIO-3, AIO-5, and CDD-11 show that CEA improves average restoration quality over strong prompt-, descriptor-, and expert-based baselines, with the clearest gains on spatially varying and compositional degradations, while maintaining favorable parameter, FLOP, and runtime efficiency. --- ### P-Guide: Parameter-Efficient Prior Steering for Single-Pass CFG Inference Authors: Xin Peng, Ang Gao Published: 2026-05-07 Categories: cs.AI arXiv: https://arxiv.org/abs/2605.06124v1 Classifier-Free Guidance (CFG) is essential for high-fidelity conditional generation in flow matching, yet it imposes significant computational overhead by requiring dual forward passes at each sampling step. In this work, we address this bottleneck by introducing \textbf{P-Guide}, a framework that achieves high-quality guidance through a single inference pass by modulating only the initial latent state. We further show that, under a first-order approximation, P-Guide is equivalent to CFG in the sense that it steers generation from the prior space, without requiring explicit velocity field extrapolation during sampling. We consider both homoscedastic and \textbf{heteroscedastic} priors, and find that jointly modeling the mean and variance enables adaptive loss attenuation and improved robustness to data uncertainty. Extensive experiments demonstrate that P-Guide reduces inference latency by approximately 50\% while maintaining fidelity and prompt alignment competitive with standard dual-pass CFG baselines. --- ### Back to the Beginning of Heuristic Design: Bridging Code and Knowledge with LLMs Authors: Nguyen Viet Tuan Kiet, Bui Dinh Pham, Dao Van Tung, Tran Cong Dao, Huynh Thi Thanh Binh Published: 2026-05-07 Categories: cs.AI arXiv: https://arxiv.org/abs/2605.06123v1 Large language models (LLMs) have recently advanced automatic heuristic design (AHD) for combinatorial optimization (CO), where candidate heuristics are iteratively proposed, evaluated, and refined. Most existing approaches search over executable programs and distill insights from execution feedback to guide later iterations. Because this process moves from low-level implementations to high-level principles, we refer to it as a bottom-up paradigm. We argue that this view is incomplete and introduce a complementary top-down perspective: knowledge becomes the primary search object and code merely instantiates and tests it, making what is learned explicit and reusable across problems and trajectories. We formalize this shift through a statistical-learning view that exposes a distortion--compression trade-off, and instantiate it in both population-based and tree-based AHD frameworks. Across CO and tasks beyond it, knowledge-first search improves discovery efficiency, transfer, and generalization, often outperforming code-centric pipelines, while combining both strategies yields further gains. Our results suggest that progress in AHD depends on iteratively constructing and evolving interpretable hypotheses that retain value beyond a single search trajectory. --- ### Pest-Thinker: Learning to Think and Reason like Entomologists via Reinforcement Learning Authors: Xueheng Li, Yu Wang, Tao Hu, Ji Huang, Ke Cao, Qize Yang, Rui Li, Jie Zhang, Chengjun Xie Published: 2026-05-07 Categories: cs.CV arXiv: https://arxiv.org/abs/2605.06121v1 Pest-induced crop losses pose a major threat to global food security and sustainable agricultural development. While recent advances in Multimodal Large Language Models (MLLMs) have shown strong potential for visual understanding and smart agriculture, their direct application to pest recognition remains limited due to the domain's unique challenges such as high inter-species complexity, intra-species variability, and the scarcity of expert-annotated data. In this work, we introduce Pest-Thinker, a knowledge-driven reinforcement learning (RL) framework that enables MLLMs to reason over fine-grained pest morphology. We first construct two high-definition pest benchmarks, QFSD and AgriInsect, comprising diverse species and expert-annotated morphological traits. Leveraging these datasets, we synthesize Chain-of-Thought (CoT) reasoning trajectories to facilitate structured learning of pest-specific visual cues through Supervised Fine-Tuning (SFT). Subsequently, we employ Group Relative Policy Optimization (GRPO) with a novel feature reward that guides the model to focus on observable morphological evidence, assessed by an LLM-as-a-Judge strategy. Extensive experiments demonstrate that Pest-Thinker substantially improves both in-domain and out-of-domain morphological understanding, marking a step toward expert-level visual reasoning for intelligent agricultural pest analysis. The datasets and source code are available upon acceptance. --- ### BoostLLM: Boosting-inspired LLM Fine-tuning for Few-shot Tabular Classification Authors: Yi-Siang Wang, Kuan-Yu Chen, Yu-Chen Den, Darby Tien-Hao Chang Published: 2026-05-07 Categories: cs.LG arXiv: https://arxiv.org/abs/2605.06117v2 Large language models (LLMs) have recently been adapted to tabular prediction by serializing structured features into natural language, but their performance in low-data regimes remains limited compared to gradient-boosted decision trees (GBDTs). In this work, we revisit the boosting paradigm, traditionally associated with tree ensembles, and ask whether it can be applied as a general training principle for LLM fine-tuning. We propose BoostLLM, a framework that transforms parameter-efficient fine-tuning into a multi-round residual optimization process by training sequential PEFT adapters as weak learners. To incorporate tabular inductive bias, BoostLLM integrates decision-tree paths as a second input view alongside raw features; analysis reveals that the path view acts as a structured teacher in early training steps before the model shifts toward feature-driven representations. Empirically, BoostLLM achieves consistent improvements over standard fine-tuning across multiple LLM backbones and datasets, matching or surpassing XGBoost across a wide range of shot counts and outperforming GPT-4o-based methods with a 4B model. We further show that the framework scales: pairing with stronger tree models and extended boosting horizons yields additional gains under appropriate stabilization. These results suggest that boosting can serve as a general training principle for LLM fine-tuning, particularly in low-data regimes for structured data. --- ### Policy-Guided Stepwise Model Routing for Cost-Effective Reasoning Authors: Wenwen Si, Insup Lee, Osbert Bastani Published: 2026-05-07 Categories: cs.AI arXiv: https://arxiv.org/abs/2605.06116v1 Inference-time computation has greatly enhanced the performance of large language models (LLMs) on challenging reasoning tasks, but this strategy can incur high inference costs. One solution is to route intermediate chain-of-thought (CoT) states to language models of different sizes; however, existing approaches rely on handcrafted routing strategies that limit performance, or on training large process reward models that may be infeasible in many applications. We formulate stepwise model routing as a constrained decision-making problem, which we solve by training a small control policy using reinforcement learning in conjunction with threshold calibration to tune the performance-efficiency tradeoff. We validate our method on three math benchmarks (GSM8K, MATH500, and OmniMath) on both open and closed models. Our method consistently improves the accuracy-cost tradeoff compared to handcrafted approaches, while achieving a comparable tradeoff to methods that require training large process reward models. --- ### CrossCult-KIBench: A Benchmark for Cross-Cultural Knowledge Insertion in MLLMs Authors: Zhen Zeng, Leijiang Gu, Feng Li, Jing Yu, Zenglin Shi Published: 2026-05-07 Categories: cs.AI arXiv: https://arxiv.org/abs/2605.06115v2 Multimodal Large Language Models (MLLMs), trained primarily on English-centric data, frequently generate culturally inappropriate or misaligned responses in cross-cultural settings. To mitigate this, we introduce the task of cross-cultural knowledge insertion, which focuses on adapting models to specific cultural contexts while preserving their original behavior in other cultures. To facilitate research in this area, we introduce CrossCult-KIBench, a comprehensive evaluation benchmark for assessing both the effectiveness of knowledge insertion and its unintended side effects on non-target cultures. The benchmark includes 9,800 image-grounded cases covering 49 culturally relevant visual scenarios across English, Chinese, and Arabic language-culture groups. It supports evaluation in both single-insert and sequential-insert settings. We also propose Memory-Conditioned Knowledge Insertion (MCKI) as a baseline method. MCKI retrieves relevant cultural knowledge from an external memory using frozen MLLM representations, prepending matched entries as conditional prompts when applicable. Extensive experiments on CrossCult-KIBench reveal that current approaches struggle to balance effective cultural adaptation with behavioral preservation, highlighting a key challenge in developing culturally-aware MLLMs. Our work thus underscores an important research direction for developing more culturally adaptive and responsible MLLMs. --- ### Dynamic Pondering Sparsity-aware Mixture-of-Experts Transformer for Event Stream based Visual Object Tracking Authors: Shiao Wang, Xiao Wang, Duoqing Yang, Wenhao Zhang, Bo Jiang, Lin Zhu, Yonghong Tian, Bin Luo Published: 2026-05-07 Categories: cs.CV, cs.AI arXiv: https://arxiv.org/abs/2605.06112v1 Despite significant progress, RGB-based trackers remain vulnerable to challenging imaging conditions, such as low illumination and fast motion. Event cameras offer a promising alternative by asynchronously capturing pixel-wise brightness changes, providing high dynamic range and high temporal resolution. However, existing event-based trackers often neglect the intrinsic spatial sparsity and temporal density of event data, while relying on a single fixed temporal-window sampling strategy that is suboptimal under varying motion dynamics. In this paper, we propose an event sparsity-aware tracking framework that explicitly models event-density variations across multiple temporal scales. Specifically, the proposed framework progressively injects sparse, medium-density, and dense event search regions into a three-stage Vision Transformer backbone, enabling hierarchical multi-density feature learning. Furthermore, we introduce a sparsity-aware Mixture-of-Experts module to encourage expert specialization under different sparsity patterns, and design a dynamic pondering strategy to adaptively adjust the inference depth according to tracking difficulty. Extensive experiments on FE240hz, COESOT, and EventVOT demonstrate that the proposed approach achieves a favorable trade-off between tracking accuracy and computational efficiency. The source code will be released on https://github.com/Event-AHU/OpenEvTracking. --- ### Schedule-and-Calibrate: Utility-Guided Multi-Task Reinforcement Learning for Code LLMs Authors: Yujia Chen, Yang Ye, Xiao Chu, Yuchi Ma, Cuiyun Gao Published: 2026-05-07 Categories: cs.SE, cs.AI arXiv: https://arxiv.org/abs/2605.06111v1 Reinforcement learning (RL) with verifiable rewards has proven effective at post-training LLMs for coding, yet deploying separate task-specific specialists incurs costs that scale with the number of tasks, motivating a unified multi-task RL (MTRL) approach. However, existing MTRL methods treat all coding tasks uniformly, relying on fixed data curricula under a shared optimization strategy, ultimately limiting the effectiveness of multi-task training. To address these limitations, we propose ASTOR, a multi-tASk code reinforcement learning framework via uTility-driven coORdination. Centered on task utility, a signal capturing each task learning potential and cross-task synergy, ASTOR comprises two coupled modules: 1) Hierarchical Utility-Routed Data Scheduling module hierarchically allocates training budget and prioritizes informative prompts, steering training toward the most valuable data and 2) Adaptive Utility-Calibrated Policy Optimization module dynamically scales per-task KL regularization, matching update constraints to each tasks current training state. Experiments on two widely-used LLMs across four representative coding tasks demonstrate that ASTOR consistently improves a single model across all tasks, outperforming the best task-specific specialist by 9.0%-9.5% and surpassing the strongest MTRL baseline by 7.5%-12.8%. --- ### On Time, Within Budget: Constraint-Driven Online Resource Allocation for Agentic Workflows Authors: Xinglin Wang, Zishen Liu, Shaoxiong Feng, Peiwen Yuan, Yiwei Li, Jiayi Shi, Yueqi Zhang, Chuyi Tan, Ji Zhang, Boyuan Pan, Yao Hu, Kan Li Published: 2026-05-07 Categories: cs.AI, cs.CL arXiv: https://arxiv.org/abs/2605.06110v2 Agentic systems increasingly solve complex user requests by executing orchestrated workflows, where subtasks are assigned to specialized models or tools and coordinated according to their dependencies. While recent work improves agent efficiency by optimizing the performance--cost--latency frontier, real deployments often impose concrete requirements: a workflow must be completed within a specified budget and before a specified deadline. This shifts the goal from average efficiency optimization to maximizing the probability that the entire workflow completes successfully under explicit budget and deadline constraints. We study \emph{constraint-driven online resource allocation for agentic workflows}. Given a dependency-structured workflow and estimates of success rates and generation lengths for each subtask--model pair, the executor dynamically allocates models and parallel samples across simultaneously executable subtasks while managing the remaining budget and time. We formulate this setting as a finite-horizon stochastic online allocation problem and propose \emph{Monte Carlo Portfolio Planning} (MCPP), a lightweight closed-loop planner that directly estimates constrained completion probability through simulated workflow executions and replans after observed outcomes. Experiments on CodeFlow and ProofFlow demonstrate that MCPP consistently improves constrained completion probability over strong baselines across a wide range of budget--deadline constraints. --- ### Shallow Prefill, Deep Decoding: Efficient Long-Context Inference via Layer-Asymmetric KV Visibility Authors: Jungsuk Oh, Hyeseo Jeon, Hyunjune Ji, Kyongmin Kong, Jay-Yoon Lee Published: 2026-05-07 Categories: cs.AI arXiv: https://arxiv.org/abs/2605.06105v1 Long-context inference in decoder-only language models is costly because long prompts are processed during Prefill, cached at every layer, and repeatedly attended to during autoregressive Decode. We introduce \emph{Shallow Prefill, dEEp Decode} (SPEED), a phase-asymmetric KV-visibility policy that materializes non-anchor prompt-token KV states only in lower layers while keeping Decode-phase tokens full-depth. Unlike previous approaches that make upper-layer prompt KV states cheaper to store or construct, SPEED removes prefill tokens from the upper-layer Decode visibility set altogether. With a minimal BoS anchor, this simple change preserves broad benchmark quality while reducing long-context cost. In a controlled Llama-3.1-8B instruction-tuning study, SPEED using only 75\% of layers for prefill tokens reaches 51.2 average score on OLMES-style benchmarks, compared with 51.4 for the full-depth baseline, while improving TTFT by 33\%, TPOT by 22\%, and reducing active KV memory by 25.0\% at 128K context. Layer-wise diagnostics suggest that this cutoff retains the main prompt-selection and representation-stabilization regions of the full-depth model. These results show that long-context prompt tokens need not always persist as full-depth KV-cache objects when Decode-phase tokens remain full-depth. --- ### Beyond Autoregressive RTG: Conditioning via Injection Outside Sequential Modeling in Decision Transformer Authors: Yongyi Wang, Hanyu Liu, Lingfeng Li, Bozhou Chen, Ang Li, Qirui Zheng, Xionghui Yang, Chucai Wang, Wenxin Li Published: 2026-05-07 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2605.06104v1 Decision Transformer (DT) formulates offline reinforcement learning as autoregressive sequence modeling, achieving promising results by predicting actions from a sequence of Return-to-Go (RTG), state, and action tokens. However, RTG is a scalar that summarizes future rewards, containing far less information than typical state or action vectors, yet it consumes the same computational budget per token. Worse, the self-attention cost of Transformers grows quadratically with sequence length, so including RTG as a separate token adds unnecessary overhead. We propose SlimDT, which removes RTG from the autoregressive sequence. Instead, we inject RTG information into the state representations before the sequential modeling step, allowing the Transformer to process only a compact (state, action) sequence. This reduces the sequence length by one-third, directly improving inference efficiency. On the D4RL benchmark, SlimDT surpasses standard DT across various tasks and achieves performance comparable to existing state-of-the-art methods. Decoupling a sparse conditioning signal from an information-rich sequence thus yields both computational gains and higher task performance. --- ### CredibleDFGO: Differentiable Factor Graph Optimization with Credibility Supervision Authors: Liang Qian, Penggao Yan, Penghui Xu, Li-Ta Hsu Published: 2026-05-07 Categories: eess.SP, cs.AI, cs.LG, cs.RO arXiv: https://arxiv.org/abs/2605.06100v2 Global navigation satellite system (GNSS) positioning is widely used for urban navigation, but the covariance reported by the GNSS solver is often unreliable in urban canyons. Existing differentiable factor graph optimization (DFGO) methods learn measurement weighting through the solver, but they still use position-only objectives. As a result, the position estimate may improve while the reported covariance remains too small, too large, or incorrectly oriented. We propose CredibleDFGO (CDFGO), a differentiable GNSS factor graph framework that makes covariance credibility an explicit training target. A Weighting Generation Network (WGN) predicts per-satellite reliability weights, and a differentiable Gauss-Newton solver maps these weights to a position estimate and a Hessian-derived posterior covariance. We use proper scoring rules to supervise the East-North predictive distribution end to end. We study negative log-likelihood (NLL), the energy score (ES), and their combination. Results on three UrbanNav test scenes show consistent gains in covariance credibility. Positioning accuracy also improves on the medium-urban and harsh-urban scenes; on the deep-urban scene, both the mean horizontal error and the 95th-percentile error improve. On the harsh-urban Mong Kok (MK) scene, CDFGO-Combined reduces the mean horizontal error from 13.77 m to 11.68 m, reduces NLL from 40.63 to 6.59, and reduces ES from 12.31 to 9.05 relative to DFGO (MAE). Case studies link the MK improvement to better axis-wise consistency, more credible local covariance ellipses, and satellite-level reweighting. --- ### Uncovering Entity Identity Confusion in Multimodal Knowledge Editing Authors: Shu Wu, Xiaotian Ye, Xinyu Mou, Dongsheng Liu, Xiaohan Wang, Mengqi Zhang Published: 2026-05-07 Categories: cs.CL, cs.CV arXiv: https://arxiv.org/abs/2605.06096v1 Multimodal knowledge editing (MKE) aims to correct the internal knowledge of large vision-language models after deployment, yet the behavioral patterns of post-edit models remain underexplored. In this paper, we identify a systemic failure mode in edited models, termed Entity Identity Confusion (EIC): edited models exhibit an absurd behavior where text-only queries about the original entity's identity unexpectedly return information about the new entity. To rigorously investigate EIC, we construct EC-Bench, a diagnostic benchmark that directly probes how image-entity bindings shift before and after editing. Our analysis reveals that EIC stems from existing methods failing to distinguish between Image-Entity (I-E) binding and Entity-Entity (E-E) relational knowledge in the model, causing models to overfit E-E associations as a shortcut: the image is still perceived as the original entity, with the new entity's name serving only as a spurious identity label. We further explore potential mitigation strategies, showing that constraining edits to the model's I-E processing stage encourages edits to act more faithfully on I-E binding, thereby substantially reducing EIC. Based on these findings, we discuss principled desiderata for faithful MKE and provide methodological guidance for future research. ---