# Research Papers — 2026-05 · bundle page 247 of 326 > Full-text markdown of papers #12301–12350 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/247 [← Previous bundle](https://aipod.dev/papers/archive/2026-05/page/246.md) [Next bundle →](https://aipod.dev/papers/archive/2026-05/page/248.md) ### SynerMedGen: Synergizing Medical Multimodal Understanding with Generation via Task Alignment Authors: Weiren Zhao, Yi Dong, Cheng Chen Published: 2026-05-09 Categories: cs.CV arXiv: https://arxiv.org/abs/2605.08724v1 Unifying multimodal understanding and generation is a compelling frontier that is beginning to emerge in the medical field. However, the limited existing unified medical models typically treat understanding and generation as disjoint objectives, lacking a meaningful functional synergy. In this work, we identify and address a critical question in unified medical modeling: what form of understanding truly benefits generation. We present SynerMedGen, a unified framework built on the proposed principle of generation-aligned understanding, which synergizes understanding objectives with generation tasks via task alignment. SynerMedGen introduces three generation-aligned understanding tasks and a two-stage training strategy that transfers generation-beneficial representations learned during understanding training to medical image synthesis. Remarkably, even with understanding training alone, our SynerMedGen achieves strong zero-shot performance across 22 medical image synthesis tasks and demonstrates robust generalization to unseen datasets. When combined with generation training, SynerMedGen consistently outperforms state-of-the-art specialized medical image synthesis models as well as recent unified medical models. We also release a large-scale dataset named SynerMed consisting of 1M paired synthesis samples and 2M generation-derived understanding instances to support further research on understanding-generation synergy. Our project can be accessed at https://github.com/Mhilab/SynerMedGen. --- ### EAR: Enhancing Uni-Modal Representations for Weakly Supervised Audio-Visual Video Parsing Authors: Huilai Li, Xiaomeng Di, Ying Xing, Yonghao Dang, Yiming Wang, Jianqin Yin Published: 2026-05-09 Categories: cs.CV, cs.MM arXiv: https://arxiv.org/abs/2605.08723v1 Weakly supervised Audio-Visual Video Parsing (AVVP) aims to recognize and temporally localize audio, visual, and audio-visual events in videos using only coarse-grained labels. Faced with the challenging task settings, existing research advances along two main paths: pre-training pseudo-label generators for fine-grained cross-modal semantic guidance, or refining AVVP model architectures to enhance audio-visual fusion. However, since audio and visual signals are typically unaligned, achieving accurate video parsing fundamentally relies on precise perception of uni-modal events. Yet these multi-modal focused strategies excessively emphasize multi-modal fusion while inadequately guiding and preserving uni-modal semantics, resulting in noisy pseudo-labels and sub-optimal video parsing performance. This paper proposes a novel framework that enhances uni-modal representations for both the pseudo-label generator and the AVVP model. Specifically, we introduce a similarity-based label migration approach to annotate pre-training data, thereby enabling the pseudo-label generator to better understand uni-modal events. We also employ a soft-constrained manner to refine modeling of uni-modal features in parallel with multi-modal fusion. These designs enable coordinated attention to both uni-modal and cross-modal representations, thus boosting the localization performance for events. Extensive experiments show that our method outperforms state-of-the-art methods in both pseudo-label and AVVP performance. --- ### Breaking the Impasse: Dual-Scale Evolutionary Policy Training for Social Language Agents Authors: Minzheng Wang, Run Luo, Yanbo Wang, Zichen Liu, Yuqiao Tan, Tao Tan, Xu Nan, Yinhe Zheng, Wenji Mao Published: 2026-05-09 Categories: cs.CL arXiv: https://arxiv.org/abs/2605.08721v1 While Reinforcement Learning with Verifiable Rewards (RLVR) has proven effective for closed-ended tasks, extending it to open-ended social language games via self-play reveals a critical issue: evolution impasse. Due to the vast strategy space, language agents frequently converge to homogenized behaviors, leading to deterministic match outcomes that eliminate the gradient signals necessary for policy evolution. To tackle this issue, we propose Dual-scale Evolutionary Policy Training (DEPT) for social language games. DEPT introduces a time-scaled evolutionary perception mechanism that detects impasse by quantifying dual-scale value baseline divergence alongside match entropy. Upon perceiving the collapse, it then activates asymmetric advantage reshaping to dynamically modulate the optimization landscape for intervention. Thus, our method effectively restores gradient signals and enforces sustained strategic exploration. Extensive experiments on multiple social language games demonstrate that DEPT outperforms strong baselines, avoiding policy degeneration and driving the continuous evolution of social language agents. --- ### Debugging the Debuggers: Failure-Anchored Structured Recovery for Software Engineering Agents Authors: Chenyu Zhao, Shenglin Zhang, Yihang Lin, Wenwei Gu, Zhimin Chen, Yongqian Sun, Dan Pei, Chetan Bansal, Saravan Rajmohan, Minghua Ma Published: 2026-05-09 Categories: cs.SE, cs.AI arXiv: https://arxiv.org/abs/2605.08717v2 Software engineering agents are increasingly deployed in evaluable engineering environments, yet post-failure recovery remains costly, manual, and ad hoc. Existing systems expose traces or generate follow-up feedback, but they do not convert heterogeneous runtime evidence into grounded, bounded recovery guidance for a subsequent attempt. We present PROBE, a failure-anchored framework for structured recovery in software engineering agents. PROBE organizes failed-run telemetry into structured evidence, structured diagnosis, and bounded recovery guidance through a Telemetry Layer, a Diagnosis Layer, and a Guidance Gate. The Telemetry Layer preserves fine-grained runtime signals, the Diagnosis Layer fuses cross-signal evidence into grounded diagnoses, and the Guidance Gate produces diagnosis-derived guidance only when it is evidence-grounded, actionable, and within the scope of agent-side behavior. We evaluate PROBE across three settings: repository-level software repair, enterprise workflow recovery, and AIOps service mitigation. On 257 initially unresolved cases, PROBE achieves 65.37% Top-1 diagnosis accuracy and a 21.79% recovery rate, outperforming the strongest non-PROBE baseline by 43.58 and 12.45 percentage points. The results reveal a diagnosis-recovery gap: accurate diagnosis is necessary but insufficient unless translated into bounded guidance that a subsequent attempt can execute and verify. Beyond controlled evaluation, a Microsoft IcM prototype shows that PROBE can attach as a non-intrusive side channel to existing service-diagnosis workflows without changing the agent policy, toolset, or execution budget. These results suggest that telemetry-grounded, failure-anchored recovery can improve post-failure recoverability under realistic engineering constraints. --- ### Bias by Necessity: Impossibility Theorems for Sequential Processing with Convergent AI and Human Validation Authors: Jikun Wu, Dongxin Guo, Siu-Ming Yiu Published: 2026-05-09 Categories: cs.AI, cs.CL, cs.LG arXiv: https://arxiv.org/abs/2605.08716v1 Are certain cognitive biases mathematically inevitable consequences of sequential information processing? We prove that primacy effects, anchoring, and order-dependence are architecturally necessary in autoregressive language models due to causal masking constraints. Our three impossibility theorems establish: (1) primacy bias arises from asymmetric attention accumulation; (2) anchoring emerges from sequential conditioning with provable information bounds; and (3) exact debiasing by permutation marginalization requires factorial-time computation, with Monte Carlo approximation feasible at constant per-tolerance overhead. We validate these bounds across 12 frontier LLMs ($R^2 = 0.89$; $Δ$BIC $= 16.6$ vs. next-best alternative). We then derive quantitative predictions from the framework and test them in two pre-registered human experiments ($N = 464$ analyzed). Study 1 confirms anchor position modulates anchoring magnitude ($d = 0.52$, BF$_{10} = 847$). Study 2 shows working memory load amplifies primacy bias ($d = 0.41$, BF$_{10} = 156$), with WM capacity predicting bias reduction ($r = -.38$). These convergent findings reframe cognitive biases as resource-rational responses to sequential processing. --- ### AgentForesight: Online Auditing for Early Failure Prediction in Multi-Agent Systems Authors: Boxuan Zhang, Jianing Zhu, Zeru Shi, Dongfang Liu, Ruixiang Tang Published: 2026-05-09 Categories: cs.CL, cs.AI, cs.MA arXiv: https://arxiv.org/abs/2605.08715v2 LLM-based multi-agent systems are increasingly deployed on long-horizon tasks, but a single decisive error is often accepted by downstream agents and cascades into trajectory-level failure. Existing work frames this as \emph{post-hoc failure attribution}, diagnosing the responsible agent and step after the trajectory has ended. However, this paradigm forfeits any opportunity to intervene while trajectory is still unfolding. In this work, we introduce AgentForesight, a framework that reframes this problem as online auditing: at each step of an unfolding trajectory, an auditor observes only the current prefix and must either continue the run or alarm at the earliest decisive error, without access to future steps. To this end, we curate AFTraj-2K, a corpus of agentic trajectories across Coding, Math, and Agentic domains, in which safe trajectories are retained under a strict curation pipeline and unsafe trajectories are annotated at the step of their decisive error via consensus among multiple LLM judges. Built on that, we develop AgentForesight-7B, a compact online auditor trained with a coarse-to-fine reinforcement learning recipe that first equips it with a risk-anticipation prior at the failure boundary on adjacent safe/unsafe prefix pairs, then sharpens this prior into precise step-level localization under a three-axis reward jointly targeting the what, where, and who of an audit verdict. Across AFTraj-2K and an external Who\&When benchmark, AgentForesight-7B outperforms leading proprietary models, including GPT-4.1 and DeepSeek-V4-Pro, achieving up to +19.9% performance gain and 3$\times$ lower step localization error, opening the loop from post-hoc failures detection to enabling deployment-time intervention. Project page: https://zbox1005.github.io/agent-foresight/ --- ### REAP: Reinforcement-Learning End-to-End Autonomous Parking with Gaussian Splatting Simulator for Real2Sim2Real Transfer Authors: Changze Li, Zhe Chen, Shaoyu Chen, Lisen Mu, Yijian Li, Yuelong Yu, Qian Zhang, Qing Su, Ming Yang, Tong Qin Published: 2026-05-09 Categories: cs.RO, cs.AI arXiv: https://arxiv.org/abs/2605.08713v1 In recent years, autonomous parking has made significant advances, yet parking tasks still face challenges in extreme scenarios such as mechanical and dead-end parking slots, often resulting in failures. This is mainly due to traditional parking methods adopting a multistage approach, lacking the ability to optimize the parking problem as a whole. End-to-end methods enable joint optimization across perception and planning modules to eliminate the accumulation of errors, enhancing algorithm performance in extreme scenarios. Although several end-to-end parking methods use imitation or reinforcement learning, the former is limited by data cost and distribution coverage, while the latter suffers from inefficient exploration. To address these challenges, we propose a Reinforcement learning End-to-end Autonomous Parking method (REAP). REAP employs Soft Actor-Critic (SAC) within an asymmetric reinforcement learning framework to improve training efficiency and inference performance. To accelerate model convergence, we distill the capabilities of a rule-based planner into the end-to-end network through behavior cloning. We further introduce a soft predictive collision penalty mechanism to reduce collision rates by penalizing obstacle-approaching actions. To ensure that the trained reinforcement learning network can directly transfer to real-world scenarios, we have established a Real2Sim2Real simulator. In the Real2Sim step, we use 3D Gaussian Splatting (3DGS) to transform real-world scenes into digital scenes. In the Sim2Real step, we deploy the end-to-end model onto the vehicle to bridge the Sim2Real gap. Trained in the 3DGS simulator and deployed on physical vehicles, REAP successfully parks in various types of parking spaces, especially demonstrating the feasibility of end-to-end RL parking in extremely narrow mechanical slots. --- ### From Articulated Kinematics to Routed Visual Control for Action-Conditioned Surgical Video Generation Authors: Bohan Li, Shuojue Yang, Baorui Peng, Xianda Guo, Erli Zhang, Youqi Tao, Junfeng Duan, Daguang Xu, Qi Dou, Xin Jin, Wenjun Zeng, Hao Zhao et al. Published: 2026-05-09 Categories: cs.CV arXiv: https://arxiv.org/abs/2605.08712v3 Action-conditioned surgical video generation is a critical yet highly challenging problem for robotic surgery. The core difficulty is that low-dimensional control vectors must precisely govern complex image-space evolution. In this work, we propose a kinematic-to-visual lifting paradigm that converts articulated kinematics into a unified set of five image-aligned control modalities. Building on this representation, we introduce a hierarchically routed visual control framework that selectively activates the most relevant control modalities and motion scales. Instead of uniformly applying all control signals, our model performs hierarchical routing to dynamically allocate conditioning capacity. We further design kinematic-prior-guided routing loss functions to ensure physically meaningful, temporally stable, and efficient expert utilization. To improve efficiency, we propose a budgeted training and inference scheme that leverages routing-induced sparsity. By selectively discarding low-significance control pathways during training and execution, our approach enables adaptive computation that is complementary to standard distillation. We additionally construct a new benchmark with curated articulated annotations, obtained through human-in-the-loop semantic labeling and differentiable pose tracking, providing realistic supervision for action-conditioned surgical video generation. Extensive experiments demonstrate that our method consistently improves action faithfulness, visual fidelity, and cross-domain generalization over diverse baselines. Moreover, our efficient variant achieves substantial reductions in latency while maintaining strong control accuracy. --- ### From Articulated Kinematics to Routed Visual Control for Action-Conditioned Surgical Video Generation Authors: Bohan Li, Shuojue Yang, Baorui Peng, Xianda Guo, Erli Zhang, Youqi Tao, Junfeng Duan, Daguang Xu, Qi Dou, Xin Jin, Wenjun Zeng, Hao Zhao et al. Published: 2026-05-09 Categories: cs.CV arXiv: https://arxiv.org/abs/2605.08712v1 Action-conditioned surgical video generation is a critical yet highly challenging problem for robotic surgery. The core difficulty is that low-dimensional control vectors must precisely govern complex image-space evolution. In this work, we propose a kinematic-to-visual lifting paradigm that converts articulated kinematics into a unified set of five image-aligned control modalities. Building on this representation, we introduce a hierarchically routed visual control framework that selectively activates the most relevant control modalities and motion scales. Instead of uniformly applying all control signals, our model performs hierarchical routing to dynamically allocate conditioning capacity. We further design kinematic-prior-guided routing loss functions to ensure physically meaningful, temporally stable, and efficient expert utilization. To improve efficiency, we propose a budgeted training and inference scheme that leverages routing-induced sparsity. By selectively discarding low-significance control pathways during training and execution, our approach enables adaptive computation that is complementary to standard distillation. We additionally construct a new benchmark with curated articulated annotations, obtained through human-in-the-loop semantic labeling and differentiable pose tracking, providing realistic supervision for action-conditioned surgical video generation. Extensive experiments demonstrate that our method consistently improves action faithfulness, visual fidelity, and cross-domain generalization over diverse baselines. Moreover, our efficient variant achieves substantial reductions in latency while maintaining strong control accuracy. --- ### When Can Human-AI Teams Outperform Individuals? Tight Bounds with Impossibility Guarantees Authors: Dongxin Guo, Jikun Wu, Siu-Ming Yiu Published: 2026-05-09 Categories: cs.AI arXiv: https://arxiv.org/abs/2605.08710v1 Human-AI teams fail to outperform their best member in 70% of studies, yet no theory specifies when complementarity is achievable. We derive tight bounds for the broad class of confidence-based aggregation rules by integrating signal detection theory with information-theoretic analysis, yielding four results: (1) a complementarity theorem (teams outperform individuals iff error correlation $ρ_{HM} < ρ^*$, with $ρ^* \approx a$ in the symmetric near-chance regime); (2) minimax bounds showing gains scale as $Θ(\sqrt{Δd})$ with metacognitive sensitivity difference; (3) an impossibility result proving no confidence-based aggregation rule achieves complementarity when $ρ_{HM} \geq ρ^*$; and (4) multi-class generalization $ρ^*_K \approx ρ^*/\sqrt{K-1}$. Predictions match observed team accuracy ($R = 0.94$ on ImageNet-16H, $R = 0.91$ on CIFAR-10H) and the multi-class threshold scaling holds on human data ($R = 0.93$, $K = 16$), with robustness under non-Gaussian distributions. The framework explains why complementarity is rare and provides actionable design formulas; results apply to aggregation, not to interactive deliberation that generates novel answers. --- ### UniShield: Unified Face Attack Detection via KG-Informed Multimodal Reasoning Authors: Hongrui Li, Yichen Shi, Hongyang Wang, Yuhao Gao, Hui Ma, Jun Feng, Zitong Yu Published: 2026-05-09 Categories: cs.CV arXiv: https://arxiv.org/abs/2605.08709v1 Unified face attack detection (UAD) requires recognizing physical spoofing and digital forgery within a shared decision space, yet existing discriminative or prompt-based methods largely rely on appearance correlations and provide limited evidence-grounded reasoning. We propose UniShield, a knowledge-grounded multimodal reasoning framework for unified face attack defense. UniShield constructs a Face Attack Knowledge Graph (FAKG) that links attack categories to diagnostic visual cues and attack-conditioned relations, and uses it to synthesize 52,025 FAKG-QA examples for Attack-Graph Instruction Tuning (AGIT). To improve rationale consistency, we further introduce Graph-Consistent Reasoning Optimization (GCRO), a GRPO-based objective with a KG-consistency reward that encourages generated rationales to match graph-supported cues while penalizing incompatible claims. Experiments on our multimodal UAD benchmark show that UniShield achieves strong performance across binary, coarse-grained, and fine-grained protocols, with consistently high ACC and low HTER. These results suggest that structured attack knowledge can improve both detection accuracy and reasoning reliability over discriminative baselines and general-purpose MLLMs. Our code will be released at https://anonymous.4open.science/r/Unishield-A6A3/. --- ### Minimax Optimal Estimation of Transport-Growth Pairs in Unbalanced Optimal Transport Authors: Donlapark Ponnoprat, Noboru Isobe, Masaaki Imaizumi Published: 2026-05-09 Categories: math.ST, math.PR, stat.ME, stat.ML arXiv: https://arxiv.org/abs/2605.08705v1 Unbalanced optimal transport (UOT) extends classical optimal transport to measures with different total masses, but statistical guarantees for Monge-type estimation remain limited. We study unbalanced transport with quadratic cost and Kullback-Leibler marginal penalties and argue that the natural population target is not a map alone, but a transport-growth pair. Consequently, we develop two estimators for the transport-growth pairs under several setups: an optimal transport plan-based estimator for a general case, and a kernel-based estimator for a case with smooth densities. We also show that an error of the estimator achieves the minimax optimal rate by deriving a matching lower bound of the minimax risk. Our main technical contribution is a value-based stability reduction that converts perturbations of the UOT objective into transport and growth risks through a UOT gap condition. These results provide a statistical foundation for Monge-type estimation in unbalanced optimal transport. --- ### AgentPSO: Evolving Agent Reasoning Skill via Multi-agent Particle Swarm Optimization Authors: Hyunmin Hwang, Jaemin Kim, Choonghan Kim, Hangeol Chang, Jong Chul Ye Published: 2026-05-09 Categories: cs.AI arXiv: https://arxiv.org/abs/2605.08704v3 Multi-agent reasoning has shown promise for improving the problem-solving ability of large language models by allowing multiple agents to explore diverse reasoning paths. However, most existing multi-agent methods rely on inference-time debate or aggregation, which can be vulnerable to incorrect peer influence and biased consensus. Moreover, the agents' reasoning procedures remain static, as they do not accumulate reusable behaviors across tasks. In this paper, we introduce \textbf{AgentPSO}, a particle-swarm-inspired framework for evolving multi-agent reasoning skills. AgentPSO treats each agent as a particle-like reasoner whose state is a natural-language skill and whose velocity is a semantic update direction, iteratively guiding agents toward higher-performing skill configurations. Across training iterations, each agent updates its skill by combining its previous velocity, personal-best skill, global-best skill, and a self-reflective direction derived from peer reasoning trajectories. This enables agents to learn reusable reasoning behaviors by drawing on their own experience and on the strongest skills found by the population, without updating the parameters of the backbone language model. Experiments on mathematical and general reasoning benchmarks show that AgentPSO improves over static single-agent skills and test-time-only multi-agent reasoning baselines. The evolved skills further transfer across benchmarks and to another backbone model, suggesting that AgentPSO captures reusable reasoning procedures rather than merely optimizing benchmark-specific prompts. Code is publicly available at https://github.com/HYUNMIN-HWANG/AgentPSO/. --- ### AgentPSO: Evolving Agent Reasoning Skill via Multi-agent Particle Swarm Optimization Authors: Hyunmin Hwang, Jaemin Kim, Choonghan Kim, Hangeol Chang, Jong Chul Ye Published: 2026-05-09 Categories: cs.AI arXiv: https://arxiv.org/abs/2605.08704v2 Multi-agent reasoning has shown promise for improving the problem-solving ability of large language models by allowing multiple agents to explore diverse reasoning paths. However, most existing multi-agent methods rely on inference-time debate or aggregation, which can be vulnerable to incorrect peer influence and biased consensus. Moreover, the agents themselves remain static, as their underlying reasoning skills do not evolve across tasks. In this paper, we introduce \textbf{AgentPSO}, a particle-swarm-inspired framework for evolving multi-agent reasoning skills. AgentPSO treats each agent as a particle-like reasoner whose state is a natural-language skill and whose velocity is a semantic update direction, iteratively guiding agents toward higher-performing skill configurations. Across training iterations, each agent updates its skill by combining its previous velocity, personal-best skill, global-best skill, and a self-reflective direction derived from peer reasoning trajectories. This enables agents to learn reusable reasoning behaviors by drawing on their own experience and on the strongest skills found by the population, without updating the parameters of the backbone language model. Experiments on mathematical and general reasoning benchmarks show that AgentPSO improves over static single-agent skills and test-time-only multi-agent reasoning baselines. The evolved skills further transfer across benchmarks and to another backbone model, suggesting that AgentPSO captures reusable reasoning procedures rather than merely optimizing benchmark-specific prompts. Code is publicly available at https://github.com/HYUNMIN-HWANG/AgentPSO/. --- ### RewardHarness: Self-Evolving Agentic Post-Training Authors: Yuxuan Zhang, Penghui Du, Bo Li, Cong Wei, Junwen Miao, Huaisong Zhang, Songcheng Cai, Yubo Wang, Dongfu Jiang, Yuyu Zhang, Ping Nie, Wenhu Chen et al. Published: 2026-05-09 Categories: cs.AI, cs.CL, cs.CV, cs.LG arXiv: https://arxiv.org/abs/2605.08703v1 Evaluating instruction-guided image edits requires rewards that reflect subtle human preferences, yet current reward models typically depend on large-scale preference annotation and additional model training. This creates a data-efficiency gap: humans can often infer the target evaluation criteria from only a few examples, while models are usually trained on hundreds of thousands of comparisons. We present RewardHarness, a self-evolving agentic reward framework that reframes reward modeling as context evolution rather than weight optimization. Instead of learning from large-scale annotations, RewardHarness aligns with human preferences by iteratively evolving a library of tools and skills from as few as 100 preference demonstrations. Given a source image, candidate edited images, and an editing instruction, an Orchestrator selects the most relevant subset of tools and skills from the maintained library, and a frozen Sub-Agent uses them to construct a reasoning chain that produces a preference judgment. By comparing predicted judgments with ground-truth preferences and analyzing successes and failures in the reasoning process, the Orchestrator automatically refines its library of tools and skills without additional human annotation. Using only 0.05% of the EditReward preference data, RewardHarness achieves 47.4% average accuracy on image-editing evaluation benchmarks, surpassing GPT-5 by 5.3 points. When used as a reward signal for GRPO fine-tuning, RL-tuned models achieve 3.52 on ImgEdit-Bench. Project page: https://rewardharness.com. --- ### Gate-and-Merge: Zero-shot Compositional Personalization of Vision Language Models Authors: Guodong Ding, Angela Yao Published: 2026-05-09 Categories: cs.CV, cs.AI arXiv: https://arxiv.org/abs/2605.08702v1 This paper tackles compositional personalization of vision-language models (VLMs). In this problem, multiple user-defined concepts must be recognized or described jointly at test time. We introduce Gate-and-Merge, a zero-shot framework that enables compositional personalization without the need for co-occurrence training. During personalization, each concept is learned independently as a lightweight LoRA adapter, paired with a concept token. The base model remains unchanged and concepts are kept disentangled. At inference, we enable composition by merging concept-specific LoRA updates directly in weight space. To suppress irrelevant activations and prevent interference, a gating mechanism is employed to estimate textual and visual cues and select only the modules that contribute to the prediction. We further stabilize composition by combining only the most meaningful and mutually consistent updates, helping preserve each concept's identity. Our quantitative and qualitative analyses show consistent gains in performance across multiple personalization tasks in both single-concept and compositional settings. --- ### METBRA25Y: Brazil Surface Meteorology Archive with Harmonized Variables and Quality Control Authors: Matheus Lima Castro, William Dantas Vichete, Leopoldo Lusquino Filho Published: 2026-05-09 Categories: cs.LG, physics.ao-ph arXiv: https://arxiv.org/abs/2605.08701v1 This data paper describes METBRA25Y, a harmonized archive of hourly surface meteorological observations from Brazil derived from public historical records of the Instituto Nacional de Meteorologia (INMET). The dataset was designed to support reproducible environmental, climatological, hydrological, agricultural, urban-risk, and machine-learning studies that require station-level meteorological time series with standardized variable names and explicit quality-control metadata. The processing workflow ingests annual INMET archives, parses station metadata from raw file headers, normalizes heterogeneous Portuguese column names into a canonical schema, constructs hourly timestamps, consolidates observations by city and station, and exports compressed CSV files together with station manifests, per-station quality flags, daily precipitation aggregates, variable-level failure summaries, and missing-data audits. The quality-control protocol follows a two-stage strategy: first, physically implausible values are converted to missing values and flagged; second, temporal and cross-variable consistency checks generate diagnostic flags without necessarily overwriting the original measurements. The resulting package covers observations between 2000 and 2025, with stationspecific temporal coverage, and includes key meteorological variables such as precipitation, air temperature, dew point, relative humidity, atmospheric pressure, wind speed, wind gust, wind direction, and global solar radiation. Based on the summary files included in the current release snapshot, the archive contains 616 unique station codes across variable summaries, of which 605 have coordinates within a broad Brazil plausibility envelope. This paper documents the dataset provenance, file organization, harmonized schema, quality-control rules, technical validation outputs, limitations, and recommended usage practices. --- ### Supersampling Stable Diffusion and Beyond: A Seamless, Training-Free Approach for Scaling Neural Networks Using Common Interpolation Methods Authors: Md Abu Obaida Zishan, Jannatun Noor, Annajiat Alim Rasel Published: 2026-05-09 Categories: cs.CV, cs.LG arXiv: https://arxiv.org/abs/2605.08698v2 Stable Diffusion (SD) has evolved DDPM (Denoising Diffusion Probabilistic Model) based image generation significantly by denoising in latent space instead of feature space. This popularized DDPM-based image generation as the cost and compute barrier was significantly lowered. However, these models could only generate fixed-resolution images according to their training configuration. When we attempt to generate higher resolutions, the resulting images show object duplication artifacts consistently. To solve this problem without finetuning SD models, recent works have tried dilating the convolution kernels of the models and have achieved a great level of success. But dilated kernels are harder to fine-tune due to being zero-gapped. Apart from this, other methods, such as patched diffusion, could not solve the object-duplication problem efficiently. Hence, to overcome the limitations of dilated convolutions, we propose kernel interpolation of SD models for higher-resolution image generation. In this work, we show mathematically that interpolation can correctly scale convolution kernels if multiplied by a constant coefficient and achieve competitive empirical results in generating beyond-training-resolution images with Stable Diffusion using zero training. Furthermore, we demonstrate that our method enables interpolation of deep neural networks to adapt to higher-dimensional training data, with a worst-case performance drop of $2.6\%$ in accuracy and F1-Score relative to the baseline. This shows the applicability of our method to be general, where we interpolate fully-connected layers, going beyond convolution layers. We also discuss how we can reduce the memory footprints of training neural networks, using our method up to at least $4\times$. --- ### MBP-KT: Learning Global Collaborative Information from Meta-Behavioral Pattern for Enhanced Knowledge Tracing Authors: Yuhao Jia, Duantengchuan Li, Jinsong Chen, Zhongjie Mao, Mingwen Tong, Yue Li, Xiaoguang Wang Published: 2026-05-09 Categories: cs.AI arXiv: https://arxiv.org/abs/2605.08697v1 The emerging collaborative information-based knowledge tracing (KT) has been a promising way to enhance modeling of learners' knowledge states. The core idea is to extract the collaborative information from interaction sequences of other learners to assist the prediction on the target one. Despite effectiveness, existing methods are built on the raw interaction sequences with tailored modules, which inevitably limits their capacity in deeply capturing learning behavioral patterns and generalization. To this end, we propose a general meta-behavioral pattern-aware framework (MBP-KT) for KT. Specifically, MBP-KT introduces a novel meta-behavioral sequence construction to transform the raw interaction sequences into the combinations of different meta-behavioral patterns. In this way, the learning behavioral patterns of learners can be effectively preserved. Then, MBP-KT develops a parameter-free module to extract the global collaborative representations from the constructed meta-behavioral sequences. Moreover, MBP-KT provides general injection strategies to introduce the extracted global collaborative information into various downstream KT models, ensuring the universality of the collaborative information. Extensive results on real-world datasets demonstrate that MBP-KT can consistently boosts the performance of a wide range of KT models. --- ### Structured Recurrent Mixers for Massively Parallelized Sequence Generation Authors: Benjamin L. Badger Published: 2026-05-09 Categories: cs.CL, cs.LG arXiv: https://arxiv.org/abs/2605.08696v4 Over the last two decades, language modeling has experienced a shift from the use of predominantly recurrent architectures that process tokens sequentially during training and inference to non-recurrent models that process sequence elements in parallel during training, which results in greater training efficiency and stability at the expense of lower inference throughput. Here we introduce the Structured Recurrent Mixer, an architecture that allows for algebraic conversion between a sequence parallel representation at train time and a recurrent representation at inference, notably without the need for specialized kernels or device-specific memory management. We show experimentally that this dual representation allows for greater training efficiency, higher input information capacity, and larger inference throughput and concurrency when compared to other linear complexity models. We postulate that recurrent models are poorly suited to extended sequence length scaling for information-rich inputs typical of language, but are well suited to scaling in the sample (batch) dimension due to their constant memory per sample. We provide Mojo/MAX inference implementations of SRMs exhibiting 12x the throughput and 170x the concurrency of similarly powerful Transformers inferenced on vLLM, increases characteristic of Pytorch implementations resulting in a 30\% increase in compute-constant GSM8k Pass@k. We conclude by demonstrating that SRMs are effective reinforcement learning training candidates. --- ### EditSleuth: A Dataset of Grounded Reasoning Chains for Image-Edit Forensics Authors: Van-Loc Nguyen, AprilPyone MaungMaung, Minh-Triet Tran, Isao Echizen Published: 2026-05-09 Categories: cs.CV arXiv: https://arxiv.org/abs/2605.08695v1 Forensic analysis of AI-edited images requires more than binary real-versus-fake prediction: a useful system should localize the edit, identify its semantic type, and ground its decisions in visual evidence. Existing image-forensics datasets typically emphasize detection or localization, while reasoning-supervised vision-language datasets rarely target image manipulation and often rely on LLM-generated rationales whose faithfulness is difficult to verify. We introduce EditSleuth, a dataset of 257,725 image-edit triplets constructed from existing image-editing corpora for grounded image-edit forensic reasoning. Each example includes an edited image, its source image, a binary edit mask, a 12-class edit taxonomy label, a difficulty score, and a six-step reasoning chain. EditSleuth chains are generated deterministically from triplet-grounded upstream artifacts, with each statement tied to a specific computable source of evidence. Our analysis reveals that a naive four-component difficulty formulation suffers from a rank-2 correlation collapse among magnitude features; a simplified three-component formulation substantially increases score dispersion on both Pico-Banana and MagicBrush. Difficulty also varies meaningfully within most edit categories, indicating that the score is not a proxy for edit type. As an initial learning study, we fine-tune Qwen2-VL-2B with LoRA and find that chain-as-target supervision matches a label-only baseline on classification accuracy among parseable answers, while additionally yielding grounded explanatory prose that label-only supervision cannot produce. We release the dataset, the deterministic construction pipeline, and pilot training scripts. --- ### SkillMaster: Toward Autonomous Skill Mastery in LLM Agents Authors: Min Yang, Jinghua Piao, Xu Xia, Xiaochong Lan, Jiaju Chen, Yongshun Gong, Yong Li Published: 2026-05-09 Categories: cs.AI arXiv: https://arxiv.org/abs/2605.08693v2 Skills provide an effective mechanism for improving LLM agents on complex tasks, yet in existing agent frameworks, their creation, refinement, and selection are typically governed by external teachers, hand-designed rules, or auxiliary modules. As a result, skills remain external resources to be invoked, rather than capabilities that agents can develop, adapt, and internalize through experience. To endow LLM agents with autonomous skill mastery, we propose SkillMaster, a training framework that teaches agents to create new skills, refine existing skills, and select accumulated skills during task solving. This capability is achieved through three key designs. First, we train agents through trajectory-informed skill review, teaching agents to propose, update, or retain skills based on evidence from completed episodes. Second, each candidate skill edit is designed to be evaluated by its counterfactual utility on related probe tasks, providing a direct learning signal for training skill-editing decisions. Third, we introduce DualAdv-GRPO, which separately estimates advantages for task-solving actions and skill-editing decisions, stabilizing joint training across task solving and skill management. Experiments on ALFWorld and WebShop show that SkillMaster improves the overall success rate over state-of-the-art baselines by 8.8% and 9.3%, respectively, achieving the best performance among all compared methods. Further analysis reveals a marked shift in agent capability: agents trained with SkillMaster can identify skill failures, refine procedural knowledge from trajectory evidence, and transfer improvements to future tasks with limited skill-bank edits. Overall, SkillMaster moves LLM agents beyond mere skill use toward self-improving agents capable of developing, adapting, and applying their own skill repertoires. --- ### AAAC: Activation-Aware Adaptive Codebooks for 4-bit LLM Weight Quantization Authors: Beshr IslamBouli, David Jin Published: 2026-05-09 Categories: cs.LG, cs.CL arXiv: https://arxiv.org/abs/2605.08692v2 Post-training weight-only quantization to 4 bits is widely used to reduce the memory and compute costs of large language model inference. Existing PTQ methods, such as AWQ and GPTQ, improve how weights are mapped onto a fixed 4-bit grid through scaling, clipping, or error compensation. To further improve accuracy, methods such as OmniQuant and QuIP\# uses gradient-assisted algorithms at the cost of hours of quantization time. In this work, we propose AAAC (Activation-Aware Adaptive Codebooks), a lightweight method for 4-bit LLM weight quantization. AAAC replaces the fixed scalar codebook used in standard quantization with two small learned scalar codebooks (64 bytes) per layer. Each group of weights selects the codebook that minimizes activation-weighted reconstruction error, encoding the choice in the unused sign bit of the group's positive scale and adding zero storage overhead. AAAC completes in 3--30 minutes on a single GPU, and adds no memory beyond the model itself. We evaluate against AWQ, GPTQ, IF4, GPTVQ, OmniQuant, SqueezeLLM, and QuIP\# across model families. AAAC outperforms baselines at orders-of-magnitude less quantization time. --- ### Structure-Centric Graph Foundation Model via Geometric Bases Authors: Xiaodong He, Haolan He, Ruiyi Fang, Ming Sun, Zhao Kang Published: 2026-05-09 Categories: cs.LG, cs.AI, cs.SI arXiv: https://arxiv.org/abs/2605.08689v1 Graph foundation models (GFMs) seek transferable representations across graph domains but are limited by structural heterogeneity and incompatible node feature spaces. We propose Structure-Centric Graph Foundation Models (SCGFM), which treat graph topology as the primary source of transferable knowledge. Modeling graphs as metric measure spaces, SCGFM introduces learnable geometric bases that define a shared structural coordinate system. Graphs are aligned to these bases via Gromov-Wasserstein distances, yielding structure-aligned latent representations that accommodate heterogeneous graph topologies. To address feature incompatibility, SCGFM employs a structure-aware feature re-encoding mechanism that unifies node representations without assuming a fixed feature dimensionality or requiring dataset-specific preprocessing. Experiments on graph- and node-level tasks demonstrate strong in-domain and cross-domain generalization, outperforming existing GFM approaches. --- ### Reconciling Consistency-Based Diagnosis with Actual-Causality-Based Explanations Authors: Leopoldo Bertossi Published: 2026-05-09 Categories: cs.AI, cs.DB, cs.LO arXiv: https://arxiv.org/abs/2605.08688v2 We establish, from the point of view of Explainable AI (XAI), connections between Consistency-Based Diagnosis (CBD), on one side, and Actual Causality and Causal Responsibility, on the other. CBD has received little attention from the XAI community. Connections between these two areas could have a fruitful impact on XAI and Explainable Data Management. --- ### Reconciling Consistency-Based Diagnosis with Actual-Causality-Based Explanations Authors: Leopoldo Bertossi Published: 2026-05-09 Categories: cs.AI, cs.DB, cs.LO arXiv: https://arxiv.org/abs/2605.08688v1 We establish, from the point of view of Explainable AI (XAI), connections between Consistency-Based Diagnosis (CBD), on one side, and Actual Causality and Causal Responsibility, on the other. CBD has received little attention from the XAI community. Connections between these two areas could have a fruitful impact on XAI and Explainable Data Management. --- ### PrepBench: How Far Are We from Natural-Language-Driven Data Preparation? Authors: Jingzhe Xu, Rui Wang, Jiannan Wang, Guoliang Li Published: 2026-05-09 Categories: cs.DB, cs.AI arXiv: https://arxiv.org/abs/2605.08687v1 Data preparation is a central and time-consuming stage in data analysis workflows. Traditionally, commercial tools have relied on graphical user interfaces (GUIs) to simplify data preparation, allowing users to define transformations through visual operators and workflows. Recent advances in large language models (LLMs) raise the possibility of a paradigm shift toward natural language (NL)-driven data preparation, in which users can specify preparation intents in NL directly. However, it remains unclear how far current LLM-based agents are from this paradigm shift in practice. Existing code generation benchmarks do not capture key characteristics of data preparation, including ambiguous user intents, imperfect real-world data, and the need to translate code into interpretable workflows for validation. To bridge this gap, we present PrepBench, a benchmark designed to evaluate NL-driven data preparation along three core capabilities: interactive disambiguation, prep-code generation, and code-to-workflow translation. We crawl data from the Preppin' Data Challenges, and then extend it into a systematically designed benchmark. The benchmark covers diverse domains, and each task involves 3 to 18 data preparation steps. Nearly half of the tasks require over 100 lines of Python code, and the longest solutions approach 300 lines. Our evaluation shows that, despite recent progress, realizing this paradigm shift remains challenging for state-of-the-art LLMs. PrepBench provides a principled benchmark for measuring this gap and helps identify key challenges toward realizing NL-driven data preparation. --- ### Iterative Critique-and-Routing Controller for Multi-Agent Systems with Heterogeneous LLMs Authors: Wenzhi Fang, Liangqi Yuan, Guangchen Lan, Dong-Jun Han, Christopher G. Brinton Published: 2026-05-09 Categories: cs.AI arXiv: https://arxiv.org/abs/2605.08686v1 Multi-agent large language model (LLM) systems often rely on a controller to coordinate a pool of heterogeneous models, yet existing controllers are typically limited to one-shot routing: they select a model once and return its output directly. Such routing-only designs provide no mechanism to critique intermediate drafts or support iterative refinement. To address this limitation, we propose a critique-and-routing controller that casts multi-agent coordination as a sequential decision problem. At each turn, the controller evaluates the current draft, decides whether to stop or continue, and, if needed, selects the next agent for further refinement. We formulate this process as a finite-horizon Markov Decision Process (MDP) with explicit agent-utilization constraints, design a composite reward for controller decisions across turns, and optimize the controller via policy gradients under a Lagrangian-relaxed objective. Extensive experiments across multiple heterogeneous multi-agent systems and seven reasoning benchmarks show that our method consistently outperforms state-of-the-art baselines and substantially narrows the gap to the strongest agent, while using it for fewer than 25% of total calls. --- ### Event Fields: Learning Latent Event Structure for Waveform Foundation Models Authors: Li Na, Yuanyun Zhang, Shi Li Published: 2026-05-09 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2605.08685v1 We propose a new class of waveform foundation models that departs from conventional sequence based representations by modeling physiological time series as realizations of latent event processes. Rather than treating signals as collections of local tokens or patches, our approach assumes that clinically meaningful structure arises from temporally extended, interacting events whose boundaries and dynamics are not directly observed. To capture this structure, we introduce a self supervised learning framework that enforces consistency across stochastic segmentations and time frequency projections of the same waveform, encouraging representations that are invariant to signal level perturbations while preserving event level organization. The resulting model combines a segmentation aware encoder with a latent interaction operator that captures dependencies among inferred events, and naturally extends to multimodal settings by aligning modalities through shared event representations. Across a range of physiological benchmarks, including arrhythmia classification, hemodynamic prediction, and waveform retrieval, the proposed method improves performance, robustness, and label efficiency relative to strong sequence based baselines. These results suggest that shifting from signal centric to event centric representations provides a more appropriate inductive bias for modeling physiological dynamics and offers a complementary path to scaling foundation models in healthcare. --- ### Core-Halo Decomposition: Decentralizing Large-Scale Fixed-Point Problems Authors: Haixiang, Yang Xu, Jiefu Zhang, Xudong Wu, Zihan Zhou, Jun He, Jiayu Chen Published: 2026-05-09 Categories: stat.ML, cs.AI, cs.LG, math.NA arXiv: https://arxiv.org/abs/2605.08681v1 We study solving large-scale fixed-point equation \(x^\star=\bar F(x^\star)\) with decomposition. Standard strict decomposition assigns each agent a disjoint block and evaluates updates using only owned coordinates. For most operators, however, a block update may depend on variables outside the block. Truncating these dependencies by strict decomposition changes the mean operator and creates structural bias that cannot be removed by more samples, smaller stepsizes, or additional consensus. We therefore propose Core-Halo decomposition, which separates write ownership from read-only evaluation context: each agent updates its own core and reads from an overlapping halo. By aligning the Core-Halo decomposition with the block-dependence structure of $\bar F$, the original fixed-point problem can be implemented faithfully in a decentralized multi-agent system. We further characterize the fundamental obstruction faced by strict decomposition through a Bellman closure condition and a blockwise bias lower bound, showing that local-only updates can alter the original fixed-point operator. Finally, we conduct extensive experiments across a range of application settings, and demonstrate that Core-Halo achieves near-centralized performance while retaining the parallelism benefits of decentralization. --- ### Semantic Voting: Execution-Grounded Consensus for LLM Code Generation Authors: Shan Jiang, Zijian Yi, Chenguang Zhu Published: 2026-05-09 Categories: cs.SE, cs.AI, cs.LG arXiv: https://arxiv.org/abs/2605.08680v1 LLM code-generation pipelines often sample multiple candidates and select one final answer without access to a complete oracle. Existing pipelines mix textual voting, ranking, and execution-based agreement, but the relative contribution of each component remains unclear. We study 18 configurations across different models, thinking levels, and benchmarks, comparing output-pattern majority voting, weighted voting, MBR-Exec, and SemanticVote - a method that clusters candidates by execution fingerprints on LLM-generated inputs. Three findings emerge. (1) The best execution-based selector exceeds output-pattern majority voting by 19-52 percentage points on every configuration, with every execution-based selector exceeding it by at least 18 points. (2) Once candidates are executed on diverse inputs, aggregation rule has limited effect: SemanticVote, weighted voting, and MBR-Exec are statistically indistinguishable across all 18 configurations. The largest factor is input quality: sketch-based input generation consistently outperforms direct LLM generation by 0.6-2.1 pp and random fuzzing by up to 11.3 pp. (3) Thinking level interacts differently with selection families: deeper thinking improves majority voting by 12 pp but execution-based methods stay flat or degrade as candidate diversity falls. These results frame inference-time code selection as a signal-quality problem rather than an aggregation-rule problem: when oracles are unavailable, the behavioral evidence matters more than the aggregation rule. --- ### Attention-based graph neural networks: a survey Authors: Chengcheng Sun, Chenhao Li, Xiang Lin, Tianji Zheng, Fanrong Meng, Xiaobin Rui, Zhixiao Wang Published: 2026-05-09 Categories: cs.SI, cs.AI, cs.LG arXiv: https://arxiv.org/abs/2605.08679v1 Graph neural networks (GNNs) aim to learn well-trained representations in a lower-dimension space for downstream tasks while preserving the topological structures. In recent years, attention mechanism, which is brilliant in the fields of natural language processing and computer vision, is introduced to GNNs to adaptively select the discriminative features and automatically filter the noisy information. To the best of our knowledge, due to the fast-paced advances in this domain, a systematic overview of attention-based GNNs is still missing. To fill this gap, this paper aims to provide a comprehensive survey on recent advances in attention-based GNNs. Firstly, we propose a novel two-level taxonomy for attention-based GNNs from the perspective of development history and architectural perspectives. Specifically, the upper level reveals the three developmental stages of attention-based GNNs, including graph recurrent attention networks, graph attention networks, and graph transformers. The lower level focuses on various typical architectures of each stage. Secondly, we review these attention-based methods following the proposed taxonomy in detail and summarize the advantages and disadvantages of various models. A model characteristics table is also provided for a more comprehensive comparison. Thirdly, we share our thoughts on some open issues and future directions of attention-based GNNs. We hope this survey will provide researchers with an up-to-date reference regarding applications of attention-based GNNs. In addition, to cope with the rapid development in this field, we intend to share the relevant latest papers as an open resource at https://github.com/sunxiaobei/awesome-attention-based-gnns. --- ### MLS-Bench: A Holistic and Rigorous Assessment of AI Systems on Building Better AI Authors: Bohan Lyu, Yucheng Yang, Siqiao Huang, Jiaru Zhang, Qixin Xu, Xinghan Li, Xinyang Han, Yicheng Zhang, Huaqing Zhang, Runhan Huang, Kaicheng Yang, Zitao Chen et al. Published: 2026-05-09 Categories: cs.LG arXiv: https://arxiv.org/abs/2605.08678v3 Modern AI progress has been driven by ML methods that are generalizable across settings and scalable to larger regimes. As large language models demonstrate advanced capabilities in reasoning, coding, and engineering tasks, it is increasingly important to understand whether they can discover such methods rather than only apply existing ones. We introduce MLS-Bench, a benchmark for evaluating whether AI systems can invent generalizable and scalable ML methods. MLS-Bench contains 140 tasks across 12 domains, each requiring an agent to improve one targeted component of an ML system or algorithm and demonstrate that the improvement generalizes across controlled settings and scales. We find that current agents remain far from reliably surpassing human-designed methods, and that engineering-style tuning is easier for them than genuine method invention. We further study the effects of test-time scaling, adaptive compute allocation, and context provision on agents' discovery performance, together with case studies of their behavior. Our analyses suggest that the bottleneck is not only in proposing new methods, but also in the scientific insight needed to plan, validate, and scale claims about them. More search, compute, or context alone does not remove this bottleneck. We build and maintain a community platform for cumulative and comparable iteration, and release the data and code at https://mls-bench.com. --- ### PHIDA: Persistence-Guided Node-to-Cluster Mapping for Online Clustering Authors: Naoki Masuyama, Yusuke Nojima, Stefan Wermter, Yuichiro Toda, Hisao Ishibuchi, Chu Kiong Loo Published: 2026-05-09 Categories: cs.LG arXiv: https://arxiv.org/abs/2605.08673v1 Online clustering methods that adaptively create and update nodes as data arrive often make node learning explicit, whereas the mapping from the learned node state to output clusters often remains implicit or simplified. Implicit mappings make output clusters sensitive to weak graph bridges or local relations based on distance in the graph over learned nodes, leaving no explicit constraint on which node groups remain intact during mapping. This paper addresses this gap by proposing PHIDA, a persistence-guided node-to-cluster mapping method for online clustering with learned nodes. PHIDA implements this mapping within Adaptive Resonance Theory (ART)-based online clustering by combining Inverse-Distance ART (IDA) node learning with node-to-cluster mapping constrained by Persistent Homology (PH). Experiments on 24 benchmark datasets show that PHIDA achieves the best average ranks in stationary comparisons that include the recent stationary-only clustering methods, while also improving aggregate performance in the nonstationary setting over the evaluated online methods that adaptively create and update nodes. Ablations and comparisons with conventional node-to-cluster mappings indicate that the observed gains are associated with PH-constrained mapping that preserves raw PH components, together with the use of the PH component view during node learning. Source code is available at https://github.com/Masuyama-lab/PHIDA --- ### Posterior Concentration of Bayesian Physics-Informed Neural Networks for Elliptic PDEs Authors: Yuxuan Zhao, Yulong Lu Published: 2026-05-09 Categories: math.ST, math.NA, stat.ML arXiv: https://arxiv.org/abs/2605.08672v1 We study the posterior contraction rate of Bayesian Physics-Informed Neural Networks (PINNs) for solving a general class of elliptic partial differential equations (PDEs). We focus on learning of the elliptic equation with a non-homogeneous Dirichlet boundary condition from independent and noisy measurements collected both inside the domain and on the boundary. Assuming that the PDE admits a strong solution in a Hölder space and using with a suitably constructed prior on the neural network weights, we prove that the posterior distribution concentrates around the exact solution at a near-minimax rate. Furthermore, the chosen prior is rate-adaptive: the posterior contracts at an (almost) optimal rate without prior knowledge of the smoothness level of the exact solution. Our results provide statistical guarantees for uncertainty quantification of PDEs via Bayesian PINNs. --- ### Explanation Fairness in Large Language Models: An Empirical Analysis of Disparities in How LLMs Justify Decisions Across Demographic Groups Authors: Gautam Veldanda Published: 2026-05-09 Categories: cs.CL, cs.AI arXiv: https://arxiv.org/abs/2605.08671v1 Large language models (LLMs) are increasingly deployed not only to make decisions but to explain them. While AI decision fairness has been studied extensively, the fairness of AI explanations (whether LLMs justify decisions with equal quality, depth, tone, and linguistic sophistication across demographic groups) has received little attention. This paper introduces the Explanation Fairness Taxonomy (EFT), a framework comprising five formally defined, operationalizable dimensions: Verbosity Disparity, Sentiment Disparity, Epistemic Hedging Disparity, Decision-Linked Explanation Disparity, and Lexical Complexity Disparity. The taxonomy is instantiated in a controlled empirical study across 80 prompt templates, four consequential decision domains (hiring, medical triage, credit assessment, legal judgment), and five LLMs: GPT-4.1, Claude Sonnet, LLaMA 3.3 70B, GPT-OSS 120B, and Qwen3 32B. Two novel black-box metrics are introduced: the Hedging Density Score (HDS) and the Explanation Faithfulness Proxy (EFP), a heuristic indicator of decision-linked explanation variation. Across up to 400 prompt pairs, all eight EFT metrics show statistically significant disparities (Cohen's d ranging from small to large, all p_BH < 10^(-62)). Model choice is strongly associated with disparity magnitude: Qwen3 32B exhibits verbosity disparities 5.9x larger than LLaMA 3.3 70B. Two prompting-based mitigations show significant reductions in EFP disparity (78-95%) but no significant effect on stylistic dimensions, consistent with the hypothesis that stylistic explanation inequalities are encoded in pre-training distributions and are not resolvable through deployment-level instruction alone. A reproducible measurement framework is offered for explanation-level fairness auditing, with implications for AI regulation and deployment practice. --- ### MIND-Skill: Quality-Guaranteed Skill Generation via Multi-Agent Induction and Deduction Authors: Yixuan Li, Mingshu Cai, Ziyang Xiao, Wanyuan Wang, Yanchen Deng, Bo An Published: 2026-05-09 Categories: cs.AI, cs.CL, cs.MA arXiv: https://arxiv.org/abs/2605.08670v1 Large language model (LLM) powered AI agents have emerged as a promising paradigm for autonomous problem-solving, yet they continue to struggle with complex, multi-step real-world tasks that demand domain-specific procedural knowledge. Reusable agent skills, which encapsulate successful problem-solving strategies, offer a natural remedy by enabling agents to build on prior experience. However, curating such skills has largely remained a manual endeavor, requiring human experts to distill rich domain knowledge into actionable guidelines. In this work, we present $\textbf{M}$ulti-agent $\textbf{IN}$duction and $\textbf{D}$eduction for $\textbf{Skill}$s ($\textbf{MIND-Skill}$), a framework that automatically induces generalizable skills from successful trajectories with robust quality guarantees. MIND-Skill consists of an induction agent which is tasked to abstract reusable skills from successful trajectories, and a deduction agent which aims to reconstruct trajectories by following the induced skills. To guarantee the quality of the generated skills, we introduce a reconstruction loss that compares input and reconstructed trajectories, an outcome loss that enforces the correctness of the reconstructed trajectories, and a rubric loss that assesses the documentation quality and regularizes the abstraction level of the generated skills according to predefined criteria. These textual losses are jointly optimized with TextGrad, and the resulting skills are evaluated on held-out tasks unseen during optimization. Experiments on AppWorld and BFCL-v3 show that MIND-Skill consistently outperforms concurrent skill generation methods. --- ### The Cancellation Hypothesis in Critic-Free RL: From Outcome Rewards to Token Credits Authors: Tianhao Cheng, Zeyu Huang, Zihan Qiu, Yu Cheng, Edoardo Ponti, Yinghui Xu, Ivan Titov, Zenglin Xu Published: 2026-05-09 Categories: cs.LG arXiv: https://arxiv.org/abs/2605.08666v1 A commonly accepted explanation of critic-free RL for LLMs, based on sequence-level rewards, is that it reinforces successful rollouts with a positive advantage while penalizing failed ones. In contrast, we study critic-free RL from a token-level perspective, revealing the token-flipping phenomenon: positive and negative rollouts exhibit remarkably similar proportions of tokens whose probabilities are boosted or suppressed during RL training. To explain this phenomenon, we further show that a token's change in probability is not fully determined by its own advantage; coupled gradient interactions with other tokens also play a non-negligible role. Specifically, these token coupling effects occur primarily between identical tokens that are both predicted with low confidence. Building upon this analysis, we propose the cancellation hypothesis: as a result of coupling, opposing signals cancel out for tokens shared by positive and negative rollouts, while tokens more specific to successful rollouts receive stronger reinforcement, thereby inducing hidden token-level credit assignment from rollout-level rewards. We support this hypothesis with complementary empirical evidence. (1) Compared with training on only positive rollouts, critic-free RL shifts updates from template and formatting tokens toward reasoning tokens; (2) Tokens boosted by critic-free RL consistently demonstrate higher value than suppressed tokens, regardless of whether they originate from positive or negative rollouts. Guided by this view, we implement two batching interventions to encourage or preserve cancellation in critic-free RL training: query-preserved mini-batching and reward-balanced batching. Despite their simplicity, these interventions improve RLVR training across multiple model scales, supporting cancellation as both an explanatory principle and a practical design criterion for critic-free RL training. --- ### Hint Tuning: Less Data Makes Better Reasoners Authors: Siqi Fan, Minghao Li, Xiaoqian Ma, Xiusheng Huang, Zhuo Chen, Bowen Qin, Liujie Zhang, Shuo Shang, Weihang Chen Published: 2026-05-09 Categories: cs.CL arXiv: https://arxiv.org/abs/2605.08665v2 Large reasoning models achieve high accuracy through extended chain-of-thought but generate 5--8 more tokens than necessary, applying verbose reasoning uniformly regardless of problem difficulty. We propose Hint Tuning, a data-efficient approach that teaches models to calibrate reasoning depth. Our key insight: the corresponding instruct model serves as an ideal difficulty probe. By testing what the instruct model can solve with varying guidance, we automatically construct training data across three states: No-Hint (direct answer), Sparse-Hint (minimal prefix), and Full-Hint (complete reasoning). This converts the abstract challenge of difficulty labeling into a measurable consistency check between the instruct and reasoning models. With only 1K self-annotated samples, Hint Tuning achieves 24--66% token reduction (31.5% average) across mainstream reasoning models (Qwen3-Thinking, DeepSeek-R1-Distill) at multiple scales (4B--32B) while maintaining competitive accuracy on five benchmarks. Unlike methods requiring massive distillation datasets or expensive RL, we achieve superior efficiency through simple alignment with the instruct model's capabilities. Code and data are available at https://github.com/redai-infra/hint-tuning. --- ### IPAD-CLIP: Teaching CLIP to Detect Image Local Perceptual Artifacts Authors: Juan Wang, Xinyu Sun, Ke Zhang, Jin Wang, Bing Li, Weiming Hu, Liang Wang Published: 2026-05-09 Categories: cs.CV arXiv: https://arxiv.org/abs/2605.08664v1 Current image quality assessment methods are heavily biased towards global distortions (e.g., noise, blur), neglecting local perceptual artifacts such as ghosting, lens flare, and moire effects. Although significant progress has been made in artifact removal, the fundamental problem of automatic artifact detection remains largely unexplored. In this paper, we formalize the Image Perceptual Artifact Detection (IPAD) task to address this gap. We contribute a benchmark dataset comprising 3,520 artifact images, including 520 real-captured and 3,000 synthetic samples, each paired with pixel-level masks across three representative artifact categories. The core challenge of IPAD lies in the localized, subtle, and semantically weak nature of these artifacts, which makes them prone to missed detection. To overcome this, we introduce IPAD-CLIP, a novel framework built upon CLIP that enhances artifact discrimination in both textual and visual spaces while preserving generalization capabilities. Our key insight is that local artifacts often exhibit strong correlations with specific semantic contexts. Accordingly, we learn artifact-aware text embeddings to explicitly model the object-artifact relationships, resulting in enhanced representations that clear differentiate between clean and artifact prompts. These text embeddings are then used as anchors to shift the visual encoder's attention from high-level semantics to subtle, low-level artifacts. Extensive experiments demonstrate that IPAD-CLIP offers a resource-efficient adaptation of CLIP for detection, significantly outperforming advanced image anomaly detection and manipulation detection methods on our benchmark. To the best of our knowledge, this is the first study addressing multi-class local perceptual artifact detection in terms of both dataset and model. --- ### CAST: Channel-Aware Spatial Transfer Learning with Pseudo-Image Radar for Sign Language Recognition Authors: Md. Shakhoyat Rahman Shujon, Sheikh Md. Galib Mahim, Md. Milon Islam, Md Rezwanul Haque, Md Rabiul Islam, Hamdi Altaheri, Fakhri Karray Published: 2026-05-09 Categories: cs.CV arXiv: https://arxiv.org/abs/2605.08663v1 We propose CAST, a dual-stream architecture that utilizes channel-aware spatial transfer learning for isolated sign language recognition addressing the challenges of magnitude-only 60~GHz radar Range-Time Maps (RTM). The proposed framework combines three physics-aware architectures with pretrained vision backbones, which operate under radar-only constraints across clinical and alphabetical gestures. First, an explicit decibel-to-linear inversion is combined with a windowed fast Fourier transform that extracts Cadence Velocity Diagrams (CVD) while avoiding the harmonic artifacts that arise from the spectral analysis of log-compressed signals. Second, a cross-antenna spatial attention module applies attention to raw antenna channels before the convolution, preserving inter-receiver amplitude covariance. Third, an asymmetric cross-attention mechanism fuses representations from parallel ConvNeXt-Tiny (CVD) and EfficientNetV2-S (RTM) backbones. Extensive experiments reveal that the architecture achieves a Top-1 accuracy of 80.5% under 5-fold cross-validation, establishing a 3.3% improvement over the best single-model baseline (77.2%). The findings suggest that physics-aware signal representations form a promising direction for radar-only sign language recognition under constrained sensor modalities. The source code is available at: https://github.com/Shakhoyat/CAST-at-SignEval2026. --- ### Optimised Support Vector Regression for California Housing Price Prediction: The Critical Role of Feature Engineering and Hyperparameter Tuning Authors: Emmanuel Adutwum Published: 2026-05-09 Categories: cs.LG arXiv: https://arxiv.org/abs/2605.08660v1 In the recent literature, Support Vector Regression (SVR) has been cited as one of the weakest performers on the California Housing benchmark dataset, with Preethi et al. (2025)specifically ranking it last among the algorithms they tested, reporting an R2 of only 0.60. This paper examines whether the previously reported performance reflects experimental configuration choices rather than an inherent algorithmic limitation. A structured experimental workflow is applied: ten domain-motivated derived features are constructed from the eight raw inputs, an exploratory ensemble feature importance analysis identifies the most predictive candidates, and a randomised search over hyperparameter combinations with three-fold cross-validation selects the optimal SVR configuration within a leakage-safe scikit-learn Pipeline. A formal four-stage ablation study isolates the contribution of each component: scaling alone accounts for +0.744 in R2 (from -0.054 to 0.690), feature engineering adds +0.026 (to 0.716), and hyperparameter tuning contributes +0.008 (to 0.723). The resulting tuned SVR achieves a test R2 of 0.723, a 0.123-point absolute improvement over the previously reported SVR result (from 0.60 to 0.723, approximately 20% relative gain). In the ten-model comparison, the tuned SVR ranks fourth with R2 = 0.723, below XGBoost (0.832), Random Forest (0.814) and Gradient Boosting (0.783), while substantially outperforming simpler baselines. Ten-fold cross-validation yields a mean R2 of 0.703 (95% CI: (0.630, 0.775)), confirming robust generalisation. The observed improvement from R2 = 0.60 to R2 = 0.723 is associated primarily with proper feature scaling within a unified preprocessing pipeline, with domain-motivated feature engineering and systematic hyperparameter tuning, providing further incremental gains. --- ### Sketch-and-Verify: Structured Inference-Time Scaling via Program Sketching Authors: Shan Jiang, Zijian Yi, Chenguang Zhu Published: 2026-05-09 Categories: cs.LG, cs.AI, cs.SE arXiv: https://arxiv.org/abs/2605.08658v1 SKETCHVERIFY is a within-tier cost-performance policy, not a universal accuracy improvement. The operational question: a practitioner stuck with a small, cheap code model (here, Gemini 3.1 Flash Lite) for latency, deployment, or budget reasons -- how should they spend a small amount of extra test-time compute? SKETCHVERIFY factorizes the search space: the LLM enumerates K distinct algorithmic strategies, writes a program sketch for each (a partial program with ?? holes), and fills each sketch M times, producing K x M structurally diverse candidates that are verified by execution and selected by fingerprint clustering. Each extra sketch is guaranteed to explore a different algorithm; each extra flat sample likely duplicates an existing one. Our central evidence is a cost-quality Pareto plot on HumanEval+ across three Gemini tiers (Lite, Flash, Pro), and a reanalysis of the 19 problems where Lite greedy fails. Two findings: (1) Within-tier, sketching dominates flat sampling at matched candidate count. On the hard subset, Lite Sketch K=2, M=5 recovers 11/19 (58%) vs. flat N=10 at 5/19 (26%, +32pp); Lite Sketch K=10, M=10 recovers 15/19 (79%) vs. flat N=100 at 10/19 (53%, +26pp). Flat cannot close the gap even at ~3x the budget: flat N=50 still loses to Sketch K=2, M=5 by +11pp. (2) Cross-tier, sketching does not replace upgrading. Pro greedy (89%) dominates Lite Sketch K=10, M=10 (79%) on both pass@1 and dollar cost. Practitioner rule: if a stronger tier is available, use greedy on it; otherwise sketching is the cost-effective way to spend extra compute. We characterize the K-vs-M trade-off via a Flash Lite scaling sweep, report HumanEval+ saturation on Flash and Pro, and show the method composes cleanly with execution-based selection from the concurrent Semantic Voting line of work. --- ### Fitting Multilinear Polynomials for Logic Gate Networks Authors: Youngsung Kim Published: 2026-05-09 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2605.08657v1 We study learnable logic gate networks that stack layers of 2-input Boolean gates to build combinational circuits. Every 2-input gate has a unique multilinear polynomial with 4 coefficients, so the 16 Boolean gates form a codebook of prototypes in a 4-dimensional space, reducing training to a vector-quantization problem. The baseline method, Soft-Mix, learns a 16-dimensional softmax over gate identities, but the codebook has rank~4: 11 of 15 simplex directions carry nullspace gradient, and at uniform initialization the backward signal vanishes exactly. We prove that no affine product reparameterization fixes the resulting interaction-coefficient starvation under STE, and show that the covariance Jacobian of soft-VQ selection bypasses it by coupling the starved coefficient to the always-active constant channel. Working in the 4-dimensional polynomial space reduces each neuron from 16 to 4 parameters. On seven datasets, at least one 4-parameter method matches or exceeds Soft-Mix on every dataset; the CovJac advantage over STE grows monotonically with interaction demand across all seven datasets. At depth, Soft-Mix collapses ($-37.3$pp on CIFAR-10 at 12 layers) while CovJac holds ($-0.5$pp on CIFAR-10, stable on MNIST). --- ### C2L-Net: A Data-Driven Model for State-of-Charge Estimation of Lithium-Ion Batteries During Discharge Authors: Khoa Tran, Tri Le, Nhu Nguyen Gia, T. Nguyen-Thoi, Vin Nguyen-Thai, Duong Tran Anh, Hung-Cuong Trinh Published: 2026-05-09 Categories: cs.AI arXiv: https://arxiv.org/abs/2605.08653v2 Accurate state-of-charge (SOC) estimation is critical for the safe and efficient operation of lithium-ion batteries in battery management systems (BMS). Although data-driven approaches can effectively capture nonlinear battery dynamics, many existing methods rely on long historical input sequences, resulting in high computational cost and introducing padding-induced positional bias at the beginning of drive cycles. To address these limitations, we propose C2L-Net, a novel context-to-latest data-driven framework for realistic online SOC estimation using only a short historical window (20 s). Unlike existing short-receptive-field or long-history models, the proposed framework explicitly separates contextual encoding from latest-measurement updating, enabling both efficient temporal modeling and rapid adaptation to dynamic battery states. The proposed model incorporates a chunk-based feature extraction mechanism that combines Theta Attention Pooling with a Fourier-based Seasonality Basis to capture local temporal patterns while reducing sequence length. A causal context encoder, integrating a gated recurrent unit (GRU) with Causal Cosine Attention, models temporal dependencies without information leakage. Furthermore, a latest-measurement decoder, inspired by recursive filtering, updates the contextual state using the most recent measurement, enhancing responsiveness to dynamic operating conditions. Extensive experiments on a public lithium-ion battery drive-cycle dataset under multiple fixed-temperature conditions demonstrate that the proposed method achieves state-of-the-art or competitive accuracy while significantly improving computational efficiency. In particular, C2L-Net achieves up to 60 times faster inference and requires fewer parameters than recent data-driven baselines, while maintaining robust performance across unseen driving profiles. --- ### C2L-Net: A Data-Driven Model for State-of-Charge Estimation of Lithium-Ion Batteries During Discharge Authors: Khoa Tran, T. Nguyen-Thoi, Vin Nguyen-Thai, Duong Tran Anh, Hung-Cuong Trinh, Tri Le Published: 2026-05-09 Categories: cs.AI arXiv: https://arxiv.org/abs/2605.08653v1 Accurate state-of-charge (SOC) estimation is critical for the safe and efficient operation of lithium-ion batteries in battery management systems (BMS). Although data-driven approaches can effectively capture nonlinear battery dynamics, many existing methods rely on long historical input sequences, resulting in high computational cost and introducing padding-induced positional bias at the beginning of drive cycles. To address these limitations, we propose C2L-Net, a novel context-to-latest data-driven framework for realistic online SOC estimation using only a short historical window (20 s). Unlike existing short-receptive-field or long-history models, the proposed framework explicitly separates contextual encoding from latest-measurement updating, enabling both efficient temporal modeling and rapid adaptation to dynamic battery states. The proposed model incorporates a chunk-based feature extraction mechanism that combines Theta Attention Pooling with a Fourier-based Seasonality Basis to capture local temporal patterns while reducing sequence length. A causal context encoder, integrating a gated recurrent unit (GRU) with Causal Cosine Attention, models temporal dependencies without information leakage. Furthermore, a latest-measurement decoder, inspired by recursive filtering, updates the contextual state using the most recent measurement, enhancing responsiveness to dynamic operating conditions. Extensive experiments on a public lithium-ion battery drive-cycle dataset under multiple fixed-temperature conditions demonstrate that the proposed method achieves state-of-the-art or competitive accuracy while significantly improving computational efficiency. In particular, C2L-Net achieves up to 60 times faster inference and requires fewer parameters than recent data-driven baselines, while maintaining robust performance across unseen driving profiles. --- ### Privacy-Aware Video Anomaly Detection through Orthogonal Subspace Projection Authors: Lei Wang, Wenxiang Diao, Andrew Busch, Jun Zhou, Yongsheng Gao Published: 2026-05-09 Categories: cs.CV, cs.AI, cs.LG arXiv: https://arxiv.org/abs/2605.08651v1 Video anomaly detection (VAD) systems often prioritize accuracy while overlooking privacy concerns, limiting their suitability for real-world deployment. We propose the Orthogonal Projection Layer (OPL), a lightweight module that removes task-irrelevant variations to produce representations focused on anomaly-relevant cues. To address privacy risks in human-centered scenarios, we introduce Guided OPL (G-OPL), which suppresses facial attributes using weak supervision from face-presence signals while preserving non-identifying features such as pose and motion. A cosine alignment objective enforces consistent capture and removal of facial information without identity labels or adversarial training. We further present a privacy-aware evaluation framework that jointly assesses detection performance and privacy preservation, and enables analysis of how sensitive information is filtered. Experiments show that embedding privacy constraints into model design reduces sensitive information while maintaining or improving detection accuracy, supporting projection-based architectures as a principled approach for privacy-aware VAD. --- ### FLUX: Geometry-Aware Longitudinal Flow Matching with Mixture of Experts Authors: Josue Ortega Caro, Yongxu Zhang, Hannah M Batchelor, Sizhuang He, Jessica Cardin, Shreya Saxena Published: 2026-05-09 Categories: cs.LG, q-bio.NC arXiv: https://arxiv.org/abs/2605.08648v1 Many biological systems evolve through continuous local dynamics while switching between latent regimes defined by learning, stimulus context, internal state, or developmental stage. These processes are often observed only as unpaired longitudinal snapshots: the same cells, neurons, or animals are not tracked as matched trajectories, even though population states are sampled across successive stages. This creates two coupled challenges. First, trajectories must respect curved low-dimensional manifolds embedded in high-dimensional biological measurements. Second, the model must identify when the transport mechanism itself changes. We introduce FLUX (FLow matching for Unpaired longitudinal data with miXture-of-experts), a geometry-aware longitudinal flow-matching framework for joint transport modeling and unsupervised regime discovery. FLUX learns a data-dependent metric from pooled labeled and unlabeled observations, uses that metric to construct geometry-aware conditional paths between adjacent marginals, and decomposes the resulting velocity field into sparse expert vector fields selected by a Straight-Through Gumbel-Softmax router. Across manifold controls, a regime-switching Lorenz system, widefield cortical calcium imaging during associative learning, and embryoid body single-cell differentiation, FLUX reconstructs longitudinal transport while recovering interpretable regime structure. Ablations show that mixture-of-experts routing alone is insufficient: FLUX without geometric learning can fit local transport but fails or weakens regime discovery when regimes are encoded in local dynamics. These results suggest that geometry-aware velocity decomposition provides a general strategy for discovering latent biological state transitions from unpaired longitudinal snapshots. --- ### AgentCollabBench: Diagnosing When Good Agents Make Bad Collaborators Authors: Aritra Mazumder, Shubhashis Roy Dipta, Nusrat Jahan Lia, Tanzila Khan, Kainat Raisa Hossain, Nehaa Shri, Shubhrangshu Debsarkar, Humayra Tasnim, Gour Gupal Talukder Shawon, Debjoty Mitra, Sumaiya Ahmed Rani, Al Jami Islam Anik et al. Published: 2026-05-09 Categories: cs.CL, cs.AI, cs.LG arXiv: https://arxiv.org/abs/2605.08647v1 Multi-agent systems achieve state-of-the-art outcomes through peer collaboration. However, when an agent in the pipeline silently drops a constraint, the system's final output may look correct even though the reasoning chain was quietly corrupted, and existing outcome-based evaluations are blind to such multi-hop process failures. To make these vulnerabilities measurable before deployment, we introduce AgentCollabBench, a diagnostic benchmark of 900 human-validated tasks spanning software engineering, DevOps, and data engineering. Each task isolates one of four behavioral risks: instruction decay (does a constraint survive peer pressure?), false-belief contagion (does a falsehood spread through consensus?), context leakage (does information bleed between tasks?), and tracer durability (does marked data reach the final agent?). Evaluating four modern LLMs (GPT 4.1 mini, Gemini 2.5 Flash Lite, Qwen-3.5-35B-A3B, and Llama 3.1 8B Instruct), we expose model-specific vulnerability profiles invisible to outcome-only evaluation; Qwen-3.5-35B-A3B, for example, leads on tracer durability and instruction stability, while GPT 4.1 mini leads on leakage containment and false-belief resistance. Beyond per-model differences, communication topology emerges as a primary risk factor that explains 7-40% of the variance in multi-hop information survival. The effect traces to a synthesis bottleneck specific to converging-DAG nodes: an agent weighing competing parent inputs discards constraints carried by a minority branch, a bottleneck structurally absent from linear chains. AgentCollabBench demonstrates that suboptimal topology can silently erase the safeguards of highly capable models, arguing that multi-agent reliability is fundamentally a structural problem and that scaling model intelligence alone is no substitute for architecture. --- ### PAAC: Privacy-Aware Agentic Device-Cloud Collaboration Authors: Liangqi Yuan, Wenzhi Fang, Shiqiang Wang, Christopher G. Brinton Published: 2026-05-09 Categories: cs.LG, cs.CL, cs.DC arXiv: https://arxiv.org/abs/2605.08646v1 Large language model (LLM) agents face a structural tension: cloud agents provide strong reasoning but expose user data, while on-device agents preserve privacy at the cost of overall capability. Existing device-cloud designs treat this boundary as a compute split rather than a trust boundary suited to agentic workloads, and existing sanitizers force a choice between policy flexibility and the structural fidelity tool calls require. In this work, we develop PAAC, a privacy-aware agentic framework that aligns planner--executor decomposition with the device-cloud boundary so that role specialization itself becomes the privacy mechanism. The cloud agent reasons over typed placeholder tokens that preserve each sensitive value's reasoning role while discarding its content, while the on-device agent identifies sensitive spans and distills each step's execution outcome into compact key findings. Sanitization confines the on-device LLM to proposing which spans to mask, while a deterministic registry performs all substitution and reversal, keeping actions directly executable on device. On three agentic benchmarks under strict privacy settings, PAAC dominates the Pareto frontier of privacy and accuracy, improving average accuracy by 15-36\% and reducing average leakage by 2-6$\times$ over state-of-the-art device-cloud baselines, with the largest margins on privacy targets outside fixed entity taxonomies. We find consistent improvements on 17 additional benchmarks spanning 10 domains, including math, science, and finance. ---