# Research Papers — 2026-05 · bundle page 127 of 326 > Full-text markdown of papers #6301–6350 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/127 [← Previous bundle](https://aipod.dev/papers/archive/2026-05/page/126.md) [Next bundle →](https://aipod.dev/papers/archive/2026-05/page/128.md) ### Retrieval-Augmented Long-Context Translation for Cultural Image Captioning: Gators submission for AmericasNLP 2026 shared task Authors: Aashish Dhawan, Christopher Driggers-Ellis, Dzmitry Kasinets, Daisy Zhe Wang, Christan Grant Published: 2026-05-20 Categories: cs.CL, cs.AI, cs.CV arXiv: https://arxiv.org/abs/2605.20626v1 We present the University of Florida Gators submission to the AmericasNLP 2026 shared task on cultural image captioning for Indigenous languages. Our two-stage pipeline generates a Spanish intermediate caption with Qwen2.5-VL, then produces the target-language caption using retrieval-augmented many-shot prompting with Gemini 2.5 Flash. We achieve 164.1%, 131.7%, and 122.6% improvements over the shared task baseline for Bribri, Guaraní, and Orizaba Nahuatl captioning, respectively, in our dev set evaluation and maintain >150% improvements for the Bribri and Orizaba Nahuatl languages in the test set evaluation. We find retrieval is highly language-dependent, beneficial only for large, in-domain corpora, and that synthetic data augmentation accounts for around 28 chrF++ of the dev set Guaraní performance gain. Our submission is the overall winner of the shared task, placing second out of five finalist submissions in human evaluations of target-language captions. --- ### Accelerating Video Inverse Problem Solvers with Autoregressive Diffusion Models Authors: Taesung Kwon, Jonghyun Park, Hyungjin Chung, Jong Chul Ye Published: 2026-05-20 Categories: cs.CV, cs.AI, cs.LG arXiv: https://arxiv.org/abs/2605.20624v1 Diffusion models provide powerful priors for zero-shot video inverse problems, but their real-time deployment is hindered by two inefficiencies: high initial latency caused by holistic video restoration, and low throughput resulting from multiple VAE passes to enforce measurement consistency in pixel space. To overcome these limitations, we propose Autoregressive Video Inverse problem Solver (AVIS). The AVIS framework leverages autoregressive video diffusion models to restore videos in a streaming manner, naturally eliminating latency bottlenecks. Specifically, AVIS initializes reverse diffusion with a measurement-consistent estimate, reducing the required sampling steps. Compared to leading non-autoregressive solvers, AVIS drastically reduces initial latency from 114s to 4s and increases throughput from 0.71 to 1.18 FPS while achieving superior restoration quality. We further introduce a highly accelerated variant, dubbed AVIS Flash, that enforces measurement consistency solely on the first chunk. AVIS Flash substantially boosts throughput to 5.91 FPS on a single RTX 4090 GPU while maintaining competitive performance and achieving a favorable efficiency-performance trade-off, paving the way toward real-time deployment. --- ### Lower Bounds for Advection-Diffusion Equations: An Exploration with AI-Generated Proofs Authors: Chenyang An, Xiaoqian Xu Published: 2026-05-20 Categories: math.AP, cs.AI arXiv: https://arxiv.org/abs/2605.20623v1 We establish explicit lower bounds for advection-diffusion equations in three settings: a polynomial $\dot H^{-1}$ bound for inviscid shears with $u\in L^\infty_t W^{1,1}_y$, a uniform positive lower bound on the mixing scale for diffusive shears, and an exponential $L^2$ bound for rapidly oscillating time-periodic flows. All constants are explicit in the data. The proofs were generated entirely by a multi-agent math proving system, QED, without expert human intervention, serving as a test of AI's capability to produce rigorous mathematics. --- ### Dynamic Shapley Computation Authors: Xuan Yang, Hsi-Wen Chen, Ming-Syan Chen, Jian Pei Published: 2026-05-20 Categories: cs.LG, cs.DB, cs.GT arXiv: https://arxiv.org/abs/2605.20620v1 Shapley-based data valuation provides a principled way to quantify the contribution of training data, but its high computational cost makes it impractical in dynamic settings where tasks and training players evolve. Existing methods treat Shapley computation as a one-shot process and collapse contributions into aggregated scores, preventing reuse and requiring recomputation under any change. We introduce a new perspective that represents Shapley values as a player-by-task matrix and formulates dynamic valuation as a structured matrix maintenance problem. We exploit the fact that each task depends on a small subset of training players and that similar tasks yield similar valuations, leading to utility locality and coalition locality. Based on these insights, we propose D-Shap, a dynamic valuation framework that enables efficient updates by modifying only a small portion of the matrix: new task valuations are inferred via structure-aware interpolation, while updates induced by new players are confined to affected local matrix blocks. To eliminate the need for pre-specified evaluation tasks, we introduce self-valuation, which constructs the initial matrix directly from training data, supported by scalable subset reuse and coverage-aware anchor selection. Experiments across diverse models show that D-Shap performs task updates in milliseconds and reduces the cost of player updates by up to three orders of magnitude, while achieving valuation quality competitive with full recomputation. --- ### SURF: Steering the Scalarization Weight to Uniformly Traverse the Pareto Front Authors: Liuyuan Jiang, Chentong Huang, Lisha Chen Published: 2026-05-20 Categories: cs.LG, math.OC, stat.ML arXiv: https://arxiv.org/abs/2605.20619v1 Scalarization is widely used in multi-objective optimization owing to its simplicity and scalability. In many applications, the goal is to generate solutions that represent diverse user preferences, ideally with uniform coverage of the Pareto front (PF). However, uniformly sampling scalarization weights usually induces non-uniform coverage of the PF. We explain this mismatch through a geometric analysis of the scalarization path. As the scalarization weight varies, the corresponding solutions trace the PF with a generally non-uniform traversal speed. This speed induces an arc-length cumulative distribution function (CDF); inverting this CDF map yields a principled rule for selecting weights that produce uniform PF coverage. Building on this insight, we propose SURF (Sampling Uniformly along the PaReto Front). For structured problems, including bi-objective bandits, we derive closed-form expressions for this CDF map and the resulting PF-aware weight sampling rule. For general problems, SURF alternates between CDF reconstruction and weight sampling. Theoretically, we show that under provable conditions, SURF converges linearly to an unavoidable finite-sampling floor. Empirically, experiments on bandits, multi-objective-gymnasium, and multi-objective LLM alignment demonstrate that SURF efficiently achieves more uniform PF coverage than baselines. --- ### COAgents: Multi-Agent Framework to Learn and Navigate Routing Problems Search Space Authors: Oleksandr Yakovenko, Mahdi Mostajabdaveh, Cheikh Ahmed, Abdullah Ali Sivas, Xiaorui Li, Zirui Zhou, Mao Kun Published: 2026-05-20 Categories: cs.AI arXiv: https://arxiv.org/abs/2605.20618v1 Although Vehicle Routing Problems (VRP) are essential to many real-world systems, they remain computationally intractable at scale due to their combinatorial complexity. Traditional heuristics rely on handcrafted rules for local improvements and occasional \textit{jumps} to escape local minima, but often struggle to generalize across diverse instances. We introduce \textbf{COAgents}, a cooperative multi-agent framework that models the search process as a graph: nodes represent solutions, and edges correspond to either local refinements or large perturbations for diversification (i.e., jumps). A \textit{Partial Search Graph} (PSG) is dynamically constructed during search, enabling COAgents to train a Node Selection Agent and a Move Selection Agent to guide intensification, and a Jump Agent to trigger well-timed explorations of new regions. Unlike end-to-end learning approaches, COAgents cleanly separates problem-agnostic search control from compact domain-specific encoding, facilitating adaptability across tasks. Extensive experiments on the CVRP and VRPTW benchmarks show that COAgents remains competitive with several learn-to-search baselines on CVRP and sets a new state of the art among learning-based methods on the more challenging VRPTW instances, reducing the gap to the best-known solutions by 14\% at $N\!=\!100$ and 44\% at $N\!=\!50$ relative to the strongest neural solver (POMO), and by 21\% and 40\% respectively relative to ALNS. Code is available at https://github.com/mahdims/COAgents. --- ### Auto-Dreamer: Learning Offline Memory Consolidation for Language Agents Authors: Chongrui Ye, Yuxiang Liu, Yu Wang, Haofei Yu, Yining Zhao, Ge Liu, Julian McAuley, Jiaxuan You Published: 2026-05-20 Categories: cs.CL arXiv: https://arxiv.org/abs/2605.20616v1 Language agents increasingly operate over streams of related tasks, yet existing memory systems struggle to convert accumulated experience into reusable knowledge. Retrieval-augmented and structured memory methods record per-session observations effectively, but often couple acquisition and consolidation into a single online process, leaving the agent without a global view across sessions to discover recurring patterns, abstract shared procedures, or prune redundant entries. Inspired by complementary learning systems theory, we propose Auto-Dreamer, a learned offline consolidator for language-agent memory. Auto-Dreamer decouples fast per-session memory acquisition from slow cross-session consolidation. Given a selected working region of a typed memory bank, the consolidator treats the region as read-only evidence, performs bounded tool-use to inspect entries and provenance-linked source trajectories, and synthesizes a fresh compact replacement set that abstracts across sessions and supersedes the original region. We train Auto-Dreamer via GRPO, using end-to-end agent performance as the reward signal to learn how to consolidate memories acquired through fast online experience. Trained on ScienceWorld trajectories alone, Auto-Dreamer outperforms fixed, RL-trained, and prompted memory baselines on ScienceWorld by 7 points while using an active memory bank 12$\times$ smaller than the strongest baseline, and continues to lead on held-out ALFWorld and WebArena without retraining -- using 6$\times$ less memory than the strongest baseline on ALFWorld. --- ### HRM-Text: Efficient Pretraining Beyond Scaling Authors: Guan Wang, Changling Liu, Chenyu Wang, Cai Zhou, Yuhao Sun, Yifei Wu, Shuai Zhen, Luca Scimeca, Yasin Abbasi Yadkori Published: 2026-05-20 Categories: cs.CL arXiv: https://arxiv.org/abs/2605.20613v1 The current pretraining paradigm for large language models relies on massive compute and internet-scale raw text, creating a significant barrier to foundational research. In contrast, biological systems demonstrate highly sample-efficient learning through multi-timescale processing, such as the functional organization of the frontoparietal loop. Taking this as inspiration, we introduce HRM-Text, which replaces standard Transformers with a Hierarchical Recurrent Model (HRM) that decouples computation into slow-evolving strategic and fast-evolving execution layers. To stabilize this deep recurrence for language modeling, we introduce MagicNorm and warmup deep credit assignment. Furthermore, instead of standard raw-text pretraining, we train exclusively on instruction-response pairs using a task-completion objective and PrefixLM masking. Serving as an empirical existence proof of efficient pretraining, a 1B-parameter HRM-Text model trained from scratch on only 40 billion unique tokens and $1,500 budget achieves 60.7% on MMLU, 81.9% on ARC-C, 82.2% on DROP, 84.5% on GSM8K, and 56.2% on MATH. Despite utilizing roughly 100-900x fewer training tokens and 96-432x less estimated compute than standard baselines, HRM-Text performs competitively with 2-7B parameter open models. These results demonstrate that co-designing architectures and objectives can radically reduce the compute-to-performance ratio, making pretraining from scratch accessible to the broader research community. --- ### Matryoshka Concept Bottleneck Models Authors: Ziye Chen, Hongbin Lin, Jie Li, Lijie Hu Published: 2026-05-20 Categories: cs.LG arXiv: https://arxiv.org/abs/2605.20612v3 Concept Bottleneck Models (CBMs) have emerged as a prominent paradigm for interpretable deep learning, learning by grounding predictions in human-understandable concepts. However, their practical deployment is hindered by the high cost of test-time intervention, as correcting model errors typically requires human experts to manually inspect and verify a large set of predicted concepts. Existing approaches suffer from a fundamental structural limitation: they either adopt a single static concept set, forcing experts to exhaustively annotate concepts and incurring prohibitive intervention costs, or train multiple models tailored to different concept budgets, resulting in substantial computational and maintenance overhead. To address this challenge, we propose the Matryoshka Concept Bottleneck Model (MCBM), a unified architecture that enables adaptive concept utilization within a single model. Inspired by Matryoshka Representation Learning, MCBM organizes concepts into a nested hierarchy based on maximum relevance and minimum redundancy, allowing inference at multiple levels of conceptual granularity without retraining. Theoretically, we show that MCBM reduces the expected intervention costs from linear to logarithmic order, $O(\log K)$, while guaranteeing monotonic performance improvement. Empirically, extensive experiments demonstrate that MCBM matches the performance of independently trained models while enabling dynamic and efficient expert interaction. --- ### Beyond Routing: Characterising Expert Tuning and Representation in Vision Mixture-of-Experts Authors: Gene Tangtartharakul, Katherine R. Storrs Published: 2026-05-20 Categories: cs.CV, cs.AI arXiv: https://arxiv.org/abs/2605.20610v1 Mixture-of-Experts (MoE) models are often interpreted by analysing which categories are routed to which experts. However, routing alone does not reveal what each expert actually encodes. We train sparsely-gated convolutional MoE models with a contrastive objective on natural images and characterise expert specialisation using tools from visual neuroscience. Extending from gating-level to expert-level analyses, we measure per-expert category separability, and per-expert tuning using the most exciting inputs. Extending from category-level to feature-level explanations, we interpret tuning via semantic dimensions derived from a dataset of human behavioural judgements (THINGS). Finally, we use tuning and representational similarity analysis to assess the stability of expertise-allocation across independent initialisations. We find that an animate-inanimate distinction dominates expert partitioning, apparent from gating through to expert readout, and is stable across independently trained models. Although routing statistics suggest relatively sparse, categorical preferences, expert analyses reveal broader tuning to continuous visual and semantic dimensions that extend beyond category boundaries. Experts exhibit similar category-separability to one another, despite distinct feature tuning, demonstrating the explanatory benefits of moving beyond category-level analyses. Together, these results show that expert specialisation in vision MoEs extends well beyond category routing and is better understood by probing fine-grained expert-level tuning and representational structure. --- ### Compositional Transduction with Latent Analogies for Offline Goal-Conditioned Reinforcement Learning Authors: Junseok Kim, Dohyeong Kim, Mineui Hong, Songhwai Oh Published: 2026-05-20 Categories: cs.LG arXiv: https://arxiv.org/abs/2605.20609v1 Compositional generalization is essential for reaching unseen goals under novel contextual variations in offline goal-conditioned reinforcement learning (GCRL), where a generalist goal-reaching agent must be learned from limited data. Most prior approaches pursue this via trajectory stitching over temporally contiguous segments, which limits composing behaviors across varying contexts. To overcome this limitation, we formalize analogy transduction as synthesizing new plans by composing task-endogenous analogies with given contexts and propose a novel analogy representation tailored for it. Grounded in our theory, this analogy representation captures what changes under optimal task execution, remains invariant to contextual variations, and is sufficient for optimal goal reaching. We further contend that generalization to unseen analogy-context pairs is a practical obstacle in analogy transduction, and introduce a new approach for offline GCRL that enables analogy transduction beyond seen pairs to unseen combinations. We empirically demonstrate the effectiveness of our approach on OGBench manipulation environments, substantially outperforming prior methods that do not perform analogy transduction. Project page: https://rllab-snu.github.io/projects/CTA/ --- ### From Automated to Autonomous: Hierarchical Agent-native Network Architecture (HANA) Authors: Binghan Wu, Shoufeng Wang, Yunxin Liu, Ya-Qin Zhang, Joseph Sifakis, Ye Ouyang Published: 2026-05-20 Categories: cs.AI, cs.NI arXiv: https://arxiv.org/abs/2605.20608v1 Realizing Level 4/5 Autonomous Networks (AN) demands a shift from static automation to agent-native intelligence. Current operations, reliant on rigid scripts, lack the cognitive agency to handle off-nominal conditions. To address this, this letter proposes a hierarchical multi-agent reference architecture enabling high-level autonomy. The framework features a Dual-Driven Orchestrator that coordinates specialized Executive Agents, supported by a shared Public Memory for unified domain knowledge. A key innovation is the integration of agent self-awareness, which empowers the system to harmonize deliberative strategic governance with reflexive fault recovery. We instantiate and validate this architecture within a 5G Core environment. Case studies demonstrate that the system sustains critical throughput under congestion and reduces Mean Time to Repair (MTTR) by 86%, confirming its efficacy in unifying strategic planning with operational resilience. --- ### Mechanistic Interpretability for Learning Assurance of a Vision-Based Landing System Authors: Romeo Valentin, Olivia Beyer Bruvik, Marc R. Schlichting, Mykel J. Kochenderfer Published: 2026-05-20 Categories: cs.LG, cs.CV, cs.RO arXiv: https://arxiv.org/abs/2605.20607v1 EASA's learning-assurance guidance requires data-driven aviation systems to build and monitor their own situation representation, yet for neural networks the technical means to provide such evidence remain an open problem. We address this gap for a vision-based aircraft landing system: we propose that a minimally assurable model must at least be shown to separate content from style in its own situation representation. Showing that the model's predictions then rely largely on the contentful representation components leads to a concrete assurance path. To demonstrate this assurance path on a concrete model we train a vision transformer model for runway keypoint regression on the LARDv2 dataset. The model, which acts as the subject for our assurance demonstration, produces per-patch embeddings that we decompose into interpretable atoms via K-SVD sparse dictionary learning. A qualitative visualization confirms that contentful atoms track task-relevant runway structure and stylistic atoms track domain-specific appearance, and the regression head is shown to place almost all of its linear weight on contentful atoms. We further build on the content/style separation and define out-of-model-scope (OOMS) detection, a novel runtime assurance approach directly monitoring the model's situation representation. OOMS monitoring is complementary to operational design domain and output-space out-of-distribution monitoring and addresses concrete requirements of the recent EASA guidance. By directly analyzing a model's situation representation both at test time and runtime, this work delivers the first concrete piece of the representation-level evidence that EASA learning-assurance guidance demands, and points to mechanistic interpretability as a practical building block of future aviation safety cases. --- ### Mind Your Margin and Boundary: Are Your Distilled Datasets Truly Robust? Authors: Muquan Li, Yingyi Ma, Yihong Huang, Hang Gou, Ke Qin, Ming Li, Yuan-Fang Li, Tao He Published: 2026-05-20 Categories: cs.CV arXiv: https://arxiv.org/abs/2605.20606v2 Dataset distillation (DD) compresses a large training set into a small synthetic set for efficient training, but most DD methods optimize only clean accuracy and leave robustness uncontrolled. Recent robust DD methods improve robustness, yet they often suffer from a poor accuracy-robustness trade-off because they (i) treat all adversarially perturbed examples uniformly, despite robust risk being dominated by near-zero robust margins, and (ii) do not explicitly increase inter-class separation in the decision boundary where attacks concentrate. We present Contrastive Curriculum for Robust Dataset Distillation (C$^2$R), a framework that couples an attack-aware curriculum with a contrastive robustness objective. From a robust-margin perspective, we derive a perturbation score that approximates each sample's robust hinge, enabling a curriculum that prioritizes the smallest-margin adversaries that most directly drive robust error. In parallel, a class-balanced contrastive robustness loss enforces adversarial invariance while explicitly widening boundary separation across classes. Experiments on CIFAR-10/100, Tiny-ImageNet, and multiple ImageNet-1K subsets under six attacks show that C$^2$R achieves the best robust accuracy, outperforming prior robust DD by $2.8$% on average. --- ### Self-Training Doesn't Flatten Language -- It Restructures It: Surface Markers Amplify While Deep Syntax Dies Authors: Ming Liu Published: 2026-05-20 Categories: cs.CL, cs.AI, cs.LG arXiv: https://arxiv.org/abs/2605.20602v1 Successive self-training on a language model's own outputs is widely characterized as a process of flattening: diversity drops, distributions narrow, and the text becomes "more like itself." We provide evidence that this characterization is incomplete. Across eleven generations of self-training on five models (GPT-2 124M, Pythia-410M, Pythia-1.4B, OPT-1.3B, Pythia-2.8B), language is not flattened uniformly -- it is restructured. Surface markers (discourse connectives, hedges, em-dashes) rise, while mid- and deep-syntactic structures (questions, parentheticals, passives, subjunctives) collapse. We formalize this asymmetric collapse as the Structural Depth Hypothesis (SDH): the per-generation decay rate of a linguistic feature is predicted primarily by its structural depth -- the number of nested syntactic dependencies it requires -- and only secondarily by its generation-zero output frequency. Pooling 17-feature panels from five models spanning three architecture families (N=85), the pooled Spearman correlation is rho=0.540 (p < 10^{-6}; cluster-bootstrap 95% CI (0.434, 0.634)), while frequency is a substantially weaker predictor (rho=0.225). A matched human-text fine-tuning control yields rho=0.039 (p=0.88), confirming the gradient is self-training-specific. We further document a Superficial Complexity Paradox: aggregate complexity proxies (dep-tree depth, TTR, word length) all rise as the underlying clause structure dies, with direct implications for training-data curation and LLM-text detection. --- ### Head-Aware Key-Value Compression for Efficient Autoregressive Image Generation Authors: Guotao Liang, Baoquan Zhang, Zhiyuan Wen, Yunming Ye Published: 2026-05-20 Categories: cs.CV arXiv: https://arxiv.org/abs/2605.20600v1 Autoregressive (AR) visual generation has achieved remarkable performance but suffers from high memory usage and low throughput, as it requires caching previously generated visual tokens. Recent research has shown that retaining only a few lines of cache tokens can maintain high-quality images while significantly reducing memory usage and improving throughput. However, these methods allocate a fixed budget to each attention head, overlooking the heterogeneity among attention heads, leading to suboptimal memory allocation. In this paper, we observe that attention heads across different layers exhibit diverse attention patterns, where some heads focus on local neighborhoods while others capture broader contextual dependencies. Based on this insight, we propose a novel head-aware key-value (KV) cache compression framework for autoregressive image generation, called HeadKV, which assigns smaller budgets to locality-biased heads and larger budgets to heads with broader attention. A key challenge lies in identifying the type of each attention head to guide cache compression. We further observe that, within the same layer, each head exhibits consistent attention patterns across token positions, \emph{i.e.}, a head's behavior for early tokens remains consistent with that for later tokens. This insight suggests that head types can be identified during the early stage and reused for KV compression throughout generation. Its advantage is that it requires no additional training or dataset-level statistics and generalizes seamlessly across different inputs. Moreover, we design a Stratified Token Eviction strategy to effectively preserve long-range information. Extensive experiments demonstrate its effectiveness across multiple autoregressive image generation models. --- ### Unsupervised clustering and classification of upper limb EMG signals during functional movements: a data-driven Authors: L. F. Salazar Álvarez, D. Escobar-Saltarén, M. B. Salazar Sánchez, S. C. Henao-Aguirre Published: 2026-05-20 Categories: cs.LG arXiv: https://arxiv.org/abs/2605.20599v1 This study presents a comprehensive approach for the clustering and classification of upper-limb surface electromyography (sEMG) signals during functional reach and grasp movements. The methodology was applied to the NINAPRO DB4 dataset, which provides multichannel EMG recordings of 52 gestures. A four-stage pipeline was designed, including signal preprocessing, fea-ture extraction, gesture selection via hierarchical clustering, and comparative model evaluation. Preprocessing involved a fourth-order low-pass filter (0.6 Hz) and Hilbert envelope transformation, effectively reducing noise and enhancing signal clarity. Feature extraction yielded 26 temporal and frequency-domain met-rics, which were later refined using visual analysis, mutual information, principal component analysis, and decision tree importance scores. A final subset of five key features was selected for classification tasks. Gesture selection was per-formed through hierarchical clustering using Mahalanobis distance, resulting in six representative movements that balanced biomechanical diversity and compu-tational efficiency. A 200 ms window was identified as optimal for temporal seg-mentation based on stability and physiological plausibility. Classifier models were evaluated in two stages. Automated comparison using PyCaret identified Extra Trees (ET) and Artificial Neural Networks (ANN) as top performers. Sub-sequent independent training confirmed their stability and generalization capac-ity, with ANN showing progressive learning and ET maintaining robust, con-sistent results. The findings support the implementation of adaptive, low-latency control strategies for myoelectric prostheses and provide a scalable pipeline for future real-time applications. --- ### ReversedQ: Opportunities for Faster Q-Learning in Episodic Online Reinforcement Learning Authors: Sofia R. Miskala-Dinc, Aviva Prins Published: 2026-05-20 Categories: cs.LG arXiv: https://arxiv.org/abs/2605.20592v1 We study model-free Q-learning in finite-horizon episodic Markov Decision Processes (MDPs) with stationary dynamics across episodes. We identify a central issue in nascent model-free posterior-sampling works: the reliance on delayed learning in order to prove theoretical guarantees. In particular, we identify three opportunities for faster learning - (i) value-function update order, (ii) update frequencies, and (iii) value-function initialization. Using Wang et al.'s RandomizedQ as a basis, we illustrate these changes and their individual (as well as cumulative) impact in multiple empirical studies. We find that our combined modifications, termed ReversedQ, improve scaled mean cumulative reward compared to RandomizedQ, from 9.53% to 78.78% in the Bidirectional Diabolical Combination Lock (BDCL), and from 21.76% to 61.81% in a chain MDP. --- ### Do No Harm? Hallucination and Actor-Level Abuse in Web-Deployed Medical Large Language Models Authors: Sunday Oyinlola Ogundoyin, Muhammad Ikram, Rahat Masood Published: 2026-05-20 Categories: cs.CL, cs.CY arXiv: https://arxiv.org/abs/2605.20591v1 Medical large language models (LLMs), including custom medical GPTs (MedGPTs) and open-source models, are increasingly deployed on web platforms to provide clinical guidance. However, they pose risks of hallucination, policy noncompliance, and unsafe design. We conduct a large-scale assessment of 6,233 MedGPTs, evaluating a stratified sample of 1,500, together with 10 open-source LLMs. We introduce two frameworks: MedGPT-HEval for hallucination detection and an LLM-based pipeline for assessing policy violations and developer intent. Our results show that 25-30% of MedGPTs exhibit low factual accuracy, with bottom- and middle-tier models at highest risk; 33.6-54.3% violate operational thresholds, and 57.06% of Action-enabled models lack adequate privacy disclosures. Compared with open-source models, MedGPTs achieve higher factual accuracy and semantic alignment, though open-source models are more stable. These results reveal systemic gaps in hallucination and compliance, highlighting the need for multi-metric evaluation and stronger safeguards. We release HAA-MedGPT, a structured dataset that supports future research on the safety of web-facing medical LLMs. --- ### Direct Translation between Sign Languages Authors: Zetian Wu, Bowen Xie, Wuyang Meng, Milan Gautam, Stefan Lee, Liang Huang Published: 2026-05-20 Categories: cs.CL, cs.CV arXiv: https://arxiv.org/abs/2605.20588v1 The field of sign language translation has witnessed significant progress in the translation between sign and spoken languages, but the translation between sign languages remains largely unexplored and out of reach. The latter can help 1.5 billion deaf and hard-of-hearing (DHH) people worldwide communicate across language barriers without relying on hearing interpreters or written-language fluency. The cascade approach composing separate sign-to-text, text-to-text, and text-to-sign systems suffers from error propagation and extra latency as well as the loss of information unique in the visual modality. We aim to develop direct sign-to-sign translation. However, a large-scale open-domain parallel corpus has not been curated between sign languages. To enable direct translation between sign language utterances, we use back-translation to produce synthetic sign-sign pairs from unaligned individual language utterance-sign corpora. Using this data, we jointly train a single MBART-based model for both text->sign (T2S) and sign->sign (S2S). On synthetically generated paired sets between American Sign Language (ASL), Chinese Sign Language (CSL), and German Sign Language (DGS), our direct S2S method outperforms the cascaded baseline on geometric sign error metrics (20% lower DTW-aligned MPJPE) and language matching metrics after predicted sign utterances are translated back to sentences (50% high BLEU-4) while achieving a roughly 2.3* speedup. On a small set of pre-existing cross-lingual sign data, we find similar improvements for our proposed method. --- ### QwenSafe: Multimodal Content Rating Description Identification via Preference-Aligned VLMs Authors: Dishanika Denipitiyage, Aruna Seneviratne, Suranga Seneviratne Published: 2026-05-20 Categories: cs.CV arXiv: https://arxiv.org/abs/2605.20584v1 Mobile app marketplaces require developers to disclose standardized content rating descriptors (CRDs) to inform users about potentially sensitive or restricted content. Ensuring the accuracy and consistency of these disclosures remains challenging due to the multimodal nature of app content, which spans textual descriptions and visual interfaces. In this paper, we present QwenSafe, a Vision-Language Model (VLM) designed to automatically identify the presence of Apple-defined CRDs by jointly reasoning over app metadata and screenshots. To enable scalable training for this task, we introduce metadata2CRD, a data-construction pipeline that synthesizes descriptor-aligned question-answer pairs by combining app descriptions, screenshots, and formal descriptor definitions. We adapt Qwen3-VL-8B using supervised fine-tuning followed by Direct Preference Optimization (DPO) to align model predictions with descriptor-specific evidence and explanations across visual and textual modalities. We evaluate QwenSafe on 12 Apple-defined content rating descriptors and compare it against state-of-the-art vision-language models, including Qwen3-VL, LLaVA-1.6, and Gemini-2.5-Flash. QwenSafe consistently outperforms all baselines in binary CRD classification, achieving improvements in positive-class recall of 111.8%, 36.1%, and 2.1%, respectively. Our results demonstrate that descriptor-aware multimodal alignment substantially improves automated content classification and highlights the potential of vision-language models to support scalable and consistent content rating in mobile app marketplaces. --- ### TriForces: Augmenting Atomistic GNNs for Transferable Representations Authors: Ali Ramlaoui, Alexandre Duval, Hannah Bull, Victor Schmidt, Hugues Talbot, Fragkiskos D. Malliaros, Joseph Musielewicz Published: 2026-05-20 Categories: cs.LG, cond-mat.mtrl-sci arXiv: https://arxiv.org/abs/2605.20581v1 Machine learning interatomic potentials (MLIPs) achieve excellent accuracy when trained on large Density Functional Theory (DFT) data. To be useful in practice, they must often be adapted to target chemistries using small and expensive task-specific datasets. However, MLIPs transfer inconsistently across domains, with representations that often loose accessible composition and structure information. To address this, we present TriForces, a model-agnostic three-stream framework that separates composition and structure information, combined with self-supervised learning to preserve transferable representations. TriForces improves performance on MatBench and QM9 over baselines without needing DFT labels and enables efficient similar structure retrieval through its learned latent space. On OMat24, in limited-data training regime, TriForces reduces energy MAE by 57% at 20K samples only and improves force MAE across sample sizes. We release pretrained TriForces variants across multiple MLIP architectures with code at https://github.com/Ramlaoui/triforces. --- ### Deep Learning Surrogates for Emulating Stochastic Climate Tipping Dynamics Authors: Adeline Hillier, Jennifer Sleeman, Jay Brett, Caroline Tang, Jenelle Millison, Anand Gnanadesikan Published: 2026-05-20 Categories: cs.LG arXiv: https://arxiv.org/abs/2605.20580v1 This work explores a dynamics-informed Temporal Fusion Transformer (TFT) as a data-driven surrogate for computationally intensive Earth system simulations. Focusing on multivariate time series describing global ocean transport, we demonstrate the surrogate's ability to forecast tip events across thousands of time steps. The data involve up to 21 non-stationary time series in addition to static covariates describing free parameters and initial conditions. Modifications to the architecture and objective function yield a surrogate that anticipates the timing of Atlantic and Pacific collapses to high fidelity and captures the stochastic uncertainty in transition timing across ensemble predictions. The learned surrogate achieves a 465x computational speedup over the numerical simulator while maintaining differentiability with respect to parameters and initial conditions. --- ### A strongly annotated passive acoustic dataset for tropical bird monitoring Authors: Daniela Ruiz, Juan Sebastián Ulloa, Zhongqi Miao, Nicolás Betancourt, Maria Paula Toro-Gómez, Andrés Hernández, Bruno Demuro, Eliana Barona-Cortés, Angela Mendoza-Henao, Andrés Sierra-Ricaurte, Sebastián Pérez-Peña, Rahul Dodhia et al. Published: 2026-05-20 Categories: cs.SD, cs.CV arXiv: https://arxiv.org/abs/2605.20578v2 Passive acoustic monitoring enables continuous, non-invasive biodiversity assessment across diverse ecosystems. The scale of these datasets has driven the adoption of machine learning, with supervised approaches showing strong performance. However, supervised methods require time-resolved annotated datasets, which remain scarce, especially in complex tropical soundscapes. We present PteroSet, a curated dataset of strongly annotated Neotropical bird vocalizations recorded in Puerto Asis (Putumayo) and Pivijay (Magdalena), Colombia, between 2023 and 2025. The dataset comprises 563 recordings (73.62 h) and 15,372 time-frequency annotations, including 6,702 events identified to the species level across 168 species. We release the annotations in a COCO-inspired JSON schema that unifies audio files, taxonomic categories, and labels for machine learning workflows. Beyond providing annotated data, PteroSet serves as a realistic benchmark that highlights key characteristics of tropical soundscapes, including acoustic co-occurrence and domain shift across recording sites. We provide a deep learning baseline for binary bird detection, demonstrating PteroSet's usability and the challenges it presents. --- ### Mahjax: A GPU-Accelerated Mahjong Simulator for Reinforcement Learning in JAX Authors: Soichiro Nishimori, Shinri Okano, Keigo Habara, Sotetsu Koyamada, Eason Yu, Masashi Sugiyama Published: 2026-05-20 Categories: cs.AI, cs.LG arXiv: https://arxiv.org/abs/2605.20577v1 Riichi Mahjong is a multi-player, imperfect-information game characterized by stochasticity and high-dimensional state spaces. These attributes present a unique combination of challenges that mirror complex real-world decision-making problems in reinforcement learning. While prior research has heavily relied on supervised learning from human play logs to pre-train the policy, algorithms capable of learning \textit{tabula rasa} (from scratch) offer greater potential for general applicability, as evidenced by the AlphaZero lineage. To facilitate such research, we introduce \textbf{Mahjax}, a fully vectorized Riichi Mahjong environment implemented in JAX to enable large-scale rollout parallelization on Graphics Processing Units (GPUs). We also provide a high-quality visualization tool to streamline debugging and interaction with trained agents. Experimental results demonstrate that Mahjax achieves throughputs of up to \textbf{2 million} and \textbf{1 million steps per second} on eight NVIDIA A100 GPUs under the no-red and red rules, respectively. Furthermore, we validate the environment's utility for reinforcement learning by showing that agents can be trained effectively to improve their rank against baseline policies. --- ### $Δ$ynamics: Language-Based Representation for Inferring Rigid-Body Dynamics From Videos Authors: Chia-Hsiang Kao, Cong Phuoc Huynh, Chien-Yi Wang, Noranart Vesdapunt, Stefan Stojanov, Bharath Hariharan, Oleksandr Obiednikov, Ning Zhou Published: 2026-05-20 Categories: cs.CV arXiv: https://arxiv.org/abs/2605.20576v1 Inferring rigid-body physical states and properties from monocular videos is a fundamental step toward physics-based perception and simulation. Existing approaches assume specific underlying physical systems, object types, and camera poses, making them unable to generalize to complex real-world settings. We introduce $Δ$YNAMICS, a vision-language framework that uses language as a unified representation of rigid-body dynamics. Instead of directly predicting parameters, $Δ$YNAMICS generates scene configurations in a structured text format for physics simulation. We enhance the model's generalization by integrating natural language motion reasoning and leveraging optical flow as a semantic-agnostic input. On the CLEVRER dataset, $Δ$YNAMICS achieves a segmentation IoU of 0.30, a 7x improvement over leading VLMs (InternVL3-8B, Qwen2.5-VL-7B and Claude-4-Sonnet). Additionally, test-time sampling and evolutionary search further boost performance by 27% and 120% in segmentation IoU, respectively. Finally, we demonstrate strong transfer to a new dataset of 235 real-world rigid-body videos, highlighting the potential of language-driven physics inference for bridging perception and simulation. --- ### End-to-End Unmixing with Material Prompts for Hyperspectral Object Tracking Authors: Xu Han, Mohammad Aminul Islam, Lei Wang, Zekun Long, Guanmanyi Fu, Wangshu Cai, Kuldip K. Paliwal, Jun Zhou Published: 2026-05-20 Categories: cs.CV arXiv: https://arxiv.org/abs/2605.20569v1 Hyperspectral imagery encodes rich material properties that can improve tracking robustness under appearance ambiguity, illumination change, and background clutter. However, due to the limited availability of hyperspectral video data, many existing methods adapt pretrained RGB trackers via spatial or channel fusion strategies, largely neglecting the intrinsic material information in hyperspectral imagery. Moreover, the few material-aware approaches typically rely on external spectral unmixing pipelines that are decoupled from the tracking objective, limiting effective optimization of material representations for target localization. To address these limitations, we formulate hyperspectral object tracking as a joint optimization problem of material decomposition and target localization, coupling the two tasks via a weighted target-oriented unmixing loss that explicitly aligns material representations with localization accuracy. Specifically, we propose a material representation decomposition module for deep learning-based spectral unmixing with adaptive frequency decomposition. Building on the decomposed material representations, we further introduce a dual-branch wavelet-enhanced material prompt module that learns low- and high-frequency material prompts through efficient spatial-material interactions in the frequency domain. The framework is model-agnostic and can be seamlessly generalized to different unmixing backbones. Extensive experiments on standard hyperspectral tracking benchmarks demonstrate state-of-the-art performance and validate the effectiveness of the proposed end-to-end material-aware tracking framework. Code is available at https://github.com/han030927/E2EMPT. --- ### KernelBrain: Coarse-to-Fine, Budget-Aware Search for Agentic GPU Kernel Optimization Authors: Shuai Che, Gang Peng Published: 2026-05-19 Categories: cs.DC, cs.AI, cs.LG, cs.SE arXiv: https://arxiv.org/abs/2608.02611v1 Automating GPU kernel optimization remains difficult in practice: generated variants can violate correctness constraints, runtime measurements are noisy, and search often stalls early. We present a practical optimization agent that combines LLM-guided mutation, adaptive resource allocation, policy-gated evaluation, and profiler-informed diagnosis. The system screens many candidates with low-cost evaluation and allocates higher-fidelity budget only to promising survivors to optimize and evolve GPU kernels. On important Triton kernel generation tasks, this design improves both kernel quality and search efficiency, reaching 0.88x-6.72x speedup over PyTorch and up to 1.4x speedup over the state-of-the-art kernel agent, with up to 48% lower optimization time. --- ### The Asymmetric Effects of Knowledge Distillation on Bias in Small Language Models Authors: Plawan Kumar Rath Published: 2026-05-19 Categories: cs.CL, cs.AI, cs.CY arXiv: https://arxiv.org/abs/2607.28639v2 We show that knowledge distillation (KD) in small instruction-tuned language models has asymmetric effects on bias, and that measuring them correctly requires accounting for where refusal mass moves and what the parser can legitimately score. On unambiguous tasks (BBQ-disambig), response-based distillation from a Mistral-7B teacher genuinely improves context-following for the most context-biased baseline (SmolLM2-1.7B-Instruct): among committed (non-abstaining) answers, the rate of overriding correct context with a stereotype falls from 44.5% to 37.2%, with accuracy rising from 0.55 to 0.61. On ambiguous tasks (BBQ-ambig), the same distillation degrades conditional refusal: 15% of the cases where the baseline correctly abstained instead receive stereotype answers (silence-loss), and the distilled refusal pattern only weakly preserves the baseline's (Spearman rho=0.44). The harm reproduces, aggravated, on a second student family (OLMo-2-1B-Instruct): silence-loss reaches 49% and filled-silence accounts for 95% of new bias. Two apparently stronger results are artifacts. An unconditioned override metric reports a 44% -> 23% improvement under a Gemma-2-9B teacher that shrinks to 44.5% -> 39.8% once conditioned on committed answers: the model abstains on 43% of items and its accuracy collapses from 0.55 to 0.35. An apparent cross-condition independence reverses to a positive correlation (rho=0.58, p<0.01) on the valid 19-configuration grid once parser-invalid logit-KD configurations are excluded and the parser is corrected. Aggregate metrics (CrowS-Pairs, overall BBQ Stereotype Reliance Score) average over both effects and conceal the per-item harm. We propose Per-Condition Calibration Diagnosis (PCCD), a three-step protocol evaluating refusal-pattern preservation, committed-answer context-following, and capability preservation. No configuration in our grid passes all three steps. --- ### Learning Stateful Predictive Knowledge From Experience Authors: Yan Song, Xidong Feng, Bo Liu, Xinyu Cui, Haotian Fu, Zichen Liu, Mengyue Yang, Cheng Deng, Jian Zhao, Jun Wang Published: 2026-05-19 Categories: cs.CL, cs.LG arXiv: https://arxiv.org/abs/2607.28638v1 As large language model (LLM) agents increasingly learn from experience, they primarily rely on trajectory-level reflection to extract insights. Viewed through the lens of predictive knowledge, we argue that this approach operates on episodic hindsight rather than predictive foresight, yielding brittle, path-dependent heuristics. To address this, we propose Stateful Knowledge Learning (SKL). SKL shifts the agent's focus from trajectory-level summarization to maintaining Stateful Knowledge: explicit, declarative predictive assessments anchored to state. We first demonstrate a motivating example showing how stateful knowledge provides granularity, enhances generalization, and enables knowledge bootstrapping. To further scale up the idea, we introduce two algorithms via self-distillation (SKL-SD) and reinforcement learning (SKL-RL), training agents to autonomously extract state-grounded predictive knowledge from experience and learn to leverage it for policy making. Experiments on interactive environments (WebShop, ScienceWorld) and a complex reasoning task (ChessPuzzles) demonstrate that equipping models with the inherent ability to learn stateful predictive knowledge significantly outpaces current reflection-based training paradigms. --- ### ZeroR@CHiPSAL 2026: Two-Stage Vision-Language Adaptation with Contrastive Learning for Nepali Meme Classification Authors: Nitiz Khanal Published: 2026-05-19 Categories: cs.CL arXiv: https://arxiv.org/abs/2607.28637v1 This paper presents our system for the CHiPSAL 2026 shared task on multimodal hate speech and sentiment detection in Nepali memes. We address both subtasks: binary hate speech classification and three-class sentiment analysis. Our approach adapts the Robust Adaptation of Hateful Meme Detection (RA-HMD) framework using Qwen3-VL-8B-Instruct, a state-of-the-art vision-language model with native Devanagari support. We employ a two-stage training pipeline: (1) LoRA fine-tuning with an MLP projection head for generative classification, and (2) contrastive backbone fine-tuning with supervised InfoNCE loss. We handle class imbalance through minority oversampling, image augmentation, and focal loss. At inference, we ensemble Stage 1 token probabilities with Stage 2 classifier scores using validation-tuned weights. Our end-to-end approach eliminates error propagation from separate OCR and translation pipelines by leveraging the model's native Devanagari understanding. Our system achieved \textbf{2nd place} on hate speech detection (F1: 0.797) and \textbf{4th place} on sentiment analysis (F1: 0.518). We provide detailed ablations, error analysis, and insights into adapting large vision-language models for low-resource South Asian languages. --- ### Chain-of-Models: Cross-Model Auditing for Bias-Robust LLM Judges Authors: Qian Wang, Zhanzhi Lou, Zhenheng Tang, Nuo Chen, Bingsheng He Published: 2026-05-19 Categories: cs.CL, cs.CY arXiv: https://arxiv.org/abs/2607.28636v1 LLMs increasingly serve as automated judges, but their judgments remain vulnerable to cognitive biases. Existing mitigations mostly rely on prompt-driven debiasing, which is brittle across bias types, or human evaluation, which does not scale. We study \emph{Chain-of-Models} (CoM), an automated audit pipeline in which a second model inspects the first model's reasoning trace before producing the final judgment. The key design question is whether the auditor should be the same model, a same-family model, or a different-family model. Across 9 models from 6 families, 4 cognitive biases, and 4 factual datasets, we find that auditor identity matters in two ways. First, standalone bias resistance does not predict audit effectiveness: Kimi-K2.5 is the strongest standalone model on several biases, yet is a weak auditor for Qwen2.5-72B's biased traces. Second, the best auditor is bias-specific: GPT-4o is strongest on bandwagon, authority, and distraction, while GLM-5 is strongest on sycophancy. We operationalize these findings with a per-bias auditor selection rule that, given the bias type, scores candidates along functional diversity, per-bias standalone resistance, and calibrated audit effectiveness. Under a calibration/test split, the selector reaches the highest accuracy across the four biased slices ($0.884$ vs.\ $0.824$ for the strongest single fixed auditor and $0.805$ for the no-audit baseline). We release data, configurations, and an LLM-agent skill at https://anonymous.4open.science/r/chain-of-models-B585 . --- ### Imbalanced Data Clustering via Targeted Data Augmentation Using GMM and LLM Authors: Noor Khalal, Abdallah Alaa-Eddine Djamai, Imed Keraghel, Mohamed Nadif Published: 2026-05-19 Categories: cs.CL, cs.LG arXiv: https://arxiv.org/abs/2607.28635v1 In Natural Language Processing (NLP), dealing with underrepresented topics is challenging, especially in unsupervised tasks where clustering might not adequately capture minority topics. To tackle this challenge, our paper presents a novel unsupervised data augmentation method that integrates Gaussian Mixture Models (GMMs) and Large Language Models (LLMs). Due to their flexibility and robustness, GMMs can detect clusters corresponding to underrepresented areas in the data, while LLMs create synthetic documents to enrich these clusters and improve their representation. Experiments on various imbalanced text datasets demonstrate that our approach preserves clustering performance in all cases and often enhances cluster interpretability, offering a robust and scalable solution for improving data representation in unsupervised NLP tasks. --- ### DeepLens Diagnosis Agent: Agentic Workflow Design Lets a Small Reasoning Model Compete with Frontier LLMs Authors: Mahmood Bayeshi, Veysel Kocaman, Muhammed Ali Naqvi, Yigit Gul, David Talby Published: 2026-05-19 Categories: cs.AI arXiv: https://arxiv.org/abs/2607.22555v1 Medical diagnosis is a multi-stage process: extract facts, consult knowledge, generate a differential analysis, and select the best diagnosis with explanations. Frontier LLMs are strong generalists, but single-shot prompting often yields brittle diagnostic reasoning. We present the DeepLens Diagnosis Agent, a five-stage harnessing pipeline (combining model capabilities with disciplined process constraints) centered on a small medical reasoning model (JSL Medical Small 7B v2) and retrieval-augmented generation (RAG). The pipeline enforces structured clinical extraction, disciplined retrieval, constrained candidate generation, explicit evidence triangulation, and an auditable final decision. On the 915-case DiagnosisArena benchmark, the agent achieved 60.14% top-1 diagnostic accuracy, the highest among small and medium-sized models. The same model without the agent workflow achieved 23.99%, a +36-point gain from workflow design alone, despite 88.2% on standard medical benchmarks, showing that diagnostic reasoning under uncertainty requires more than knowledge recall. The agent costs USD 0.0072 per case (24K tokens on A100) with 24-second latency, 35-45% cheaper than Claude Sonnet 4.5 (USD 0.0110) and Gemini 3.1 Pro (USD 0.0128) while outperforming them by +9.70pp and +9.17pp. Harnessing can also correct frontier model failures; workflow constraints can outweigh parameter count or API cost. Beyond aggregate accuracy, the pipeline produces structured intermediate artifacts that make each stage inspectable and support error localization. These properties support high-stakes settings where traceability, reproducibility, and auditable evidence matter alongside benchmark performance. --- ### DC-Leap: Training-Free Acceleration of dLLMs via Draft-Guided Contiguous Leaping Decoding Authors: Yanhua Jiao, Tianyi Wu, Xiaoxi Sun, Yulin Li, HuiLing Zhen, Libo Qin, Baotian Hu, Zhuotao Tian, Min Zhang Published: 2026-05-19 Categories: cs.AI, cs.LG arXiv: https://arxiv.org/abs/2607.20467v1 While parallel decoding is central to the efficiency of Diffusion Large Language Models (dLLMs), current strategies are often hindered by overly conservative confidence thresholds. These thresholds, necessitated by the Joint Probability Dependence Error (JPDE), result in redundant denoising iterations and suboptimal inference speeds. To overcome this, we propose DC-Leap, a training-free framework that enables reliable acceleration of dLLMs in the moderate-confidence regime. DC-Leap introduces a Dynamic Contiguous Verification strategy that integrates strictly-ordered causal constraints into the parallel decoding process. By progressively validating token dependencies, this mechanism effectively neutralizes the JPDE, enabling reliable acceleration with comparable performance. Furthermore, DC-Leap incorporates the draft-guided decoding mechanism, where the draft helps extend the context by leaping forward across multiple tokens, providing look-ahead context and retaining the structural benefits of bidirectional attention during inference. Extensive experiments on standard benchmarks demonstrate that DC-Leap achieves substantial speedups, up to 53.19x on MBPP for long-sequence generation, and up to 105.02x when combined with KV-Cache with comparable generation quality. Code is available at https://github.com/ffh-wyls/DC-Leap . --- ### JAXBench: Benchmarking Autonomous TPU Kernel Optimization Authors: Arya Tschand, Charles Hong, Julian Walker, Nina Cai, Shangkun Wang, Suvinay Subramanian, Sundar Dev, Vijay Janapa Reddi, Amir Yazdanbakhsh, Sethu Sankaran Published: 2026-05-19 Categories: cs.AI arXiv: https://arxiv.org/abs/2607.20466v1 Rigorous benchmarks have driven progress in autonomous GPU kernel performance optimization by establishing a shared target to hillclimb on, but no equivalent exists for TPUs. We present JAXBench, a TPU-native benchmark suite for AI-generated kernel optimization on Google Cloud TPUs. JAXBench comprises 50 JAX workloads that are both relevant and provide headroom for optimization. We extract 17 production ML operators from architectures in the public MaxText library such as Llama-3.1, DeepSeek-V3, Mixtral, Mamba-2, and AlphaFold2, and translate 33 operators from KernelBench that are validated for correctness and set with new problem sizes that achieve high TPU v6e MXU utilization. Eight of the 17 production operators ship with hand-optimized Pallas kernels from the public Tokamax library and block-size tuned to establish an expert upper-bound baseline. We evaluate four feedback-driven methods on generating candidate Pallas kernels for JAXBench. Across the full suite with Gemini 3 Flash, we find that target-specific context matters more than model scale on a sparsely-documented DSL like Pallas. Conditioning on curated TPU documentation raises per-sample correctness from 5.8% to 37.3% and solves 48 of 50 benchmarks at a 1.28x geomean speedup. Search structure yields significant gains once correctness is achieved, with Autocomp's beam-search pipeline reaching a 1.36x geomean speedup over XLA. On the 8 hand-tuned kernels, Autocomp reaches 1.60x geomean over XLA, recovering most of the 2.08x Tokamax upper bound but trailing on the specialized paged and ragged attention operators. High-quality TPU kernel optimization remains a challenging task, and we release the JAXBench benchmark, evaluation harness, and baseline results to support open source contributions. --- ### DataPrep-Bench: Benchmarking LLMs as Training Data Preparators Authors: Hao Liang, Qifeng Cai, Yibo Lin, Jianzhuo Du, Qifeng Xia, Sizhe Qiu, Linzhuang Sun, Meiyi Qiang, Zhaoyang Han, Xiaochen Ma, Bohan Zeng, Ruichuan An et al. Published: 2026-05-19 Categories: cs.LG, cs.CL arXiv: https://arxiv.org/abs/2607.20465v1 The quality of training data fundamentally determines the capabilities of large language models (LLMs), yet no unified benchmark exists to measure how well LLMs, agents, and data-centric workflows actually prepare training data end to end. We view LLM-driven data preparation as comprising two complementary capabilities: data construction, which transforms raw sources into supervised training data, and data quality evaluation, which predicts the training value of candidate datasets before downstream training; throughout, "quality" refers to downstream training utility rather than surface-level textual properties. We introduce DataPrep-Bench, the first unified benchmark that jointly evaluates both capabilities under a shared downstream-grounded protocol over six domains and multiple base models. For data construction, methods consume identical raw sources and are scored by fine-tuning a base model on their outputs jointly with Dolly-15k; alongside this track we release Data-Construction-Skill, a skill-guided agent that lifts the Dolly-only baseline by nearly 20 points absolute on Llama-3.1-8B Finance and is competitive with the strongest agent- and DataFlow-based methods in knowledge-extraction-dense domains. For data quality evaluation, scoring functions are scored by Pearson correlation with downstream performance on a shared candidate pool; we release the Distributional Alignment Score (DAS), a distribution-based evaluator that uses MMD between a candidate dataset and a domain proxy. DAS attains the strongest cross-model correlation in four of six domains and is the only metric clearing r > 0.70 simultaneously in Math, Science, and Medical, outperforming existing quality-, diversity-, and heuristic-based evaluators. DataPrep-Bench provides a unified, downstream-grounded framework for measuring progress on both capabilities as co-equal targets of LLM-driven data preparation. --- ### MUX: Continuous Reasoning via Multiplexed Tokens Authors: Ayhan Suleymanzade, Halil Alperen Gozeten, Michael Bronstein, İsmail İlkan Ceylan, Jinwoo Kim Published: 2026-05-19 Categories: cs.AI, cs.CL, cs.LG arXiv: https://arxiv.org/abs/2607.18264v1 Language models solve complex problems by articulating intermediate reasoning steps in natural language. While effective, this process is computationally bottlenecked: each reasoning step conveys only a single subword, and many are spent expressing a thought instead of carrying out computation. We propose MUX, a simple method for high-bandwidth and compact reasoning based on distillation of discrete reasoning into continuous multiplexed tokens in a latent space. Here, each latent token is trained to represent a weighted linear superposition (multiplexing) of a span of discrete reasoning subwords, where this superposition is lossless by construction and the span can be fully recovered (demultiplexing). We prove that simple position-dependent weightings, such as suitable geometric decay, support lossless multiplexing, which in turn prevents shortcut behaviors caused by latent collapse. We further show that multiplexed reasoning can perform parallel exploration in problems that require search. Across 32 evaluation settings spanning four language models, MUX outperforms strong latent reasoning baselines. Ablation and probing analyses further show that the learned latent tokens encode faithful and interpretable reasoning. Our results suggest that lossless superposition as local learning targets constitutes a sufficient condition for achieving strong and efficient latent continuous reasoning. --- ### SelKV: Selective KV Cache Merging with Per-Token Merge-or-Drop and Attention Compensation Authors: Soumia Bouyahiaoui, Manel Kara laouar, Aicha Boutorh, Mohamed Hadj Ameur Published: 2026-05-19 Categories: cs.AI arXiv: https://arxiv.org/abs/2607.16213v1 Large Language Models (LLMs) generate text autoregressively, relying on a key-value (KV) cache whose memory footprint grows linearly with context length, creating a major bottleneck. Recent compression methods mitigate this cost via token merging; however, these approaches often rely on indiscriminate aggregation, which degrades representations and introduces attention sag, a mismatch where merged tokens receive the same softmax mass as individual tokens despite encoding multiple inputs. We propose a training-free, dual-component framework for KV cache compression that addresses these limitations. First, a soft cosine gate adaptively modulates merging decisions based on value-vector similarity, suppressing or discarding dissimilar tokens to preserve semantic fidelity. Second, we introduce an attention-ratio compensation mechanism that applies a decoding-time logit bias derived from prefill attention statistics, correcting the softmax imbalance induced by merging. Evaluated on LongBench (16 English datasets) while retaining only 25% of the KV cache, our framework achieves strong compressed performance against representative one-shot baselines. It is especially robust on the evaluated grouped-query attention (GQA) models, maintaining nearlossless generation quality. Furthermore, the method outperforms the full-cache baseline on complex multi-document QA tasks and delivers a 3.3x decoding speedup at 100k tokens. --- ### Symbolic Augmentation Closes a Canonical-Equivalence Blind Spot in Neural Fact-Checkers Authors: Genpei Zhang Published: 2026-05-19 Categories: cs.AI arXiv: https://arxiv.org/abs/2607.16212v1 Large language models hallucinate numbers and units when summarizing scientific text, a failure mode that can silently invert a scientific claim. We recast the detection of such errors as typed verification: we introduce a five-class typed-quantity error taxonomy and a 1500-item benchmark, rewritten from PMC and arXiv sources and labeled by two independent LLM annotators with adjudication (Krippendorff's alpha = 0.882). A ModernBERT encoder fine-tuned on this benchmark reaches macro-F1 = 0.899, far above any off-the-shelf neural fact-checker, yet four probes expose a sharp structural blind spot: on canonical-equivalent rewrites of physically equivalent quantities (e.g., 95°C and 368.15 K) its accuracy collapses to 36.5%. We propose Symbolic Augmentation, a training-time framework that runs the modules of a symbolic verifier in reverse to generate label-preserving augmented training data. The augmentation lifts canonical-equivalence robustness to 98.2% while slightly improving in-distribution accuracy (macro-F1: 0.899 to 0.902); the augmented encoder matches a closed-frontier LLM at no inference cost and transfers to an external benchmark (SciFact-Open binary macro-F1: 0.791 to 0.828). Two negative results sharpen the claim: symbolic features as auxiliary encoder inputs add nothing, and symbolic silver labels scale negatively under teacher noise. Together these results identify training-time augmentation as the right integration point between symbolic and learned components. --- ### OriginBlame: Record- and Token-Level Data Provenance for AI Training Datasets Authors: Haolin Xue Published: 2026-05-19 Categories: cs.AI arXiv: https://arxiv.org/abs/2607.13037v2 When a data contributor requests removal, model trainers face a practical gap: unlearning algorithms require a forget set, yet no tool can locate which training records belong to a given author. Existing provenance systems operate at file or dataset level, forcing catastrophic over-deletion. We present ob, a record- and token-level data provenance system that propagates author identity through data processing pipelines and resolves revocation requests into precise forget sets via deterministic queries. Evaluation on 219,555 Wikipedia pages demonstrates that record-level provenance eliminates dataset-level over-deletion (from 101x to 1.3x), while integration adds 1.3-4.0% throughput overhead (HuggingFace) and 2.1-19.0% (Datatrove) on wiki data. On a 1.7B model, provenance-based forget sets consistently reduce the collateral damage of machine unlearning (retain perplexity) relative to same-size random baselines across all evaluated authors, with membership-inference tests indicating they select genuinely memorized content. --- ### OriginBlame: Record- and Token-Level Data Provenance for AI Training Datasets Authors: Haolin Xue Published: 2026-05-19 Categories: cs.AI arXiv: https://arxiv.org/abs/2607.13037v1 When a data contributor requests removal, model trainers face a practical gap: unlearning algorithms require a forget set, yet no tool can locate which training records belong to a given author. Existing provenance systems operate at file or dataset level, forcing catastrophic over-deletion. We present ob, a record- and token-level data provenance system that propagates author identity through data processing pipelines and resolves revocation requests into precise forget sets via deterministic queries. Evaluation on 219,555 Wikipedia pages demonstrates that record-level provenance eliminates dataset-level over-deletion (from 101x to 1.3x), while integration adds 1.3-4.0% throughput overhead (HuggingFace) and 2.1-19.0% (Datatrove) on wiki data. On a 1.7B model, provenance-based forget sets improve unlearning by 42% over random baselines. --- ### The Role of Rigor in Artificial Intelligence Authors: Timothy Nguyen Published: 2026-05-19 Categories: cs.AI arXiv: https://arxiv.org/abs/2607.03634v1 Artificial intelligence (AI) has achieved extraordinary capabilities despite lacking many of the conceptual and scientific foundations associated with mature disciplines. Unlike traditional sciences, where reliable technology typically emerges from theoretical understanding, modern AI has progressed largely through performance-driven iteration and "alchemical" experimentation. This tension motivates a systematic analysis of AI through the lens of rigor. We introduce a three-part framework consisting of conceptual rigor (clarifying foundational concepts), epistemic rigor (establishing scientific understanding), and operational rigor (ensuring reliable performance and deployment). Using this framework, we analyze competing conceptions of intelligence and understanding, the strengths and limitations of the empirical approach to deep learning, the power and pitfalls of benchmarks, and the obstacles to theory development posed by modern AI systems. We argue that the distinctive trajectory of AI arises from how forms of rigor interact across paradigms, resulting in the primacy of operational rigor in modern deep learning. This perspective helps explain both AI's rapid advances and its persistent uncertainties, while clarifying the challenges involved in transforming AI into a mature science and reliable technology. --- ### High-Dimensional Concentration and Retrieval Instability in Embedding Spaces: Implications for Retrieval-Augmented Generation Authors: Ernesto Lopez Fune Published: 2026-05-19 Categories: cs.IR, cs.AI arXiv: https://arxiv.org/abs/2606.28330v1 Embedding-based retrieval systems rely on the assumption that geometric proximity in highdimensional representation spaces reflects semantic relevance. However, high-dimensional geometry induces concentration phenomena that can reduce the discriminative power of similarity measures and can destabilize nearest-neighbor retrieval. This work studies distance concentration, cosine concentration, contrast collapse, hubness, and retrieval instability through controlled numerical experiments across multiple synthetic distributions. The results show that similarity signals progressively lose contrast as dimension increases, leading to unstable retrieval behavior and structural bias in nearest-neighbor selection. A simplified Retrieval-Augmented Generation experiment further suggests that these effects can degrade grounding reliability upstream of generation. These findings motivate geometry-aware diagnostics and robustness-oriented retrieval strategies for embedding-based AI systems. The experiments are intentionally synthetic in order to isolate intrinsic geometric effects. High-dimensional embedding space Distance and cosine concentration Score-gap collapse and hubness Retrieval instability under perturbations Weak or incomplete retrieved context Potential degradation of grounding 1. --- ### $M^3 QuestionIng$: Multi-modal Multi-span Medical Question Answering Authors: Anisha Saha, Vaibhav Rathore, Abhisek Tiwari, Akash Ghosh, Sai Ruthvik Edara, Sriparna Saha Published: 2026-05-19 Categories: cs.IR, cs.AI, cs.CL, cs.MM arXiv: https://arxiv.org/abs/2606.28329v1 The growing adoption of AI in healthcare, particularly in preventive care, highlights the critical need for accessibility and precision in Medical Question Answering (MedQA). In recent years, significant efforts have been made to develop multi-span medical question-answering systems, where the answer to a query may span multiple sections or paragraphs of a source document. However, existing systems fall short of aligning with real-world scenarios, where source documents often include both textual and visual content, requiring answers to incorporate images for better comprehension. To address this gap, we propose $M^3QAFrame$, a multi-modal, multi-span medical question-answering framework that leverages visual cues to enhance the generation of comprehensive answers drawn from diverse textual and visual spans. The model takes the context, query, and images as input and outputs an answer containing both textual answers and relevant images. The text and image embeddings are processed using a transformer-based architecture to determine the sentence and image relevance. We curate a multi-modal, multi-span medical question-answering ($M^3 QuestionIng$) dataset containing queries, medical contexts, associated medical images, and extractive answers. Additionally, each query-answer pair is labeled with user intent and query type to enhance query and context comprehension. Extensive experiments show that our approach consistently outperforms existing methods across various evaluation metrics. --- ### Type Checking Project Haystack Grids using JSON Schema and Pydantic Authors: Thomas Hirsch, Samina Kadkhoda Masoumali, Gerald Schweiger Published: 2026-05-19 Categories: cs.PL, cs.AI, cs.SE arXiv: https://arxiv.org/abs/2606.24891v1 Ontologies enable scalable energy services in buildings by supporting interoperability and automation. Project Haystack is a building ontology that is widely adopted due to its flexible, tag-based semantic model, openness, and extensibility, but suffers from ambiguous tag usage and limited automated validation. Although Project Haystack is formally open, its reliance on custom file formats and domain-specific languages that originate from the Haxall ecosystem creates a de facto barrier to integration. In this paper, we address these limitations by introducing a Python-based toolchain for Haystack. We present (i) a parser for Haystack definition files (Trio file format), and (ii) a code generator that derives Pydantic models and JSON Schema definitions from these parsed specifications. The resulting models enable static type checking and enable structural validation of Haystack grids within Python, as well as schema-based validation of JSON representations outside the Python ecosystem. All tools, generated models, and schemas are released publicly under an open-source license, with the goal of strengthening the Haystack ecosystem and opening a practical pathway beyond its current technical boundaries. --- ### Zhinong AI: A Design-Science Study of an AI-Enabled Agricultural Decision-Support Platform for Smallholder Production Authors: Zhaoyang Li, Jiaqi Liu, Ruijie Zhang Published: 2026-05-19 Categories: cs.HC, cs.AI arXiv: https://arxiv.org/abs/2606.20601v1 Artificial intelligence is increasingly moving from single-purpose agricultural recognition tools toward integrated decision-support systems that connect information access, diagnosis, task execution and post-action feedback. This paper presents a design-science case study of the Zhinong AI Agricultural Decision Platform, a farmer-facing system that integrates agricultural information push services, natural-language question answering, image-based crop disease diagnosis, plot and farming-calendar management, workflow orchestration, a Hainan Free Trade Port agricultural service zone and an age-friendly care mode. Based on public project materials, policy context and prior research on smart agriculture, machine learning and design science, the paper constructs a layered system architecture and a closed-loop decision process summarized as sensing, analysis, planning, execution and feedback. It further proposes a function-pain-point mapping matrix, an evaluation indicator system and a governance framework covering data provenance, model risk, expert review, privacy and adoption risk. The study does not claim measured field performance because production logs, controlled user studies and expert-labeled local image datasets were not available at the time of writing. Instead, the contribution is a structured research framework for transforming an AI agricultural prototype into an empirically testable, accountable and localized decision-support infrastructure for smallholder production. --- ### The New Associationism: Lessons from Deep Learning Authors: Daniel Rothschild Published: 2026-05-19 Categories: cs.AI, cs.LG arXiv: https://arxiv.org/abs/2606.20600v1 What can the success of modern AI tell us about how humans learn? This paper argues that taking AI seriously as a model of human learning supports a modest but genuine associationism. The central finding is that supervised learning -- learning driven by evaluative feedback -- underlies a surprisingly wide range of contemporary AI systems, from large language models to game-playing agents, differing primarily in how much work is required to generate the relevant feedback signal. This vindicates associationist ideals of a uniform, gradual, error-driven learning mechanism operating across domains, and defuses the once-influential argument that associationist mechanisms are too limited to account for human cognitive capacities. At the same time, the successes of deep learning depend on computational architectures that go well beyond anything classical associationists envisaged, and supervised learning operates within these as one component rather than a complete account of learning. --- ### Beyond Fixed Budgets: Characterizing the Inelasticity and Limitations of Tree-of-Thought Reasoning Strategies Authors: Atkia Mahila, Avinash Maurya, M. Mustafa Rafique, Bogdan Nicolae Published: 2026-05-19 Categories: cs.AI, cs.LG arXiv: https://arxiv.org/abs/2606.20599v1 Tree of Thought (ToT) search has become a promising direction for improving the reasoning capabilities of large language models, but deploying these methods in practice raises a question that has received little systematic attention: how do different search strategies behave under varying compute budgets, model sizes, and problem difficulties? In this work, we evaluate two representative ToT methods; DPTS, a Monte Carlo tree search based approach, and SSDP, a semantic deduplication based approach, across two mathematical reasoning benchmarks (Math500 and GSM8K), two model scales (Llama-3B and Llama-8B), and four token budgets (3k--10k). Our analysis reveals that the two methods exhibit limitations that pull in opposite directions. DPTS suffers from a cold-start bottleneck at low budgets: it requires sufficient exploration before its value estimates become reliable, making it a poor fit for resource-constrained settings despite strong scaling behavior at higher budgets. SSDP, on the other hand, reaches candidate solutions efficiently but is prone to frontier depletion; its aggressive node merging permanently discards unexplored paths, leaving it unable to improve regardless of how much budget remains. Together, these findings suggest that neither a fixed exploration strategy nor a fixed pruning strategy is sufficient across compute continuum. We argue that effective search for scientific reasoning agents requires strategies that can adapt their behavior based on search progress and available resources. --- ### Using Biometrics to Understand AI-Assisted Coding Performance and its Perception Authors: Paolo Burelli, Fabio Calefato, Daniela Grassi, Mihaela Yurieva Hristova, Nicole Novielli, Alberto Antonio, Romano, Paolo Tell Published: 2026-05-19 Categories: cs.HC, cs.AI arXiv: https://arxiv.org/abs/2606.20598v1 AI-based code assistants are transforming software development, yet we lack empirical evidence on how they affect developers' cognitive processes. We present a multisite study investigating the neurophysiological correlates of AI-assisted programming through a within-subjects crossover design. We recruited participants at two universities (Bari, Italy, and Copenhagen, Denmark) and collected electroencephalography, eye-tracking, electrodermal activity, and heart rate variability data alongside a rubric-based performance score and self-reported workload across six dimensions using the NASA Task Load Index (NASA-TLX). We tested four hypotheses addressing physiological differences between AI-assisted and non-assisted conditions, the moderating role of developer experience, the association between physiology and performance, and the alignment between subjective perceptions and objective measures. Under AI assistance, the EEG $θ/α$ ratio was lower during the first task and the gaze blink rate was higher during the second, both consistent with reduced cognitive engagement when developers offload generative effort to the model. This pattern did not differ between undergraduate and graduate students. Electrodermal activity correlated with performance under the non-AI condition but not under AI. Among the six NASA-TLX dimensions of self-reported workload, only Physical demand was associated with performance under the non-AI condition but not under AI. These findings suggest that AI-assisted programming is not a faster version of solo coding but a cognitively distinct activity, with implications for the design of AI assistants and for biometric monitoring in AI-augmented development. ---