# Research Papers — 2026-05 · bundle page 46 of 326 > Full-text markdown of papers #2251–2300 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/46 [← Previous bundle](https://aipod.dev/papers/archive/2026-05/page/45.md) [Next bundle →](https://aipod.dev/papers/archive/2026-05/page/47.md) ### The Confidence Shortcut: A Reasoning Failure Mode of Masked Diffusion Models Authors: Dueun Kim, Albert No Published: 2026-05-27 Categories: cs.AI, cs.CL arXiv: https://arxiv.org/abs/2605.29123v1 Masked diffusion language models (MDMs) uniquely support any-order generation, with confidence-based decoding currently serving as the de facto standard inference policy. To optimize for this, recent training schemes attempt to align training mask patterns directly with those observed during generation. However, we argue that confidence-based decoding is inherently misaligned with the logical-flow trajectories required for complex reasoning, and that confidence-aligned training actively entrenches this misalignment. We make this concrete using multi-digit addition, where the decoding strategy prematurely predicts locally easy digits before resolving their long-range dependencies, producing high-confidence errors on challenging inputs. While traditional random masking keeps the failure rate low on this challenging tail, confidence-aligned training amplifies the error rate by an order of magnitude. Across five distinct reasoning tasks, this same pattern emerges with task-dependent severity: confidence-based decoding induces failures on highly complex inputs, and confidence-aligned training exacerbates them. In contrast, random masking -- despite its perceived inefficiency -- robustly preserves the reasoning-trajectory conditionals essential for solving the challenging tail. --- ### Robust Cross-Domain Generalization Using Unlabeled Target Data with Source-Domain Supervision Authors: Yuyue Zhou, Shrimanti Ghosh, Michael, Xie, Justin JY Kim, Jessica Knight, Steel McDonald, Vincent Man, Jacob L. Jaremko, Abhilash Hareendranathan Published: 2026-05-27 Categories: cs.CV arXiv: https://arxiv.org/abs/2605.29122v1 It is often desirable to generalize medical imaging AI models trained with dense annotations to data acquired from different ultrasound scanners or clinical sites; however, retraining these models with new annotations is often difficult and costly. We examine this challenge in pediatric wrist fracture assessment using point-of-care ultrasound (POCUS), where fractures are common and can be effectively triaged via ultrasound. AI has shown radiologist-level performance for fracture detection, often aided by high-quality bony structure segmentation. However, due to significant domain shifts, models perform poorly on data from other centers or probes, and obtaining segmentation labels across devices is impractical due to manual annotation effort and data privacy concerns. To address this, we propose a target-informed self-supervised pretraining and model-ensemble strategy. Specifically, our approach combines masked image modeling (MIM) and contrastive learning to learn target-domain structural representations without labels, and introduces a confidence-aware infusion head to adaptively integrate predictions. The source dataset, collected with a Philips Lumify probe, contained dense labels, while the target dataset, acquired with a TeleMED portable probe, was unlabeled. The datasets were kept strictly separate throughout the entire process. Our method used labeled source data for supervised training and leveraged target-domain pretraining to improve generalization. On 318 images from 62 pediatric POCUS videos, this approach significantly improved cross-device performance, achieving over 6% Dice improvement on the target domain versus the baseline. These results demonstrate a label-efficient and privacy-preserving approach for cross-device-robust ultrasound AI, offering a framework that can be extended to multi-center studies or federated learning setups. --- ### A Minimal Bifurcation Model of Load Imbalance in a Softmax Mixture-of-Experts Router Authors: O. M. Kiselev Published: 2026-05-27 Categories: math.DS, cs.AI, cs.LG arXiv: https://arxiv.org/abs/2605.29121v1 We propose a minimal dynamical model of adaptive softmax routing for a two-expert Mixture-of-Experts (MoE) layer. The model is obtained as a mean-field limit of a discrete reinforcement rule: the selected expert receives a small score increment, while all scores undergo regularizing decay. In the symmetric case the limiting system has a supercritical pitchfork bifurcation: for weak feedback there is a unique stable balanced state, whereas above a critical feedback strength two stable asymmetric states appear. When an external asymmetry is added, the pitchfork unfolds into a pair of fold bifurcations forming a cusp in the control-parameter plane. We derive exact parametric equations for the bifurcation set and the local normal form of the cusp catastrophe. Numerical experiments connect this picture to empirical expert load, a small trainable MoE model, hard top-1 PyTorch routing, and a small classification experiment on digits. The results provide a controlled low-dimensional mechanism for abrupt transitions to load imbalance in adaptive MoE routers. --- ### PRO-CUA: Process-Reward Optimization for Computer Use Agents Authors: Yifei He, Rui Yang, Hao Bai, Tong Zhang, Han Zhao Published: 2026-05-27 Categories: cs.AI arXiv: https://arxiv.org/abs/2605.29119v1 Computer use agents (CUAs) have shown strong potential for automating complex digital workflows, yet their training remains constrained by costly live environment interaction and limited high-quality supervision. Existing filtered behavior cloning pipelines suffer from imitation bottlenecks, including distribution shift from the expert demonstration and the absence of negative learning signals. Meanwhile, standard trajectory-level reinforcement learning struggles with sparse rewards, ambiguous credit assignment, and high infrastructure costs for long-horizon GUI interaction. In this work, we propose PRO-CUA, a process-reward optimization framework for training CUAs with iterative step-level reinforcement learning. PRO-CUA decouples on-policy environment interaction from policy optimization: the current policy collects states through live rollouts, generates diverse candidate actions for each state, receives step-level feedback from a process reward model (PRM), and is optimized with group-relative advantages. This design enables dense and flexible credit assignment without relying on golden answers or offline expert trajectories, while reducing distribution shift by training on the agent's own execution states. Experiments on live web benchmarks demonstrate the effectiveness of PRO-CUA and the reliability of PRM-guided step-level training. --- ### Beyond Consensus: Trace-Level Synthesis in Mixture of Agents Authors: Shreyas Fadnavis, Praitayini Kanakaraj, Felix Wyss Published: 2026-05-27 Categories: cs.AI arXiv: https://arxiv.org/abs/2605.29116v1 When multiple LLM agents solve the same problem, standard practice compresses each agent's reasoning into a majority vote or layered synthesis, treating agreement as the finish line. We show this is unnecessarily lossy: an LLM aggregator that reads complete reasoning traces recovers correct solutions even when agents unanimously agree, with beneficial corrections consistently outweighing harmful ones -- the \emph{aggregation paradox}. Majority voting has a ceiling that perturbation diversity does not raise (error correlations are identical); the aggregator's gain comes from trace-level complementarity, assembling correct intermediate steps from minority chains that voting discards. These findings motivate Self-Consistent Mixture of Agents which generates trace diversity through semantic-preserving input perturbations, safeguards the majority via anchored refinement with provable non-degradation guarantees, and always synthesizes -- never gates on consensus. A single model with perturbation-induced trace variation outperforms heterogeneous model pools across structured reasoning, PhD-level science, competition mathematics, and competitive programming. The unit of aggregation should be the reasoning trace, not the answer. --- ### unix-ctf: Procedural Environments for Unix-Competence Reinforcement Learning Authors: Geoffrey Bradway, Roger Creus Castanyer, Lorenz Wolf, Maxwill Lin, Matthew James Sargent, Augustine N. Mavor-Parker Published: 2026-05-27 Categories: cs.CR, cs.AI arXiv: https://arxiv.org/abs/2605.29115v1 Unix competence is the ability to use shell and operating-system primitives as first-class tools, not merely to write programs through a terminal. Current terminal benchmarks tend to blur this distinction: a solver fluent in Python but weak in Unix can pass a substantial fraction of Terminal-Bench 2.0, while the reverse skill profile is rarely exercised. We make the distinction operational and build a training surface for the Unix component. unix-ctf is a procedural generator of capture-the-flag tasks for shell agents. Each task hides a short token (a flag of the form flag(a3b1c9...)) inside a fresh Linux container using a single Unix feature, and the agent must recover it. Tasks are produced by an LLM-assisted synthesis pipeline that generates candidate hiding techniques, rewrites them into parameterized hide-and-find script pairs, and filters them with a bidirectional contract: the hide script must leave no plaintext trace of the flag on disk, and the find script must recover the flag in a fresh directory. Because the LLM only writes the planting and recovery steps (the container, layout, and grading harness are fixed), the pipeline lands 656 of 750 raw attempts as portable, reusable variants (87.5\%). Our reproduction of Endless Terminals' full-container-generation approach lands only 17.4\% under the same checks. The 656 variants canonicalize to 155 distinct techniques. Fine-tuning Qwen3-8B with LoRA using GRPO on this surface lifts solve rate from 11.6\% to 43.6\% on a 15-skill multi-family holdout (n=225), redistributes which InterCode-CTF tasks the model solves, and produces a +33 pp gain in Forensics while reaching 32/100 on InterCode-CTF. These results suggest that Unix competence is separable, trainable, and best evaluated directly rather than folded into programming-through-a-shell. --- ### ReasonBreak: Probing Vulnerabilities in Reasoning-Enabled Vision-Language-Action Models for Autonomous Driving Authors: Mohammadreza Teymoorianfard, Jean-Philippe Monteuuis, Jonathan Petit, Amir Houmansadr Published: 2026-05-27 Categories: cs.CR, cs.LG, cs.RO arXiv: https://arxiv.org/abs/2605.29114v1 Vision-Language-Action (VLA) models with integrated reasoning have been proposed for end-to-end autonomous driving, assuming a tight coupling between reasoning and trajectory generation. However, the robustness of such systems under realistic input perturbations remains largely unexplored. We show that these models are highly vulnerable to realistic input perturbations, achieving up to 89% attack success rate (ASR) on reasoning and up to 72% on trajectory manipulation in closed-loop simulation, leading to increased collision rates and degraded safety metrics. Using NVIDIA's recent Alpamayo models as representative industry-developed VLAs, we conduct the first systematic black-box study of reasoning-enabled VLA models under realistic textual input corruptions, evaluating their impact on reasoning and driving behavior. We introduce a reasoning-aware evaluation framework capturing both semantic and structural aspects of reasoning, along with safety-centric measures. We also introduce a benchmark for evaluating attacks and defenses on reasoning-trajectory interactions in autonomous driving. Our results highlight the need for rigorous evaluation and improved defenses to ensure the safety of reasoning-enabled VLA systems in autonomous driving. --- ### Bridging Chemists and AI: An Expert-Augmented Framework for Interpretable Route Evaluation Authors: Yujia Guo, Mikhail Kabeshov, Tat Hong Duong Le, Samuel Genheden, Marco V. Mijangos, Varvara Voinarvoska, Giulia Bergonzini, Ola Engkvist, Samuel Kaski Published: 2026-05-27 Categories: cs.LG arXiv: https://arxiv.org/abs/2605.29108v1 Selecting efficient multi-step synthetic routes is a central challenge in organic synthesis, particularly in medicinal and process chemistry, where route choice directly impacts feasibility, cost, and development efficiency. Data-driven assessment systems often oversimplify the multi-objective nature of synthesis design and rely on proxy datasets, such as patent routes, rather than universally grounded criteria. To address this, we introduce an expert-augmented, data-driven scoring framework that integrates machine learning with chemists' domain knowledge for both numerical and explainable route assessment. A DeepSets-based model is trained using tree edit distance between reference and machine-generated routes, and then fine-tuned with expert evaluations to produce both quantitative scores and interpretable qualitative categories: Good, Plausible, and Bad. The resulting system achieves a Spearman correlation coefficient of 0.78 and a Pearson correlation of 0.77 for category assessment prediction, and 60.2% top-1 ranking accuracy for score prediction, substantially outperforming the previous baseline of 17.5%. --- ### GEO-Bench: Benchmarking Ranking Manipulation in Generative Engine Optimization Authors: Ojas Nimase, Zhe Chen, Gengpei Qi, Yue Zhao, Xiyang Hu Published: 2026-05-27 Categories: cs.CR, cs.AI arXiv: https://arxiv.org/abs/2605.29107v2 Large language models (LLMs) increasingly rank products, documents, and recommendations for user queries, which makes manipulating these rankings a growing concern for fairness and information integrity. Research on generative engine optimization (GEO) has produced many manipulation methods, but each is evaluated on its own dataset with its own metrics, so their relative strength and detectability stay unclear. We present GEO-Bench, a benchmark that evaluates GEO ranking-manipulation attacks under one protocol. It unifies black-box prompt-based attacks (TAP, Zero-Shot), white-box gradient-based attacks (STS, RAF, StealthRank), and ten white-hat C-SEO strategies. We score every method on five datasets against a fixed open-weight ranker (Llama-3.1-8B-Instruct), using metrics for both effectiveness (NRG, Success@α, Promote@α) and stealth (keyword violation rate, perplexity ratio). Our evaluation shows that effectiveness and stealth trade off across adversarial attacks, that black-box content rewriting matches or exceeds gradient-based attacks on rank promotion while producing more fluent text and can evade both keyword- and perplexity-based detection on some domains, and that the access model does not predict attack strength. By standardizing datasets, attack implementations, and metrics, GEO-Bench enables the first direct comparison across these attack paradigms and supports the development of detection methods. --- ### Model Merging by Output-Space Projection Authors: Bethan Evans, Benjamin Etheridge, Stephen Roberts, Jared Tanner Published: 2026-05-27 Categories: cs.LG, cs.IT arXiv: https://arxiv.org/abs/2605.29101v1 Model merging combines fine-tuned checkpoints into a single multi-task model without retraining. Existing methods - such as task arithmetic, model soups, TIES, and DARE - are computationally efficient and empirically successful, but rely on heuristic design choices and lack formal optimality guarantees. We show that merging can be formulated as a convex quadratic programme over residual updates, yielding weights that minimise a squared-output calibration objective using calibration inputs and fine-tuned model outputs, and subsuming existing methods as special cases. Our framework yields a closed-form diagnostic - the fraction of residual energy captured by a chosen basis - that predicts downstream merge quality using only the calibration set. Empirically, the QP matches or outperforms existing methods in the single-layer setting, and we characterise when the optimal basis provides significant gains over the cheaper diagonal QP. We extend to multi-layer merging via a sequential layer-wise algorithm and demonstrate consistent gains across language and vision benchmarks. --- ### Seeing through boxes: Non-Line-of-Sight 3D Reconstruction from Radar Signals Authors: Jiachen Lu, Hailan Shanbhag, Haitham Al Hassanieh Published: 2026-05-27 Categories: cs.CV arXiv: https://arxiv.org/abs/2605.29098v2 Reconstructing object geometry from radio frequency (RF) signals is fundamentally challenging due to the lensless imaging nature of RF sensing, which leads to low spatial resolution and high noise. Unlike light signals, RF signals can penetrate occlusions and thus capture information about hidden scenes. Existing Non-Line-of-Sight (NLoS) 3D neural reconstruction methods can recover coarse surfaces inside enclosed environments but often suffer from unstable optimization, noisy surface geometry, and surface ambiguity, failing to produce accurate zero-level sets from the signed distance field (SDF). These limitations largely stem from neglecting the role of Line-of-Sight (LoS) geometry outside the enclosed region, which provides valuable physical constraints for modeling signal propagation. In this paper, we introduce a Unified LoS and NLoS neural geometry reconstruction framework GeRaF 2.0 that leverages the outside LoS geometry to model and guide RF propagation from the LoS region into the NLoS region. By integrating visual LoS priors into the neural field formulation, GeRaF 2.0 achieves stable training and physically consistent reconstruction of both visible and hidden geometry, setting a new state-of-the-art in RF-based geometry reconstruction. --- ### GeRaF: Neural Geometry Reconstruction from Radio Frequency Signals Authors: Jiachen Lu, Hailan Shanbhag, Haitham Al Hassanieh Published: 2026-05-27 Categories: cs.CV arXiv: https://arxiv.org/abs/2605.29097v2 GeRaF is the first method to use neural implicit learning for near-range 3D geometry reconstruction from radio frequency (RF) signals. Unlike RGB or LiDAR-based methods, RF sensing can see through occlusion but suffers from low resolution and noise due to its lensless imaging nature. While lenses in RGB imaging constrain sampling to 1D rays, RF signals propagate through the entire space, introducing significant noise and leading to cubic complexity in volumetric rendering. Moreover, RF signals interact with surfaces via specular reflections, requiring fundamentally different modeling. To address these challenges, GeRaF (1) introduces filter-based rendering to suppress irrelevant signals, (2) implements a physics-based RF volumetric rendering pipeline, and (3) proposes a novel lensless sampling and lensless alpha blending strategy that makes full-space sampling feasible during training. By learning signed distance functions, reflectiveness, and signal power through MLPs and trainable parameters, GeRaF takes the first step towards reconstructing millimeter-level geometry from RF signals in real-world settings. --- ### Trends in AI and Human-AI Interaction in Clinical Trials -- A Hybrid Human-AI Exploration Authors: Sandra Woolley, Tim Collins, Khalid Khattak, Illia Chernomorets, Ariane Arevalo, Chris Richardson Published: 2026-05-27 Categories: cs.AI arXiv: https://arxiv.org/abs/2605.29096v1 This paper examines records retrieved from the ClinicalTrials.gov registry to characterize temporal trends in AI terminology and the geographical distribution of AI trials. The work also reports on an exploratory hybrid human-AI approach to analyzing human-AI interaction trends in registered clinical trials. The hybrid workflow comprised a frontier generative AI model (GPT-5.5) and human review to screen and categorize records returned by an AI-focused search. The findings indicate a marked increase in AI-related trials over time, with recent growth in references to machine learning, deep learning, chatbots, GPTs, and large language models. Geographically, China and the United States accounted for the largest numbers of AI-related trials, with notable recent increases in several other countries including Italy, France, Spain, the UK and Turkey (Türkiye). In a random sample of 100 records, human and AI classifiers showed good agreement in identifying studies not substantively using AI, but lower agreement in classifying human-AI interaction, particularly where health professional interaction was ambiguous or insufficiently described. Overall, the results suggest that hybrid human-AI screening of clinical trial records is potentially viable, but clearer trial reporting and more precise interaction definitions will benefit the process. --- ### Lightweight Complementary-Cue Fusion for Robust Video Face Forgery Detection Authors: Sunghwan Baek, Tariq Anwaar, Karanveer Singh, Rita Singh Published: 2026-05-27 Categories: cs.CV, cs.LG, cs.MM arXiv: https://arxiv.org/abs/2605.29092v1 Current face video forgery detectors use wide or dual-stream backbones. We show that a single, lightweight fusion of two handcrafted cues can achieve higher accuracy with a much smaller model. Based on the Xception baseline model (21.9 million parameters), we build two detectors: LFWS, which adds a 1x1 convolution to combine a low-frequency Wavelet-Denoised Feature (WDF) with a phase-spectrum channel derived from Spatial-Phase Shallow Learning (SPSL), and LFWL, which merges WDF with Local Binary Patterns (LBP) in the same way. This extra module adds only 292 parameters, keeping the total at 21.9 million, smaller than F3Net (22.5 million) and less than half the size of SRM (55.3 million). Even with this minimal overhead, the fused models increase the average area under the curve (AUC) from 74.8% to 78.6% on FaceForensics++ and from 70.5% to 74.9% on DFDC-Preview, gains of 3.8% and 4.4% over the Xception baseline. They also consistently outperform F3Net, SRM, and SPSL in eight public benchmarks, without extra data or test-time augmentation. These results show that carefully paired, handcrafted features, combined through the lightweight fusion block, can provide competitive robustness at a significantly lower cost than comparable frequency-based detectors. Our findings suggest a need to reevaluate scale-driven design choices in face video forgery detection. --- ### OISD: On-Policy Internal Self-Distillation of Language Models Authors: Xinyu Liu, Darryl Cherian Jacob, Yang Zhou, Jindong Wang, Pan He Published: 2026-05-27 Categories: cs.LG, cs.AI, cs.CV arXiv: https://arxiv.org/abs/2605.29089v2 Recent reinforcement learning (RL) post-training approaches primarily optimize the final output policy using sparse outcome-level rewards, while largely overlooking predictive signals encoded in intermediate representations. In this paper, we introduce a new paradigm called on-policy internal self-distillation and propose the OISD framework, which improves reasoning by transferring on-policy predictive signals from the final layer to intermediate representations. During rollout and Group Relative Policy Optimization (GRPO) optimization, the final layer acts as both the policy and a detached internal teacher for selected intermediate layers, which are guided to align with it through two complementary mechanisms: logit alignment, which transfers high-level reasoning behaviors (how to think), and attention alignment, which enforces consistent attention patterns (where to look) from the final layer to the selected intermediate layer, both without requiring external privileged information. Our OISD, together with GRPO, employs signed advantage-weighted Jensen--Shannon alignment to distill informative intermediate representations while preserving policy consistency under a unified acting policy. Experimental results demonstrate the effectiveness of OISD, with substantial and consistent improvements over strong reasoning RL baselines across four mathematical reasoning tasks. The code will be released at https://github.com/THE-MALT-LAB/OISD --- ### OISD: On-Policy Internal Self-Distillation of Language Models Authors: Xinyu Liu, Darryl Cherian Jacob, Yang Zhou, Jindong Wang, Pan He Published: 2026-05-27 Categories: cs.LG, cs.AI, cs.CV arXiv: https://arxiv.org/abs/2605.29089v1 Recent reinforcement learning (RL) post-training approaches primarily optimize the final output policy using sparse outcome-level rewards, while largely overlooking predictive signals encoded in intermediate representations. In this paper, we introduce a new paradigm called on-policy internal self-distillation and propose the OISD framework, which improves reasoning by transferring on-policy predictive signals from the final layer to intermediate representations. During rollout and Group Relative Policy Optimization (GRPO) optimization, the final layer acts as both the policy and a detached internal teacher for selected intermediate layers, which are guided to align with it through two complementary mechanisms: logit alignment, which transfers high-level reasoning behaviors (how to think), and attention alignment, which enforces consistent attention patterns (where to look) from the final layer to the selected intermediate layer, both without requiring external privileged information. Our OISD, together with GRPO, employs signed advantage-weighted Jensen--Shannon alignment to distill informative intermediate representations while preserving policy consistency under a unified acting policy. Experimental results demonstrate the effectiveness of OISD, with substantial and consistent improvements over strong reasoning RL baselines across four mathematical reasoning tasks. The code will be released at https://github.com/THE-MALT-LAB/OISD --- ### A Deep Learning Iterative Framework for Sentinel-1 Stripmap Enhancement Based on Azimuth Doppler Decomposition Authors: Juan Francisco Amieva, Christian Ayala, Roberto Del Prete, Mikel Galar Published: 2026-05-27 Categories: cs.CV arXiv: https://arxiv.org/abs/2605.29088v1 Synthetic Aperture Radar (SAR) imagery enables all-weather, day-and-night Earth observation; however, it remains difficult to interpret due to speckle noise and other intrinsic imaging artifacts. Sentinel-1 (S1) constitutes one of the most widely used spaceborne SAR missions, offering systematic global coverage, high temporal resolution, dual-polarization imaging, and free data availability. Among S1 modes, Stripmap (SM) provides the highest resolution, yet speckle noise and spatial constraints often hinder applications requiring finer spatial detail. This motivates the need for effective image enhancement strategies. In this work, we propose a self-supervised enhancement framework for S1 SM imagery based on azimuth subaperture decomposition. The method exploits the physical consistency between subaperture reconstructions and the corresponding full-aperture image to generate paired training data without external sensors, simulated ground truth, or multi-temporal stacks. The proposed framework integrates single- and multi-frame learning and incorporates an iterative inference scheme that progressively refines image quality. Experiments on real S1 SM data show that the proposed approach consistently outperforms the widely adopted self-supervised deep learning baseline MERLIN, in terms of PSNR and SSIM, while MERLIN attains higher ENL, highlighting a trade-off between structural fidelity and speckle smoothing. Overall, the results demonstrate that subaperture-based supervision provides a physically grounded, reproducible, and operationally viable approach for SAR image enhancement using S1 data. It is worth noting that the proposed approach can be extended to other SAR platforms, polarizations, and acquisition modes. --- ### The Chain Holds, the Answer Folds: Trace-Answer Dissociation in Reasoning Models Under Adversarial Pressure Authors: Yubo Li, Ramayya Krishnan, Rema Padman Published: 2026-05-27 Categories: cs.AI arXiv: https://arxiv.org/abs/2605.29087v1 Reasoning models are evaluated on single-turn benchmarks but deployed in multi-turn dialogue, where users push back on correct answers. Under sustained adversarial pressure we find a previously undocumented failure mode: the chain-of-thought stays factually correct from first turn to last while the emitted answer flips wrong. We call this unfaithful capitulation (UC) and isolate it with a $2\times 2$ latent-versus-behavioral framework that flip-rate metrics and single-turn faithfulness probes both miss. Across three datasets (MT-Consistency, MMLU-Pro, GSM8K), the latent-correct rate at the behavioral flip clusters near 50% in think mode and collapses to 11-15% under no_think -- paired, within-model causal evidence that reasoning creates the gap. Across models the effect tracks the reasoning channel (high in Qwen3-32B and GPT-OSS-20B, low in inline-CoT Gemma-4-31B-it). An independent GPT-4o judge corroborates $86\%$ of UC labels; a token-level probe shows the answer-slot argmax is correct in $84\%$ of UC cells; and a naive trace-anchored defense backfires. We release all trajectories, traces, and judge labels. --- ### Same Question, Different Source, Different Answer: Auditing Source-Dependence in Medical Multi-Source RAG Authors: Yubo Li, Rema Padman, Ramayya Krishnan Published: 2026-05-27 Categories: cs.CL, cs.AI, cs.IR arXiv: https://arxiv.org/abs/2605.29084v1 A retrieval-augmented generation (RAG) system deployed over a multi-author institutional corpus can give a different answer to the same question depending on which source it retrieves -- a failure mode the dominant single-gold-answer paradigm cannot diagnose. We argue that source-dependence is a missing axis of NLP evaluation, and that auditing it means shifting the unit of evaluation from answer correctness to the inter-source relationship. We make this concrete in transplant patient education, where institutional sources demonstrably disagree, releasing three artefacts: TransplantQA, a benchmark of real patient questions, each answered by grounding generation in multiple institutional handbooks as candidate sources; HERO-QA, a hierarchical retrieval strategy that grounds and audits each answer; and a structured-output judge that scores inter-source relationships on a validated 5-label taxonomy. At scale, better retrieval reveals far more disagreement than prior estimates suggested -- understating its prevalence, not its intensity. The framework is domain-agnostic and transfers to legal and educational RAG: measuring source-dependence is a responsibility for deployed multi-source NLP generally. --- ### The Importance of Out-of-Band Metadata for Safe Autonomous Agents: The Redpanda Agentic Data Plane Authors: Tyler Akidau, Tyler Rockwood, Johannes Brüderl, Marc Millstone Published: 2026-05-27 Categories: cs.AI arXiv: https://arxiv.org/abs/2605.29082v1 AI agents are increasingly expected to operate as digital employees: accessing enterprise data, making decisions, and taking actions autonomously. But agents are simultaneously less predictable than humans -- prone to hallucination, misinterpretation, and adversarial manipulation -- and more technically capable: with deep system knowledge and high-throughput interfaces cascading damage at machine speed. This combination makes it unsafe to rely on agents to faithfully interpret or propagate security-critical metadata such as access policies, data classifications, and behavioral constraints. We present the Redpanda Agentic Data Plane (ADP), an architecture built around out-of-band metadata channels: infrastructure pathways that carry security context, policy signals, and audit trails deterministically, entirely outside the agent's read and write path and across heterogeneous infrastructure. These channels enforce governance at every stage of the agent lifecycle -- scoping data access on the way in, constraining actions during execution, and capturing tamper-proof transcripts on the way out. We demonstrate ADP with a multi-agent portfolio rebalancing system in which autonomous agents monitor markets, make trade decisions, and execute orders across isolated client accounts -- with per-client data scoping, trade approval thresholds, and tamper-proof audit trails all enforced by out-of-band channels the agents can neither see nor bypass. --- ### Bridging the Sim-to-Real Gap in Reinforcement Learning-Based Industrial Dispatching through Execution Semantics Authors: Jonathan Hoss, Noah Klarmann Published: 2026-05-27 Categories: cs.AI, cs.LG arXiv: https://arxiv.org/abs/2605.29078v1 Event-driven scheduling policies are increasingly deployed in industrial environments, where decisions are made under asynchronous and partially observed system states. As a result, decision states are not temporally consistent, action admissibility is not explicitly defined, and the origin of execution errors remains ambiguous. These issues limit both reliability and interpretability. To address this gap, a policy-neutral execution and measurement layer is proposed to mediate between scheduling policies and the industrial execution environment. The layer constructs decision-valid snapshots from asynchronous event streams, defines a standardized execution contract with explicit action admissibility, and records outcomes as divergences between policy intent, transactional outcomes, physical execution, and human intervention. This enables a separation between decision semantics and execution behavior and makes deployment mismatch observable and structurally attributable. The proposed framework is evaluated using a discrete-event simulation. The results show analytical benefits across all observation lag regimes, as undifferentiated execution failures are transformed into structured, typed outcomes with full attribution coverage. Operational benefits are strongest under low observation lag, where avoidable execution errors can be prevented before commitment. Overall, the layer turns execution uncertainty into supervisory data for evaluation and policy refinement. --- ### Structured Prompt Optimization Meets Reinforcement Learning for Global and Local Interpretability over Complex Text Authors: Tianyang Zhou, Wenbo Chen, Pierre Jinghong Liang, Leman Akoglu Published: 2026-05-27 Categories: cs.CL, cs.AI, cs.LG arXiv: https://arxiv.org/abs/2605.29076v2 LLMs have advanced text classification, yet existing paradigms face a trade-off: supervised (label only) fine-tuning is scalable but offers limited reasoning on complex text and lacks broader model transparency, while discrete prompt optimization offers human-readable instructions but struggles with performance and scalability. We introduce eXTC (eXplainable Text Classifier) with three progressive stages: (1) learning a Standard Operating Procedure (SOP, or rulebook) in natural language via a new Structured Prompt Optimization algorithm; (2) SOP-grounded reasoning distillation from a large teacher LLM into a compact LM; and (3) expanding reasoning capabilities beyond the initial SOP via reinforcement learning. This design enables eXTC to provide (i) fast inference via a compact LM, with (ii) inference-time local reasoning traces, alongside a global, modular explanation of its learned domain rules, while (iii) significantly outperforming existing paradigms across diverse benchmarks in both classification performance and explanation quality, with stage-by-stage gains. --- ### Knowledge Offloading: Decomposing LLMs into Sparse Backbones and Memory Modules Authors: Karim Galliamov, Rochelle Choenni, Ivan Titov Published: 2026-05-27 Categories: cs.LG arXiv: https://arxiv.org/abs/2605.29075v1 LLMs encode both general capabilities and domain-specific knowledge in a single set of parameters. We ask whether this capacity can be reorganized: keeping broadly useful computation in a shared backbone, while moving specialized knowledge into external memory modules. We propose \emph{knowledge offloading} (KOFF), a framework for decomposing a pretrained LLM into a sparse shared backbone and domain-specific memories. Starting from a frozen base model, we jointly learn a structured pruning mask and lightweight recovery modules, implemented as LoRA adapters and learned key-value caches. Across Llama and Qwen models from 3B to 8B, we find that non-trivial capacity can be moved out of the shared backbone without a large loss in model ability. At around 12\% global sparsity, KOFF preserves much of the unpruned model's performance, while pruning the same frozen model without memories degrades sharply. Ablations show that LoRA and learned KV memories are complementary, and specialization analyses suggest that the learned decomposition is meaningful: language-specific neurons are preferentially removed while language-general neurons largely remain in the backbone. These results suggest that knowledge can be reallocated between a shared core and swappable external memories. --- ### Embodied3DBench: Benchmarking Low-Level Embodied Spatial Intelligence of Vision Language Models Authors: Jiyao Zhang, Mingxu Zhang, Yitong Peng, Haoxuan Liu, Chenshuo Wang, Yuxing Long, Haoyang Huang, Dongjiang Li, Nan Duan, Hui Shen, Hao Dong Published: 2026-05-27 Categories: cs.CV, cs.RO arXiv: https://arxiv.org/abs/2605.29074v1 Are current Vision Language Models (VLMs) ready to comprehend and reason about complex embodied interactions in 3D environments? We introduce Embodied3DBench, a robot-centric benchmark targeting low-level spatial intelligence in embodied 3D environments. To systematically evaluate these foundational perceptual capabilities, the benchmark includes 6 task categories divided into two core groups: Spatial Structural Understanding (Grounding, Spatial Relation Prediction, and Multi-view Correspondence) and Interaction-Oriented Perception (Affordance Prediction, Grasp Point Prediction, and Trajectory Prediction). The benchmark spans 12 subcategories and contains over 21k high-quality question-answer pairs. We evaluate 13 state-of-the-art models, and the results show that while current models exhibit relatively strong high-level spatial reasoning, such as understanding object-to-object positional relations, they remain fragile in interaction-oriented perception, highlighting a significant lack of robust 3D-aware interaction priors. To actively bridge this capability gap revealed by our benchmark, we further synthesize a large-scale training dataset comprising 1.3M QA pairs. Notably, fine-tuning on this dataset yields significant improvements in low-level spatial intelligence. Ultimately, Embodied3DBench fills a critical gap by providing both a systematic evaluation framework and a scalable data solution, setting a clear target for the development of interaction-aware multimodal systems. --- ### Diffusion Model-Based Data Assimilation for Real-World Energy Consumption Forecasting Authors: Ruoyu Hu, Dahai Yu, Feng Bao, Guang Wang, Guannan Zhang Published: 2026-05-27 Categories: cs.LG, math.NA arXiv: https://arxiv.org/abs/2605.29072v3 Accurate estimation and forecasting of energy consumption are important for power-system operation, planning, and demand-side management. In practice, however, complete and timely measurements may not always be available, and the observed data can be partial, noisy, or delayed. This motivates the use of learned forecasting models for predicting the evolving consumption state, together with data assimilation methods for sequential forecast correction. In this work, we study a high-dimensional data assimilation problem for real energy-consumption data. \modeltext{The forward prediction is supplied by a pretrained black-box spatio-temporal forecasting model, which is treated as the state propagator in the filtering procedure.} We employ the Ensemble Score Filter (EnSF) to assimilate partial and noisy observations and to correct the forecast trajectory over time. The EnSF uses score-based diffusion models to approximate filtering distributions and avoids retraining neural-network score models during assimilation by using a closed-form score representation and Monte Carlo approximation. Numerical experiments demonstrate that open-loop propagation of the learned forecasting model can become unreliable over long horizons, while EnSF-based correction substantially improves state estimation. Comparisons with the Ensemble Kalman Filter (EnKF) further show that EnSF provides stronger correction under the nonlinear observation setting considered in this work. --- ### Robust and Efficient Guardrails with Latent Reasoning Authors: Siddharth Sai, Xiaofei Wen, Muhao Chen Published: 2026-05-27 Categories: cs.AI, cs.CL, cs.CR, cs.LG arXiv: https://arxiv.org/abs/2605.29068v2 Maintaining the safety of large language models (LLMs) is crucial as they are increasingly deployed in real-world applications. Existing safety guardrails typically rely on single-pass classification or, more recently, distilled reasoning. Reasoning-based guardrails significantly outperform classification-only baselines, but they incur substantial query latency and token overhead that make them impractical for highthroughput deployment. To address this challenge, we propose COLAGUARD, a guardrail model that transfers multi-step safety reasoning into a continuous latent space through a stage-wise training curriculum, enabling direct hidden-state propagation at inference. Evaluated on ten prompt- and response-moderation settings spanning eight safety benchmarks, COLAGUARD improves macro-F1 by 8.24 points over Llama Guard 3 and matches our explicit reasoning baseline, GuardReasoner, in macroF1 while delivering a 12.9X speedup and 22.4X reduction in token usage. Our results suggest that latent reasoning offers a practical alternative to explicit rationale generation for deployable guardrails, jointly improving safety robustness and inference efficiency rather than treating them as competing objectives. --- ### Robust and Efficient Guardrails with Latent Reasoning Authors: Siddharth Sai, Xiaofei Wen, Muhao Chen Published: 2026-05-27 Categories: cs.AI, cs.CL, cs.CR, cs.LG arXiv: https://arxiv.org/abs/2605.29068v1 Maintaining the safety of large language models (LLMs) is crucial as they are increasingly deployed in real-world applications. Existing safety guardrails typically rely on single-pass classification or, more recently, distilled reasoning. Reasoning-based guardrails significantly outperform classification-only baselines, but they incur substantial query latency and token overhead that make them impractical for highthroughput deployment. To address this challenge, we propose COLAGUARD, a guardrail model that transfers multi-step safety reasoning into a continuous latent space through a stage-wise training curriculum, enabling direct hidden-state propagation at inference. Evaluated on ten prompt- and response-moderation settings spanning eight safety benchmarks, COLAGUARD improves macro-F1 by 8.24 points over Llama Guard 3 and matches our explicit reasoning baseline, GuardReasoner, in macroF1 while delivering a 12.9X speedup and 22.4X reduction in token usage. Our results suggest that latent reasoning offers a practical alternative to explicit rationale generation for deployable guardrails, jointly improving safety robustness and inference efficiency rather than treating them as competing objectives. --- ### Persona Prompting in Multimodal Urban Perception: Descriptive Convergence and Interpretive Variation Authors: Neemias da Silva, Matt Ratto, Myriam Delgado, Rodrigo Minetto, Daniel Silver, Thiago H Silva Published: 2026-05-27 Categories: cs.CL, cs.CV, cs.HC, cs.MA arXiv: https://arxiv.org/abs/2605.29064v3 This study examines how persona prompting shapes language generated by two multimodal large language models in urban perception, a setting for examining subjective interpretations of shared visual evidence. We organize outputs into three functional layers: descriptive grounding (captions), intermediate semantic layer (perception tags), and interpretive framing (justifications). Using approximately 60,000 persona-conditioned annotations from each of two MLLMs, Qwen3-VL and Gemma4, we find that captions converge strongly across persona profiles and show only small attribute-associated differences. Justifications vary substantially more: economic status produces the largest difference in both models, with political orientation and personality also prominent. Paired image-level comparisons confirm larger justification than caption differences for these three attributes. For perception tags, personas sharing the same attribute level produce more similar tag sets than personas with different attribute levels, with the largest separation observed for economic status. Exploratory topic analysis further suggests persona-specific evaluative emphasis. Across models, profile-pair similarity patterns are strongly correlated for all three output types, although agreement is lowest for justifications. Overall, persona prompting affects interpretive framing more strongly than descriptive grounding. --- ### Analyzing Persona Effects in Generated Explanations from Multimodal LLM Agents in Urban Perception Authors: Neemias da Silva, Myriam Delgado, Rodrigo Minetto, Daniel Silver, Thiago H Silva Published: 2026-05-27 Categories: cs.CL, cs.CV, cs.HC, cs.MA arXiv: https://arxiv.org/abs/2605.29064v1 We study how persona prompting shapes language generated by multimodal large language models in an urban perception setting. Using 59,808 annotations from 1,200 persona-conditioned agents and two no-persona settings, we analyze captions, justifications, and perception tags across personas. Results indicate strong convergence in captions for different personas, whereas justifications display systematic variation associated with socioeconomic and political attributes, while perception tags show no statistically significant persona-related differences, though effect trends are observed. Topic analysis further reveals that personas emphasize different evaluative themes when interpreting the same scenes. --- ### Accelerating HEVC Intra Partitioning via a CNN-Hierarchical Attention Transformer Hybrid Authors: Krishna Kumar Sharma, Somdyuti Paul Published: 2026-05-27 Categories: eess.IV, cs.CV arXiv: https://arxiv.org/abs/2605.29063v1 The recursive quad-tree partitioning in High Efficiency Video Coding (HEVC) incurs considerable computational overhead, with exhaustive rate-distortion optimization for CTU partition prediction consuming the dominant share of encoding time. Although partition prediction through deep learning has emerged as a viable encoding accelerator, an architectural dichotomy remains largely unaddressed: CNNs are computationally efficient but spatially myopic due to their localized effective receptive fields, failing to capture long range semantic relationships and repetitive textures; conversely, transformer based architectures are better at capturing global context but incur prohibitive CPU latency, a critical liability that impedes deployment which is predominantly CPU-bound. This paper introduces Hybrid Fast Vision Transformer (HFViT), a hybrid architecture designed to accelerate HEVC intra-mode partition prediction. HFViT fuses a reparameterized depthwise-separable convolutional backbone with a Hierarchical Attention Transformer (HAT) mechanism, leveraging a carrier token scheme to enable efficient global information propagation at sub-quadratic complexity. Post-training structural fusion collapses batch normalization into preceding layers to further reduce latency. Comprehensive evaluation reveals the efficacy of HFViT in accelerating HEVC intra-encoding across resolutions. On standard JCT-VC test sequences, HFViT reduces the average VMAF BD-rate penalty by 2.4, 2.6, and 7.9 percentage points on Classes A, B and E, respectively, as compared to the competing ETH-CNN baseline while maintaining CPU inference latency within 8% of the CNN baseline and surpassing it on GPU by 40%, establishing practical viability for real-time encoder integration. --- ### Bosses, Kings, and the Commons: Cooperation Under Power Asymmetry in LLM Societies Authors: Abhilekh Borah Published: 2026-05-27 Categories: cs.CL arXiv: https://arxiv.org/abs/2605.29062v1 Communities can sustainably manage shared resources (commons) through self-governance and cooperative norms, a central finding of Ostrom's theory of self-governance. However, real-world commons (e.g., fisheries, forests, and irrigation systems) are often governed under asymmetric power structures, where certain individuals or institutions possess disproportionate control over resource extraction and collective outcomes. As Large Language Models (LLMs) are increasingly explored as agents in synthetic governance simulations, understanding how LLM societies behave under asymmetric power structures is becoming increasingly important, yet existing evaluations largely ignore such asymmetries. We introduce Sovereignty over the Commons Simulation (SovSim), a generative multi-agent simulation framework that incorporates an agent with asymmetric power (boss or king) into a society of symmetric agents (workers or peasants), where all agents extract from a shared resource, collectively determining its sustainability over time. Across eleven state-of-the-art models, we find that introducing asymmetric power leads to severe breakdowns in cooperation and sustainability, with up to an 87.3% degradation in survival rate relative to symmetric settings. --- ### SCDBench: A Benchmark for LLM-Based Smart Contract Decompilers Authors: Kaihua Qin, Dawn Song, Arthur Gervais Published: 2026-05-27 Categories: cs.SE, cs.AI, cs.CR arXiv: https://arxiv.org/abs/2605.29059v1 Smart contract decompilation aims to recover high-level source code from bytecode, but evaluating decompilers remains difficult because existing studies use narrow datasets, inconsistent metrics, and limited semantic consistency checks. This gap is increasingly important as large language models (LLMs) begin to generate source-like Solidity that may compile and appear plausible, even when its semantics diverge from the original contract. We introduce SCDBench, a dataset and benchmark methodology for LLM-based smart contract decompilation. The dataset contains 600 real-world Solidity contracts with paired bytecode inputs, ground-truth source code, and replayable semantic checkpoints. SCDBench evaluates decompiler outputs through four cumulative stages: format completeness, compilability, Application Binary Interface (ABI) recovery, and semantic consistency via differential replay. We evaluate Claude Opus 4.7, GPT-5.3-Codex, and GLM-5 in a zero-shot decompilation setting, including GLM-5 variants with and without extended reasoning and a zero-shot compilation-repair setting. The results show that frontier LLMs can often produce structured and compilable Solidity, but achieving semantic consistency remains far from solved: the best-performing frontier model perfectly decompiles only 42/600 contracts. We further show that introducing same-model compilation repair substantially improves performance at modest additional cost. SCDBench establishes a common ground for rigorous, reproducible evaluation and aims to accelerate the development of reliable smart contract decompilers for blockchain security and transparency. --- ### Parallel Adaptive Multi-Objective Evolutionary Learning of Discretized Bayesian Network Classifiers for Clinical Data Authors: Damy M. F. Ha, Thalea Schlender, Yvette M. van der Linden, Peter A. N. Bosman, Tanja Alderliesten Published: 2026-05-27 Categories: cs.LG arXiv: https://arxiv.org/abs/2605.29058v2 Bayesian Networks (BNs) are of interest from an explainable AI viewpoint, offering transparent probabilistic models for decision support. Baymex is a recently introduced multi-objective evolutionary algorithm for learning discretized BNs, enabling experts to trade-off different objectives of interest, such as likelihood, model complexity, and prior beliefs. While Baymex has been shown to outperform state-of-the-art BN learning approaches, Baymex still 1) requires a lot of computation time and 2) has only been evaluated on synthetic data. To improve scalability, we introduce a parallelization strategy as well as a mechanism that enables adaptively steering optimization toward networks that overfit less. We furthermore reconfigure Baymex to train a BN classifier through multi-objective optimization of cross-entropy loss and the BIC complexity term so as to evaluate its performance on real-world clinical classification tasks. Besides observing speedups up to over 54 times on a 16-core CPU, comparisons against clinically familiar baselines (decision trees, logistic regression, naive Bayes, and random forests) on two open-source (RADCURE and SUPPORT) and one in-house dataset, show that Baymex obtains statistically similar or better predictive performance while producing compact, clinically inspectable BNs. Importantly, Baymex finds multiple plausible BN classifiers that contain predictors consistent with established clinical factors. --- ### Hallucination Mitigation with Agentic AI, Nested Learning, and AI Sustainability via Semantic Caching Authors: Diego Gosmar, Deborah A. Dahl Published: 2026-05-27 Categories: cs.AI, cs.MA arXiv: https://arxiv.org/abs/2605.29055v2 This paper describes an approach to hallucination detection and mitigation using a HOPE-inspired Nested Learning architecture with Continuum Memory Systems (CMS) and semantic similarity caching, tested on a hybrid benchmark of 310 prompts (217 epistemic-uncertainty prompts, 93 fabrication-induction stress tests). A three-stage pipeline orchestrated via the Open Floor Protocol is evaluated with five KPIs; four score the response and aggregate into a Total Hallucination Score. The score improves end to end by 6.1% of its attainable range, but 83.5% of that gain is attributable to a single dimension, Explicit Contextualization, while Factual Claim Density, the dimension closest to unsupported content, stays flat; 97.7% of the gain arrives at the first review stage. The fifth indicator, observability, is reported separately, since it registers the presence of an OFP annotation channel rather than a property of the response: it rises 147% at the review stage, the only stage carrying explicit hallucination markers, then falls back at the final stage, which does not propagate the channel. Three annotators independently labelled every final-stage response on the 93 stress prompts: in 10 of 93 cases (10.8%, 95% CI 5.9-18.7) the final answer still presents the invented item as real, at alpha=0.586, below the conventional threshold; Explicit Contextualization tracks these labels monotonically. Re-scoring all 930 outputs with Llama 3.1, Gemma 4, and Qwen 3 as judges confirms the gain and ranks the cross-family judges above the original evaluator against human labels (rho=-0.772 vs -0.477). Semantic caching serves 47.7% of model calls. A nominally multi-dimensional reliability score is thus effectively one-dimensional, and only that dimension has external support. --- ### Hallucination Mitigation with Agentic AI, Nested Learning, and AI Sustainability via Semantic Caching Authors: Diego Gosmar, Deborah A. Dahl Published: 2026-05-27 Categories: cs.AI, cs.MA arXiv: https://arxiv.org/abs/2605.29055v1 Hallucination remains a major reliability barrier for production LLM systems, particularly in multi-agent pipelines where unsupported claims can propagate unchecked across stages. This paper adapts a HOPE-inspired Nested Learning architecture with Continuum Memory Systems (CMS) and semantic similarity caching to a hybrid benchmark of 310 prompts combining 217 epistemic-uncertainty prompts and 93 fabrication-induction stress-test prompts. A three-stage agentic pipeline orchestrated via the Open Floor Protocol (OFP) is evaluated with five KPIs -- FCD (Factual Claim Density), FGR (Factual Grounding References), FDF (Fictional Disclaimer Frequency), ECS (Explicit Contextualization Score), and OSR (Observability Score Ratio) -- aggregated into THS (Total Hallucination Score) across five weighting configurations to study mitigation-observability trade-offs. FDF, ECS, OSR, and FGR are subtracted as mitigation signals, so that a more negative THS indicates stronger mitigation. The FrontEndAgent is configured as a high-stochasticity generator (temperature = 1.0) to produce a realistic hallucination baseline, while the SecondLevelReviewer and ThirdLevelReviewer operate as progressive correctors. This asymmetric design yields end-to-end THS reductions of -31.3% to -35.9% across five weighting configurations. Semantic caching achieves 440 cache hits over 930 potential calls (47.3% hit rate), reducing LLM invocations to 490, lowering energy and CO2e footprint, and making multi-stage review pipelines operationally viable at production scale. ExtremeObservability attains the most negative final THS (-0.0709), confirming that observability-heavy configurations reinforce rather than compromise mitigation. These findings suggest that memory-augmented multi-agent designs can jointly improve factual reliability, operational efficiency, and auditability without model retraining. --- ### Converted, Not Equivalent: Benchmarking Codebase Conversion via Observational Equivalence Authors: Linxin Song, Jiefeng Chen, Yue Huang, Bhavana Dalvi Mishra, Chi Wang, Jieyu Zhao, Jinsung Yoon, Tomas Pfister Published: 2026-05-27 Categories: cs.SE, cs.CL arXiv: https://arxiv.org/abs/2605.29054v2 Coding agents increasingly act as codebase-scale collaborators that can assist with codebase conversion, but this progress has exposed a critical weakness: agents often over-trust their own local validation routines and declare success on artifacts that satisfy surface checks while violating the semantic contracts users actually care about. This problem is especially acute in codebase conversion, where prior evaluation is largely outcome-driven and therefore unstable: two implementations can match on a shallow outcome, such as a single forward loss, while diverging in gradients, optimizer behavior, or short-horizon training dynamics. We introduce T2J-Bench, a benchmark for codebase conversion that reformulates conversion as transfer under a fixed equivalence contract. A fixed verifier then compares source and converted codebases through three ordered stages: Spec (interface admissibility), Numeric (forward outputs, losses, gradients, and objective-specific tensors), and Behavioral (short training dynamics under fixed seeds). Across 355 blind conversion attempts, the best system reaches only 26.7--28.9% overall pass rate despite Spec pass rates up to 91.1%; a 4.7x token-budget spread yields only a 2.2x pass-rate spread; and all systems overestimate success by 66.6--97.8 points relative to the fixed evaluator. This suggests that failures stem more from contract-misaligned self-validation than from limited budget or backbone strength. --- ### LLMBridge: An LLM Pipeline for End-to-end Referential Bridging Resolution in English Authors: Lauren Levine, Amir Zeldes Published: 2026-05-27 Categories: cs.CL arXiv: https://arxiv.org/abs/2605.29048v2 In this paper, we introduce LLMBridge, a new LLM based system for the task of end-to-end referential bridging resolution in English. Our bridging resolution pipeline combines heuristic pre/post-processing with the natural language inference ability that comes from LLMs. We evaluate our bridging resolution pipeline on three datasets which have been used for referential bridging resolution evaluation in English: ISNotes, BASHI, and GUMBridge. Comparison to previous bridging resolution systems shows that the performance of LLMBridge surpasses previous state-of-the-art (SoTA) systems for all 3 datasets in the challenging End-to-end Evaluation Setting, as well as the Basic Bridging Resolution Evaluation Setting (gold bridging anaphor given). We also conduct a thorough error analysis of the LLMBridge performance, examining what varieties of bridging remain difficult for LLM based systems to identify. With this paper, we release the code for the LLMBridge pipeline. --- ### LLMBridge: An LLM Pipeline for End-to-end Referential Bridging Resolution in English Authors: Lauren Levine, Amir Zeldes Published: 2026-05-27 Categories: cs.CL arXiv: https://arxiv.org/abs/2605.29048v1 In this paper, we introduce LLMBridge, a new LLM based system for the task of end-to-end referential bridging resolution in English. Our bridging resolution pipeline combines heuristic pre/post-processing with the natural language inference ability that comes from LLMs. We evaluate our bridging resolution pipeline on 3 datasets which have been used for referential bridging resolution evaluation in English: ISNotes, BASHI, and GUMBridge. Comparison to previous bridging resolution systems shows that the performance of LLMBridge surpasses previous state-of-the-art (SoTA) systems for all 3 datasets in the challenging End-to-end Evaluation Setting, as well as the Basic Bridging Resolution Evaluation Setting (gold bridging anaphor given). We also conduct a thorough error analysis of the LLMBridge performance, examining what varieties of bridging remain difficult for LLM based systems to identify. With this paper, we release the code for the LLMBridge pipeline. --- ### Differentiable Belief-based Opponent Shaping Authors: Aarav G Sane, Karthik Sivachandran, Rohan Paleja Published: 2026-05-27 Categories: cs.AI, cs.LG arXiv: https://arxiv.org/abs/2605.29042v1 Human coordination often relies on the ability to influence the beliefs of others through strategic action. In multi-agent reinforcement learning, opponent shaping attempts to replicate this influence, though existing methods typically operate within an opponent's parameter, policy, or value space. Meanwhile, belief-manipulation techniques in hidden-role games often rely on hard-coded objectives, such as deception or belief saturation. We propose Differentiable Belief-based Opponent Shaping (D-BOS), a first-order method that treats each observer's belief as the shaped opponent state and differentiates through $k$-step softmax-Bayes belief dynamics. Rather than explicitly rewarding deceptive or cooperative behavior, our method treats the belief state as the target for shaping. This allows the optimal strategy to emerge naturally from the environment's reward structure. This belief-space formulation provides an opponent-shaping signal by differentiating through opponent belief updates, and naturally extends to multiple observers by aggregating gradients over their individual inferred belief trajectories. Empirically, D-BOS outperforms PPO and BBM in hidden-role games, with the largest gains in mixed-motive settings. --- ### Practitioner Beliefs and Behaviors in AI-Enhanced Education: DOT Framework Survey Evidence Authors: David Gibson, M. Elizabeth Azukas, Gerald Knezek Published: 2026-05-27 Categories: cs.AI arXiv: https://arxiv.org/abs/2605.29041v1 This study reports findings from a cross-sectional survey (n = 72) of higher education practitioners examining beliefs, behaviors, and institutional conditions related to artificial intelligence (AI) integration in teaching and learning. Grounded in the DOT Framework, which integrates design thinking and open systems theory, the study investigates AI familiarity, usage patterns, design-oriented practices, and pedagogical beliefs. Exploratory factor analysis of 19 belief items identified a three-factor structure: AI Functional Capabilities, Oversight and Governance, and Instructor Collaboration and Planning (α = .90). Results indicate that practitioners hold favorable views of AI as a pedagogical support while maintaining strong commitments to human oversight and critical evaluation. Reported practices emphasize iterative prompting and content generation, with less consistent use of needs assessment and feedback loops. Institutional barriers including limited policy, training, and infrastructure were widely reported. These findings provide preliminary empirical support for the DOT Framework as a descriptive model of practitioner beliefs and practices, while also highlighting gaps between design-oriented theory and current implementation. The study contributes an initial measurement structure and identifies directions for confirmatory validation and outcome-based research linking AI-supported design practices to instructional quality. --- ### Moment Matching Q-Learning Authors: Yiyan, Liang, Sifei Liu, Weitong Zhang Published: 2026-05-27 Categories: cs.LG arXiv: https://arxiv.org/abs/2605.29033v1 Score-based and flow-based generative models exhibit remarkable expressive capacity in capturing complex distributions, and have been extensively deployed in tasks ranging from image generation to reinforcement learning. Nevertheless, these models suffer from prolonged inference latency, which imposes a significant computational bottleneck in RL with iterative sampling. To overcome this limitation, we propose a new framework named Moment Matching Q-Learning (MoMa QL), which utilizes a technique from statistical hypothesis testing known as maximum mean discrepancy (MMD) that intend to match all orders of statistics between the original and target distribution. By enforcing strong regularization on all moment statistics, this algorithm guarantees distribution-level convergence for conditional score function and remains stable under various hyperparameters. Empirically, we show that our method MoMa QL is more computationally efficient with a comparable if not competitive performance in various D4RL tasks. Remarkably, by accelerating the action sampling process for flow-based policies, MoMa QL demonstrates superior performance in offline-to-online RL tasks because of faster and stronger adaptability for online interactive finetuning. --- ### Theoretical Foundations and Effective Algorithms for Policy-Aware Simulator Learning Authors: Christoph Dann, Yishay Mansour, Mehryar Mohri Published: 2026-05-27 Categories: cs.LG, stat.ML arXiv: https://arxiv.org/abs/2605.29032v3 Model-based reinforcement learning (MBRL) agents typically learn world models by minimizing predictive loss. However, powerful RL optimizers inevitably exploit minor model inaccuracies, leading to simulator exploitation and a reality gap where policies succeed in simulation but fail in the real world. We propose that the objective for learning simulators should be strategic robustness rather than predictive accuracy, and formulate this as a zero-sum minimax game between a model player and an adversarial policy player. We provide a comprehensive theoretical analysis: (1) an online learning guarantee showing the game is learnable with sublinear regret bounds; (2) a tractable critic-based simplification bounding the global policy-value gap by the local critic's loss; and (3) an Error-MDP duality, proving that finding the worst-case policy is formally dual to a standard RL problem where the reward is the one-step critic error. This duality yields a provably convergent active data selection algorithm. Experiments on continuous control tasks demonstrate that our approach reduces prediction error in strategically important regions by $1.5$-$2.2\times$ and enables policies trained purely in simulation to match near-optimal real-world performance. --- ### Theoretical Foundations and Effective Algorithms for Policy-Aware Simulator Learning Authors: Christoph Dann, Yishay Mansour, Mehryar Mohri Published: 2026-05-27 Categories: cs.LG, stat.ML arXiv: https://arxiv.org/abs/2605.29032v2 Model-based reinforcement learning (MBRL) agents typically learn world models by minimizing predictive loss. However, powerful RL optimizers inevitably exploit minor model inaccuracies, leading to simulator exploitation and a reality gap where policies succeed in simulation but fail in the real world. We propose that the objective for learning simulators should be strategic robustness rather than predictive accuracy, and formulate this as a zero-sum minimax game between a model player and an adversarial policy player. We provide a comprehensive theoretical analysis: (1) an online learning guarantee showing the game is learnable with sublinear regret bounds; (2) a tractable critic-based simplification bounding the global policy-value gap by the local critic's loss; and (3) an Error-MDP duality, proving that finding the worst-case policy is formally dual to a standard RL problem where the reward is the one-step critic error. This duality yields a provably convergent active data selection algorithm. Experiments on continuous control tasks demonstrate that our approach reduces prediction error in strategically important regions by $1.5$-$2.2\times$ and enables policies trained purely in simulation to match near-optimal real-world performance. --- ### Return-to-Go Is More Than a Number: Q-Guided Alignment for Return-Conditioned Supervised Learning Authors: Yuxiao Yang, Weitong Zhang Published: 2026-05-27 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2605.29028v1 Conditioned Sequence Models (CSMs) learn policies by treating return-to-go (RTG) as a control signal. However, existing CSMs often treat the RTGs as simple numerical inputs rather than aligning them with the performance of their policies. In this paper, we propose Q-ALIGN DT, a framework that enforces this alignment by ensuring the $Q$-value of the output policy is consistent with the input RTG. By leveraging a $Q$ function to provide dense guidance to CSMs and further fine-tuning it using an RTG-perturbation technique with the CSM, our method ensures that higher RTGs are consistently mapped to trajectories with higher expected returns. Theoretically, we show that Q-ALIGN DT can efficiently learn the desired policy and output a near-optimal one when the RTG is sufficiently high. Empirically, we demonstrate through extensive experiments that Q-ALIGN DT achieves superior controllability and performance across the D4RL benchmark. Remarkably, our model effectively learns a structured family of policies that maintains precise alignment and generalizes to tasks like velocity-tracking where prior methods fail. --- ### Mind Your Tone: Does Tone Alter LLM Performance? Authors: Om Dobariya, Akhil Kumar Published: 2026-05-27 Categories: cs.AI, cs.CL, cs.HC arXiv: https://arxiv.org/abs/2605.29027v1 The use of Large Language Models (LLMs) is proliferating, yet their performance is observed to vary based on prompting styles and tones. In this study, we investigate both whether and how tonal variations in prompts lead to disparate LLM accuracy for objective multiple-choice questions. We use two datasets: a 50-base question dataset with five tone variants and a 570-base question MMLU subset spanning 57 subjects with seven tone variants. Experiments were conducted to evaluate the performance of four cost-efficient, popular LLMs: ChatGPT-4o, ChatGPT-5-nano, Gemini 2.5 Flash, and Gemini 2.5 Flash Lite. Across models, tonal effects are systematic but highly model-dependent. Some models show small, yet statistically significant, shifts, while others exhibit large accuracy swings across tones. Further, we identify subject-level differences in tone sensitivity and present a routing framework to explain how tones may attune internal reasoning modes. Our findings caution users against assuming tone-robust reliability in LLM deployments. --- ### When Models Disagree: Rethinking LLM Evaluation for Public Comment Analysis Authors: Aisha Najera, Alvin Moon, Vedant Srinivasan, Rajesh Veeraraghavan Published: 2026-05-27 Categories: cs.AI, cs.CY, cs.HC arXiv: https://arxiv.org/abs/2605.29025v1 Federal agencies are deploying large language models (LLMs) to categorize public comment corpora, where the model's organization of the record shapes what policymakers see and which arguments register. Standard evaluation, anchored on stance accuracy against a small validated set, cannot detect when different models produce materially different categorizations of the same public input. We propose an Interpretive Audit Pipeline that treats multi-model disagreement as diagnostic of interpretive complexity and directs human review toward genuinely ambiguous public input. Analyzing 1,260 public comments on a federal USDA docket across four LLMs, we find that inter-model thematic divergence exceeds within-model prompt variation, and that an expert rubric suppresses deep interpretive disagreement without resolving it. In a two-stage labeling study on a stratified 40-comment subsample, four LLMs and a human annotator labeled independently and then revised after seeing the others' labels. Revision behavior varied across labelers, and the human annotator's revisions frequently introduced framings absent from the ensemble's collective output. We argue disagreement-based evaluation is a necessary complement to accuracy metrics for LLM-assisted interpretive coding. --- ### Designing Active Tether-Net Systems for Space Debris Capture with Graph-Learning-Aided Mixed-Combinatorial Optimization Authors: Feng Liu, Achira Boonrath, Gishnu Madhu, Eleonora M. Botta, Souma Chowdhury Published: 2026-05-27 Categories: cs.LG arXiv: https://arxiv.org/abs/2605.29021v1 Active tether-net systems are a promising solution for capturing large non-cooperative targets, such as space debris, by deploying a flexible net manipulated by maneuverable units (MUs). However, concurrent systematic explorations of design and control choices of the tether-net system to understand its full potential remain limited, partly due to the complex, constrained, nonlinear optimization problem that it presents -- one that involves a mixture of continuous, integer and categorical variables, with the latter two arising from net connectivity and component choices, respectively. Classical binary encoding methods are often ineffective for solving highly nonlinear and multimodal Mixed Combinatorial Nonlinear Programmings (MCNLPs) in engineering design, while integer coding approaches can introduce spurious relations among combinations. Given the graph-structured characteristics of the combinatorial space, this paper adopts and extends a new graph-learning-aided optimization approach to solve this MCNLP problem. Here, a Graph Neural Network (GNN) is trained to score (as output) and thereof recommend candidate combinations represented as nodes in a graph, with the continuous variable vector portion of a candidate design given as input. As a result, the MCNLP optimization reduces to an NLP, which can be solved using standard solvers. While this reduction approach is agnostic to the choice of the NLP solver, here a state-of-the-art Particle Swarm Optimization (PSO) algorithm with gradient-based fine-tuning is used as the solver. Demonstrated on the problem of concurrently designing the morphology of the net, choice of mass and thrusters in the MUs and aiming points used by the controller of the tether-net system, the GNN-based recommender is shown to provide significantly faster convergence to similar optimal solutions, compared to direct solution of the MCNLP problem. --- ### Adopt $\neq$ Adapt: Longitudinal Analyses of LLM Conversations in the Wild Authors: Rebecca M. M. Hicke, Kiran Tomlinson Published: 2026-05-27 Categories: cs.AI, cs.CL arXiv: https://arxiv.org/abs/2605.29018v2 Although a growing body of research has begun to describe user--LLM interactions, the picture it paints is largely static; little is known about how individual users change their behavior over time. To address this gap, we analyze the conversational trajectories of ~12,000 randomly sampled Microsoft Bing Copilot users and compare these with data from WildChat-4.8M. While the Copilot data contains significant population-level trends, we find that trends in individual user trajectories are much weaker; user habits prove to be overwhelmingly sticky. We also find stark differences between users of different activity levels: more active users have more successful conversations and use the LLM for more complex and professionally oriented tasks. Some user trends also appear in WildChat-4.8M, but we find evidence that this dataset is significantly skewed towards highly proficient "power" users. Ultimately, our results suggest that existing user behavior is difficult to change and demonstrate the extent of user heterogeneity. Our comparison between datasets highlights that WildChat does not represent typical user--AI interactions, an important caveat for downstream uses of the data. --- ### Adopt $\neq$ Adapt: Longitudinal Analyses of LLM Conversations in the Wild Authors: Rebecca M. M. Hicke, Kiran Tomlinson Published: 2026-05-27 Categories: cs.AI, cs.CL arXiv: https://arxiv.org/abs/2605.29018v1 Although a growing body of research has begun to describe user--LLM interactions, the picture it paints is largely static; little is known about how individual users change their behavior over time. To address this gap, we analyze the conversational trajectories of $\sim$12,000 randomly sampled Microsoft Bing Copilot users and compare these with data from WildChat-4.8M. While the Copilot data contains significant population-level trends, we find that trends in individual user trajectories are much weaker; user habits prove to be overwhelmingly sticky. We also find stark differences between users of different activity levels: more active users have more successful conversations and use the LLM for more complex and professionally oriented tasks. Some user trends also appear in WildChat-4.8M, but we find evidence that this dataset is significantly skewed towards highly proficient "power" users. Ultimately, our results suggest that existing user behavior is difficult to change and demonstrate the extent of user heterogeneity. Our comparison between datasets highlights that WildChat does not represent typical user-AI interactions, an important caveat for downstream uses of the data. --- ### Three-dimensional Conditional Diffusion Models for Cosmological 21 cm Lightcone Emulation Authors: Bin Xia, John H. Wise Published: 2026-05-27 Categories: astro-ph.IM, astro-ph.CO, cs.LG arXiv: https://arxiv.org/abs/2605.29016v2 We investigate conditional diffusion modeling for three-dimensional 21 cm lightcone emulation, focusing on cubes with a sky-plane size of $64\times64$ and a line-of-sight depth up to 1024 cells. Relative to earlier 2D studies, the 3D setting is substantially harder because memory limits enforce very small micro-batches while the underlying voxel distribution is highly skewed and long tailed. We perform controlled comparisons across preprocessing choices, dynamic-range compression settings, architecture depth, and training duration using $25{,}600$ training lightcones and validation ensembles at fixed parameter points. For validation, each reference parameter point contains 800 21cmFAST realizations with independent initial conditions, and we use 800 samples per model and per reference set for the reported ensemble comparisons. We evaluate generated lightcones with complementary diagnostics in both image and summary-statistic spaces: brightness-temperature slices, the global signal, the power spectrum, and reduced scattering coefficients. Across the tested configurations, preprocessing is the dominant factor governing stable training and the resulting physical fidelity. Among the configurations explored here, Yeo-Johnson preprocessing combined with moderate amplitude compression gives the most consistently favorable trade-off, with the strongest quantitative support coming from rankings based on the standard-deviation-normalized mean absolute error ($\mathrm{MAE}_{\rm std}$) of the global signal and qualitatively compatible behavior in the complementary diagnostics. At the same time, visually plausible 3D samples still retain measurable biases in two-point and higher-order statistics. We therefore view the present work as a simulation-level baseline for three-dimensional 21 cm emulation and for future studies that incorporate more realistic observational effects. ---