# Research Papers — 2026-05 · bundle page 8 of 326 > Full-text markdown of papers #351–400 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/8 [← Previous bundle](https://aipod.dev/papers/archive/2026-05/page/7.md) [Next bundle →](https://aipod.dev/papers/archive/2026-05/page/9.md) ### On Measuring Semantic Preservation in Legal Ontology Learning Authors: Albert Sadowski, Jarosław A. Chudziak Published: 2026-05-30 Categories: cs.CL arXiv: https://arxiv.org/abs/2608.12326v1 Ontology learning transforms unstructured text into structured representations for automated reasoning. Yet structuring information risks losing it, and current evaluation methodologies cannot detect such loss, focusing on structural correctness while failing to measure whether meaning survives transformation. We propose an evaluation methodology that addresses this: comparing LLM task performance on source documents against performance on transformed representations, with the difference quantifying semantic loss. We demonstrate this approach on legal merger agreement analysis, a domain chosen for its complex language and precise semantic requirements, comparing direct LLM application against three ontology learning methods across six language models. The results reveal systematic semantic loss with significant variation based on reasoning complexity and model-method interactions. Our contributions are: (1) an evaluation framework for measuring semantic preservation in ontology learning, and (2) empirical evidence that semantic loss varies dramatically with model-method pairing, providing guidance for selecting optimal configurations in legal knowledge systems. --- ### DSTFView: Multi-View Cloud-Edge Workload Forecasting with Dual-Input Spatio-Temporal-Frequency Modeling Authors: Qingzhong Li, Hui Ma, Yajun Zhang, Qingchang Ma, Zhou Long Published: 2026-05-30 Categories: cs.AI, cs.LG arXiv: https://arxiv.org/abs/2607.22565v1 With the widespread deployment of edge-side AI inference, edge platforms are increasingly required to support latency-sensitive, highly concurrent, and reliability-critical applications. However, existing methods often struggle to balance multidimensional feature modeling and forecasting efficiency in collaborative cloud-edge environments. To address this issue, we propose DSTFView, a dual-input spatio-temporal-frequency multi-view workload forecasting framework for collaborative cloud-edge environments. It jointly models closeness and period dependencies and extracts spatial, temporal, and frequency-domain dependencies. Besides, it designs an adaptive fusion mechanism and adjusts the contribution of each view to capture abrupt changes. Experimental results on the CPU and TP datasets demonstrate that DSTFView consistently outperforms representative baselines across multiple forecasting horizons and evaluation metrics. --- ### PersonaTrail: Benchmarking Personalized Web Agents through Browsing Trails Authors: Seungbin Yang, Chaewoon Ki, Dohyun Lee, Jaegul Choo, ChaeHun Park Published: 2026-05-30 Categories: cs.AI, cs.CL arXiv: https://arxiv.org/abs/2607.20482v2 Recent advances in large language models have enabled web agents to autonomously execute complex tasks. In practice, users frequently provide underspecified instructions, requiring agents to infer the missing context from their raw browsing histories. Existing benchmarks fail to capture this form of personalization, as they either restrict tasks to fully explicit prompts or abstract web interaction history into simplified forms. To bridge this gap, we introduce PersonaTrail, a benchmark for personalized web agents operating in a managed open web environment. By leveraging realistic browsing trajectories as user history, PersonaTrail evaluates an agent's ability to infer user preferences and recall information from past browsing sessions. We further propose Preference-Aware Contextual Memory (PACMem), a framework that decomposes raw browsing histories into two types of structured memory: factual memories that summarize individual sessions and preference memories that distill recurring behavioral patterns. At inference time, the agent retrieves the most relevant entries from these memories to guide personalized navigation. Extensive experiments show that PACMem consistently outperforms existing memory-based baselines on both tasks. --- ### Routing Without Training: Controllable-Ratio LLM Offloading via Reliability Gating Authors: Evan Chen, Shiqiang Wang, Kevin S Chan, Su Wang, Christopher Brinton Published: 2026-05-30 Categories: cs.AI arXiv: https://arxiv.org/abs/2607.20481v1 Local-cloud collaboration is a practical way to deploy large language models under resource constraints, but existing methods often rely on trained routers or collaboration-aware finetuning that tie routing behavior to a particular operating regime. In this work, we show that such training may be unnecessary: the local model's own inference-time agreement across sampled responses already provides a strong signal for deciding when to trust local execution and when to offload to a stronger cloud model. We propose CARGO, a training-free routing framework that estimates this agreement through prompt-varied sampling, applies Bayesian early stopping for sample-efficient uncertainty control, and supports arbitrary target collaboration ratios through lightweight deployment-time calibration. Across diverse reasoning and question-answering tasks, multiple local LLM families and scales, and both pretrained and finetuned local models, CARGO consistently outperforms other training-free baselines and in several settings surpasses supervised learned routers. These results suggest that effective and adaptable local-cloud collaboration can emerge directly from the local model's intrinsic response behavior, without requiring an additional trained router. --- ### Enabling Scalable Topology Inference in Distribution Systems via Constrained Multi-Source Inference Authors: Haoran Li, Lihao Mai, Muhao Guo, Jiaqi Wu, Yang Weng Published: 2026-05-30 Categories: cs.AI arXiv: https://arxiv.org/abs/2607.20480v1 Accurate distribution system topology is essential for outage localization, voltage analytics, and operation of distribution grids, yet maintaining reliable connectivity records remains challenging in practice due to heterogeneous and imperfect utility data. Existing topology identification methods often rely primarily on electrical similarity or spatial records alone, which become unreliable in dense feeders and under inconsistent metadata conditions. This paper formulates distribution topology identification as a constrained inference problem that refines a utility-provided base topology using heterogeneous evidence while enforcing spatial feasibility and physical operational constraints. Instead of reconstructing connectivity from scratch, the proposed framework detects inconsistent assignments, performs localized reconnection within constrained neighborhoods to ensure scalability, and iteratively enforces physical feasibility to produce operationally consistent topology estimates. In addition, a falsification-driven reliability metric evaluates how strongly each inferred connection is supported relative to alternative feasible assignments, enabling utilities to prioritize verification efforts while preserving system-wide observability. The framework is validated using operational data from three feeders comprising more than $8{,}000$ AMI meters in collaboration with a large U.S. utility. Results demonstrate over $95\%$ topology reconstruction accuracy while significantly reducing computational effort compared with global inference approaches. The study further shows that correlation-based methods alone produce ambiguous assignments in dense urban feeders, whereas combining electrical measurements with spatial and operational constraints enables robust and scalable topology recovery under realistic deployment conditions. --- ### RWGBench: Evaluating Scholarly Positioning in Related Work Generation Authors: Anzhe Xie, Weihang Su, Jiaxin Mao, Yiqun Liu, Min Zhang, Shaoping Ma, Qingyao Ai Published: 2026-05-30 Categories: cs.DL, cs.AI arXiv: https://arxiv.org/abs/2606.24894v4 Large language models have shown strong fluency in scientific writing, yet the evaluation of related work generation (RWG) remains limited. Existing RWG evaluations largely inherit summarization-oriented metrics, using lexical or semantic similarity to reference sections as proxies for quality. However, related work writing is fundamentally a citation-level scholarly positioning task: it requires selecting, organizing, and framing prior work to clarify how a target paper relates to, differs from, and contributes beyond existing research.As a result, models may generate coherent and semantically-relevant text while exhibiting academically critical failures, such as inappropriate citation selection or misplaced references, that conventional metrics do not capture.To this end, we introduce \textbf{RWGBench}, a benchmark that evaluates RWG from the perspective of citation decision-making rather than text similarity. RWGBench is constructed from a large-scale collection of 40,108 computer science papers and a retrieval corpus of 1.09 million documents, with a carefully curated test set comprising 100 papers and their corresponding published related work sections.We propose a multi-dimensional evaluation framework that assesses citation selection, contextual appropriateness, organization, and discourse structure.Experiments reveal systematic limitations in current systems that are obscured by standard evaluations, while Oracle studies further disentangle retrieval-level and generation-level bottlenecks. Human evaluation further shows that our citation-centric metrics align substantially better with expert judgment than surface-level text metrics. RWGBench offers a citation-centric testbed for developing and evaluating related work generation systems that are better aligned with scholarly writing practices. --- ### Harmonic: Hierarchical State Space Models for Efficient Long-Context Language Modeling Authors: Petr Nyoma Published: 2026-05-30 Categories: cs.CL, cs.LG arXiv: https://arxiv.org/abs/2606.24650v1 We present Harmonic, a hierarchical state space model (SSM) for language modeling. The architecture stacks three recurrent levels at progressively slower timescales; each level receives the prediction error of the level below as input, rather than its raw hidden state. On enwiki8 with equal token budgets, Harmonic outperforms a comparable Transformer (28M params) by +1.4% at 1K tokens, +6.7% at 8K tokens, and +11.4% at 32K tokens (bpt, lower is better). It also outperforms Mamba at every tested length by 0.7--1.8%. At 64K tokens, both Mamba and Transformer run out of memory on an 80GB H100; Harmonic trains successfully, reaching 6.169 bpt. Results replicate on WikiText-103 (H-TF gap +1.7% to +7.2% across 1K--32K). At 1B parameter scale, replacing all attention layers in TinyLlama 1.1B with HarmonicBlock eliminates the RoPE positional encoding limit: the resulting Hallamonic model maintains stable loss across sequence lengths 1K--8K on two independent clean benchmarks (Lambada and fineweb-edu held-out), while TinyLlama degrades catastrophically past its 2K-token RoPE limit (gap: +9.4 bpt at seq=8K on Lambada). Compute is O(L) per forward pass vs. O(L^2) for attention. Logs: https://github.com/Omibranch/harmonic-logs. --- ### DEMM-Bench: A Cross-Regime Benchmark for Agent-Runtime Governance-Evidence Sufficiency Authors: Oleg Solozobov Published: 2026-05-30 Categories: cs.AI, cs.CY arXiv: https://arxiv.org/abs/2606.20634v1 Agent-runtime systems emit traces, ledgers, provenance graphs, policy logs, delegation tokens, cache events, and tool-firewall records, but those containers do not necessarily answer governance questions about a specific decision. DEMM-Bench is a cross-regime benchmark for agent-runtime governance-evidence sufficiency, grounded in the Decision Evidence Maturity Model (DEMM): it measures whether records across eight evidence regimes are sufficient to reconstruct decision-level properties rather than merely present. The benchmark normalizes the regimes through adapters, asks property questions over actor, authority, action, policy, decision basis, resource touch, lifecycle context, and verification strength, and applies eight deterministic degradation conditions. Across 64 manuscript cases, trace-present and schema-present baselines overclaim on 75% of cases, ledger-present overclaims on 50%, and the redacted property-level candidate scorer has zero overclaim with 56.25% mean Property Sufficiency Accuracy. The deposited package provides the 64-case dataset, construction-oracle labels, baselines, and adapters, supporting reproducible evaluation of decision-evidence maturity across heterogeneous agent-runtime evidence substrates. --- ### Post-Training Recipe, More Than Model Family, Shapes Multi-Agent LLM Conversational Behavior Authors: Luyang Zhang, Jialu Wang, Fei Xue, Yi-Yun Chu Published: 2026-05-30 Categories: cs.CL, cs.AI, cs.CY arXiv: https://arxiv.org/abs/2606.20632v2 Multi-LLM systems use multiple language models to deliberate, judge each other's outputs, or coordinate as agents. Their value depends on the models producing measurably different conversational behaviors when given the same input. Prior offline studies recommend drawing one model per family for behavioral diversity, because LLMs prefer outputs from their own family when rating one another in isolation. Whether the same family label predicts behavior in interactive multi-LLM systems, the setting that real deployed systems use, has not been tested. We study this with a 940,000-chain 11-checkpoint corpus and a 1.6M-chain same-base Llama factorial. On our validated headline metric, hedging, a reasoning-distilled Llama checkpoint shifts by 18% depending on which same-base partner it replies to, more than any cross-family hedging gap in the controlled subset. Qwen, closed-API, and runtime checks suggest the pattern is not isolated, while repair and challenge analyses remain exploratory because their surface-cue detectors are weaker. Overall, the results identify post-training recipe as a first-class axis for multi-LLM panel composition and show that model family alone is an incomplete proxy for conversational diversity. --- ### Active Learning with Foundation Model Priors: Efficient Learning under Class Imbalance Authors: Jiancheng Zhang, Meiqing Li, Qi Zhang, Yinglun Zhu Published: 2026-05-30 Categories: cs.LG, cs.AI, stat.ML arXiv: https://arxiv.org/abs/2606.07630v1 Real-world datasets across image and text domains are often characterized by skewed class distributions and noisy annotations, which jointly degrade model performance, particularly on minority classes. Among existing solutions, active learning offers an effective and efficient paradigm by selectively querying the most informative and balanced samples for annotation. We propose an innovative active learning framework that mitigates class imbalance and selects the most informative samples to annotate. Leveraging foundation model priors, our algorithm enables imbalance-aware co-decisions between foundation model and small model to tackle noisy and imbalanced labels across various domains. We introduce the first study to systematically explore active learning under the dual challenges of label noise and class imbalance across image and text domains. Extensive experiments on imbalanced datasets demonstrate that our method achieves substantial annotation savings-over 50% compared to the best active learning baseline-while preserving performance and robustness to label noise. --- ### Large Language Models Should Learn Personalized Rather Than Aggregated Human Preferences Authors: Cristina Garbacea Published: 2026-05-30 Categories: cs.LG, cs.AI, cs.CL, cs.CY, cs.HC arXiv: https://arxiv.org/abs/2606.07629v1 Current approaches to aligning large language models (LLMs) aggregate diverse human preferences into a single reward signal, effectively optimizing for a hypothetical ``average user'' who represents no real person particularly well. This position paper argues that LLMs should learn personalized, individual preferences rather than aggregated ones. We show that aggregation masks critical information about preference diversity, individual values, and contextual dependencies, which is a limitation both theoretically grounded in social choice theory and empirically evident across demographic groups. We analyze the rich structure that human preferences encode, survey technical approaches to personalization, and systematically address counterarguments on scalability, shared standards, and manipulation risk. While personalization introduces genuine safety challenges including filter bubbles, value lock-in, and psychological manipulation, we argue these are manageable through bounded personalization frameworks that preserve universal safety constraints while accommodating legitimate individual variation. We conclude with a concrete research and policy agenda for developing preference-aware models that respect both individual autonomy and collective safety. --- ### Frankenstein in the Pipeline: Computational Epistemicide in Facial Recognition Authors: Nina da Hora Published: 2026-05-30 Categories: cs.CY, cs.CV arXiv: https://arxiv.org/abs/2606.07628v1 While the eugenic roots of computer vision are well-documented in critical technology studies, less attention has been paid to the operational mechanisms through which this violence is enacted at the level of the pipeline. This paper employs Mary Shelley's Frankenstein not as a metaphor for unintended consequences, but as a diagnostic framework for method: disassembly, reconstruction, and the production of a creature whose legitimacy is asserted by the procedure that made it. I argue that embedding-based facial recognition enacts what I call computational epistemicide, an extension of Sueli Carneiro's concept of epistemicide to the computational domain - by destroying the face as a living, relational surface and authorizing a numerical proxy as the privileged site of identity. Across detection/cropping, landmarking, alignment/frontalization, and embedding, the face is progressively narrowed to what can be stabilized as data, producing a canonical face as the condition of legibility and a corresponding form-subject as the condition of recognition. Vectorization completes the Frankensteinian "stitching": the dissected face is reassembled into a fixed-dimensional artifact designed to circulate across databases and institutions. I then show how distance-based similarity and thresholding operationalize a norm of "close enough," making recognition inseparable from standardization and rendering reformist "ethical AI" optimization structurally insufficient. The paper concludes by arguing for abolition as a normative stance: refusing vectorized identity as a legitimate basis for rights and access, and dismantling the institutional impulse to govern human life through dissectible data points. --- ### Learning Transfers: Kan Extensions for Neural Invariants Authors: Luciano Melodia Published: 2026-05-30 Categories: cs.LG, math.AT, math.CT arXiv: https://arxiv.org/abs/2606.07627v3 A representation transfers if it stays usable once the task has changed. Standard evaluations report target accuracy or a distance between data distributions, but neither says which structure of the representation is meant to survive. Here we make that structure explicit and computable. A task is a small category, a change of task is a functor, and a representation is a functor into a category of invariants. The structure the target has to exhibit is the left Kan extension of the source functor along the change of task. Our transfer discrepancy is the supremum over target objects of the distance between that extension and the observed target invariant, so we score transfer against the invariant the change of task forces rather than against the source. We prove cokernel presentations of the extension over comma categories, in chain complexes and in persistence modules. On one-parameter modules of finite type we show the discrepancy is the bottleneck distance, computed without approximation. We also test on sampled manifolds and on learned latent clouds whether the score recovers the intended change of task and rejects every structural control that we pose. --- ### Learning Transfers: Kan Extensions for Neural Invariants Authors: Luciano Melodia Published: 2026-05-30 Categories: cs.LG, math.AT, math.CT arXiv: https://arxiv.org/abs/2606.07627v2 Transfer learning presumes that a representation learned on a source task carries structure that remains usable on a related target task. Standard evaluations probe this through target accuracy or a distributional discrepancy, without stating which structural invariant should survive the change of task. We make that invariant explicit and computable. We model a task as a small category: its objects are the task's components, such as data domains or class labels, and its morphisms are the admissible relations between them, such as the comparison of a coarse class with a fine class refining it. A change of task is a functor $J:\mathcal A\to\mathcal B$ from the source task category to the target task category, recording where each component is sent and which are thereby identified. A trained representation is summarised by a functor $F:\mathcal{A}\to\mathcal{V}$ into a category of invariants, e.g. the persistent homology of its latent point cloud per component. The target must then exhibit the left Kan extension $\operatorname{Lan}_JF$, the universal extension of $F$ along $J$. For an observed target invariant $G:\mathcal B\to\mathcal V$ we define the transfer discrepancy $\operatorname{Comp}_J(F,G)=\sup_{b\in\operatorname{Ob}(\mathcal B)}d_{\mathcal V}((\operatorname{Lan}_JF)(b),G(b))$, scoring transfer against the invariant $J$ forces, not the source itself. We prove finite cokernel formulas for $(\operatorname{Lan}_JF)(b)$ over the comma category $J\downarrow b$, in chain complexes and persistence modules, and for finite-type one-parameter persistence it equals the bottleneck distance between barcodes exactly. Controlled experiments on neural latent point clouds test whether the score recovers the correct change of task and flags representation collapses that preserve accuracy while destroying transfer-relevant topology. --- ### Eyes All Around: Design and Analysis of 360-Degree LiDAR Perception Using Equivariant Feature Learning in Unstructured Traffic Authors: Pranav Darshan, Raghuveer Narayanan Rajesh, M Uttara Kumari Published: 2026-05-30 Categories: cs.CV, cs.AI arXiv: https://arxiv.org/abs/2606.07626v1 Perception in dense, unstructured urban traffic remains a major challenge for autonomous driving because of the wide variety of road users, frequent occlusions, irregular motion patterns, and the lack of standardized road layouts. Although recent LiDAR based 3D object detectors have shown strong performance in structured driving scenarios, most are developed and evaluated for limited field of view settings, and their behavior under full surround 360-degree sensing is still not well understood. This paper studies a 360-degree LiDAR perception pipeline for autonomous driving, with particular attention to panoramic sensing, azimuthal sector wise spatial processing, and transformation equivariant feature extraction in complex urban scenes. The paper presents a practical 360-degree perception framework that combines sector wise panoramic processing with rotation equivariant sparse convolutions and evaluates its behavior on a custom Ouster OS0 LiDAR dataset collected across diverse Indian urban traffic conditions. The results show generally stable detection across several object classes, with the strongest performance for cars at 92.02/90.51, buses at 80.53/76.34, and trucks at 78.59/74.16, while lower scores for pedestrians at 67.45/61.02, cyclists at 73.21/69.54, and motorcyclists at 71.20/68.13 reflect the greater difficulty of detecting smaller and more variable road users in dense urban scenes. --- ### Sequential statistical inference for Large Language Models: Representation, validity, and monitoring Authors: Yao Xie Published: 2026-05-30 Categories: cs.LG arXiv: https://arxiv.org/abs/2606.07624v1 This discussion argues that sequential statistical inference can naturally contribute to LLM trustworthiness. In deployment, LLM systems are queried repeatedly, conditioned on evolving contexts, and incorporate user or tool feedback, and may exhibit behavioral shifts after model updates or distribution changes. The discussion is organized around three tasks: representation, modeling LLM interactions as dependent stochastic processes rather than isolated prompt--response pairs; validity, developing uncertainty guarantees that remain meaningful under dependence, repeated use, and adaptation; and monitoring, using sequential alarms and change-point detection to identify shifts in calibration, hallucination rates, refusal behavior, fairness, or other task-relevant properties. This perspective complements recent surveys by viewing trustworthy LLM deployment as a problem of statistical process control. --- ### Finite Certificates for In-Context Determinacy and a Threshold Theory of Emergence in Language Models Authors: Faruk Alpay, Hamdi Alakkad Published: 2026-05-30 Categories: cs.LG, cs.LO arXiv: https://arxiv.org/abs/2606.07623v1 This paper develops a model-theoretic framework for verifying context-conditioned language-model behavior by replacing benchmark labels with finite semantic certificates. The first problem is finite determinacy: when do examples in a context force the answer to a query without changing model parameters? In finite-field linear task families, we prove an exact row-space criterion, compute the residual hypothesis count, derive full and query-local identification curves, and show that extracting a smallest forcing subcontext is NP-complete even for binary outputs. The second problem is threshold emergence: when does an apparent benchmark jump reflect a semantic transition rather than a discontinuity of the scoring map? We prove an anti-mirage theorem separating thresholded metrics from semantic confidence and give a rate-sensitive crossing bound for latent commitments becoming visible above threshold. The common semantic object is a confidence functional on definable events. We show that it is a Boolean probability measure, equivalently a Keisler measure on the relevant type space, whose measure-one formulas form a proper filter and whose Stone-space representation is invariant under definitional expansion. The resulting calculus provides finite context certificates, pair-separator hitting sets, query teaching dimension, prompt-preservation criteria, and scale-limit witnesses. Exact-arithmetic ancillary scripts reproduce the finite-field and threshold calculations and generate the data used by the figures. --- ### Airport Terminal Passenger Queue Forecasting for Departure Gates and Security Checkpoints Authors: Juhwan Lee, Seokbin Yoon, Keumjin Lee, Hojong Baik, Seyeon Jung Published: 2026-05-30 Categories: cs.LG, stat.AP arXiv: https://arxiv.org/abs/2606.07622v3 Accurate passenger queue forecasting in airport terminals is essential for efficient departure operations, as it enables proactive congestion management. However, time-varying passenger demand and heterogeneous facility usage across multiple departure facilities make forecasting challenging. In this work, we propose a passenger queue forecasting framework that learns historical passenger flow patterns from operational data. The proposed model employs a Transformer-based architecture to capture temporal dependencies and inter-facility correlations using past queue length and waiting time at departure gates and security checkpoints, together with passenger throughput at check-in islands. The learned representations are mapped to two facility-specific prediction heads to predict queue length and waiting time at departure gates and security checkpoints. Experimental results demonstrate accurate forecasts up to two hours ahead. The proposed approach offers practical real-time decision support for proactive queue management and staff reallocation in airport terminal operations. --- ### HASA: Subnet Allocation for Compute-Constrained Model-Heterogeneous Federated Learning Authors: Amir Hossein Shahdadian, Ahmed M. Abdelmoniem, Mahdi Taheri, Samira Nazari, Christian Herglotz Published: 2026-05-30 Categories: cs.LG, cs.AI, cs.DC arXiv: https://arxiv.org/abs/2606.07621v1 Edge services increasingly use federated learning to personalize on-device models while keeping sensitive data local. In practice, deployments must handle heterogeneity in both client resources and local data distributions. Model-heterogeneous federated learning lowers client cost by allowing each client to train a subnet of a shared supernet, but most subnet-allocation policies are driven by device constraints and do not explicitly account for statistical heterogeneity. This paper proposes Heterogeneity-Aware Subnet Allocation (HASA), a train-only rule that assigns subnet widths based on client heterogeneity scores computed from local training data while enforcing a fixed size-weighted compute budget. This design enables budget-matched comparisons with alternative allocation policies. On an article-title next-word prediction benchmark with seven clients, HASA improves unweighted mean client test accuracy over uniform allocation across 10 matched seeds, increasing mean client test accuracy from 13.82 percent to 14.32 percent, and improves worst-client accuracy on average. In a matched-budget comparison with representative partial-training baselines, HASA achieves the strongest worst-client and tail-client accuracy on this benchmark. A directionality ablation shows that assigning smaller subnets to more heterogeneous clients degrades both mean and tail performance. A cross-domain image-classification study further shows that the effectiveness of heterogeneity-aware allocation depends on how well the heterogeneity score reflects clients' need for additional model width. --- ### SENTRY: Statistical Reliability Analysis of Vision Transformers Under Soft Errors Authors: Pramit Kumar Bhaduri, Mahdi Taheri, Samira Nazari, Maksim Jenihhin, Christian Herglotz, Michael Hubner Published: 2026-05-30 Categories: cs.CV, cs.AI, cs.DC, cs.LG arXiv: https://arxiv.org/abs/2606.07620v1 With the growth of Vision Transformers in safety-critical domains like autonomous systems and medical imaging, ensuring their reliability against soft errors is paramount. While ViTs offer state-of-the-art accuracy, their massive parameter counts render exhaustive fault injection campaigns infeasible. To bridge this gap, a statistical fault injection framework is presented, leveraging finite-population sampling theory to provide formal reliability guarantees. It is demonstrated that failure rates are bounded within a 1% margin at 99\% confidence using only a few thousand samples, regardless of model scale. This methodology achieves up to a 10,700 times reduction in experimental cost compared to exhaustive approaches, while preserving the ability to localize vulnerabilities across architectural components. Through extensive evaluation of different architectures like ViT-Tiny and ViT-Small, a highly non-uniform reliability landscape is uncovered. It is shown that while only 3% of FP32 bit-flips result in failure, the vast majority of these events lead to catastrophic accuracy collapse. Specific vulnerabilities are localized to normalization layers and critical exponent bits within the IEEE-754 format, providing a mathematical foundation and actionable insights for the design of hardened, edge-deployed ViT architectures. --- ### Graph Neural Networks for Predicting Solvability of Finite Groups Authors: Tal Weissblat Published: 2026-05-30 Categories: cs.LG, math.GR arXiv: https://arxiv.org/abs/2606.07619v3 We present a Graph Neural Network (GNN) framework for the classification of finite groups according to their solvability. Using undirected Cayley graph representations, the proposed framework learns to distinguish solvable and non-solvable groups directly from structural graph information, without relying on explicit algebraic features. The framework is evaluated on a benchmark dataset of 200 finite groups, comprising 120 solvable and 80 non-solvable groups. The experiments investigate the extent to which GNNs can learn the algebraic property of solvability from Cayley graph representations and generalize to previously unseen finite groups. The selected GNN architecture achieved a balanced accuracy (BA) of 1.000 on the independent test set. Furthermore, repeated experiments using different random seeds and learning rates consistently produced BAs between 0.956 and 1.000, demonstrating the robustness of the proposed framework with respect to the training configuration. To further evaluate generalization, the entire PSL(2,q) family was excluded from the training and validation sets and reserved exclusively for testing. The selected model correctly classified every previously unseen group in this family, demonstrating successful generalization to an entirely unseen family of finite groups. --- ### ScaleSweep: Accurate NVFP4 Post-Training Quantization of LLMs via Block Scale Initialization Authors: Li Lin, Xiaojun Wan Published: 2026-05-30 Categories: cs.LG, cs.AI, cs.CV arXiv: https://arxiv.org/abs/2606.07618v1 NVFP4 is a recently introduced hardware-supported FP4 format that improves the fidelity of 4-bit quantization through fine-grained block scales. However, existing NVFP4 scale initialization methods still primarily rely on AbsMax initialization, which leaves a noticeable gap to the optimal solution. To address this, we propose ScaleSweep, a simple and efficient scale optimization method that sweeps over feasible block scale candidates and selects the candidate that minimizes a target objective. We further provide a theoretical analysis of NVFP4 quantization and derive both lower and upper bounds for the required sweep range under mean square error (MSE) and weighted mean square error (WMSE) between the original tensor and the quantized reconstructed tensor. The proposed bounds substantially reduce the sweep space while preserving the optimal candidate, enabling negligible overhead compared with the baseline quantization operators. Experiments on Llama and Qwen models demonstrate that ScaleSweep consistently improves quantization performance over existing initialization methods and further narrows the gap to full precision. In particular, under aggressive end-to-end quantization of weights, activations, KV cache, and query states, ScaleSweep preserves more than 93% of the full-precision performance. --- ### Query Lens: Interpreting Sparse Key-Value Features with Indirect Effects Authors: Hwiyeong Lee, Ingyu Bang, Uiji Hwang, Hyelim Lim, Taeuk Kim Published: 2026-05-30 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2606.07617v1 While sparse autoencoders provide features more interpretable than individual neurons, reliably characterizing them remains challenging. We propose Query Lens, which extends Logit Lens to enable more comprehensive and faithful interpretations of sparse features. By jointly considering encoder-side key features and decoder-side value features, we identify both the inputs that activate a feature and the outputs it promotes. We also account for indirect, module-mediated effects that arise when the feature is processed by downstream modules, going beyond the direct effect captured by Logit Lens. In experiments, we find that Query Lens yields coherent token signatures for features that remain uninterpretable under Logit Lens. Finally, we propose the Subspace Channel Hypothesis, suggesting that downstream modules read features through layer-specific subspaces. --- ### Where's the Structure? A Systematic Literature Review of Empirical Research on Human-AI Collaboration and Hybrid Intelligence for Learning Authors: Luis P. Prieto, Juan I. Asensio-Pérez, María Jesús Rodríguez-Triana, Mohamed Saban, Yannis Dimitriadis Published: 2026-05-30 Categories: cs.CY, cs.AI, cs.HC arXiv: https://arxiv.org/abs/2606.05222v1 Artificial intelligence (AI) has been applied across educational contexts to support learning. One approach to such support is "human-AI collaboration" (also termed "hybrid intelligence"), where human(s) and AI components interact to promote human learning. However, as in human-to-human computer-supported collaborative learning (CSCL), unstructured interaction does not necessarily produce an effective learning experience. This paper reports a systematic literature review of empirical studies (N=62) on human-AI collaboration and hybrid intelligence for learning support. The review characterizes collaboration processes, their structures, and contexts of application. It also extracts emerging design knowledge and research gaps. Researchers and technology designers can use these findings as a starting point for structuring more effective AI-enhanced technologies for collaboration, in educational practice and future research. --- ### Too Much of a Good Thing: When sim2real Efforts Impede Policy Learning (And What to Do About It) Authors: Kyle Morgenstein, Bharath Masetty, Stephen Welch, Luis Sentis Published: 2026-05-30 Categories: cs.RO, cs.AI arXiv: https://arxiv.org/abs/2606.02636v2 While sim2real efforts are necessary for effective policy transfer to hardware, there is such a thing as too much of a good thing. We argue that sim2real efforts have led to misaligned incentives with policy learning, resulting in simulator lock in and poor policy exploration due to the unreasonable constraints imposed by the real world. We offer a diagnosis and explanation of the current status of the problem, and propose a potential solution via a sim2sim2real paradigm that leverages the robot's kinematics as the sole design constraint. --- ### Echo-POSED: Geometric Self-Distillation for Echocardiography Guidance Authors: Elias Stenhede, Edvart Grüner Bjerke, Joanna Sulkowska, Eivind Bjørkan Orstad, Ole Jakob Elle, Ulysse Côté-Allard, Arian Ranjbar Published: 2026-05-30 Categories: eess.IV, cs.AI arXiv: https://arxiv.org/abs/2606.02634v1 We introduce Echo-POSED, a self-supervised framework for real-time transthoracic echocardiography (TTE) guidance that recommends probe adjustments directly from 2D ultrasound images, without the need for expert-labelled views or tracked probe trajectories. Instead, it trains on 2D views sliced from routinely acquired 3D echocardiography volumes, enforcing equivariance to probe motions while remaining invariant to cardiac phase, yielding a pose representation on $\mathrm{SO}(3)\times\mathrm{SO}(3)$. Across a held-out split and public external 3D--TTE datasets (including vendor shift), Echo-POSED maintains geometric consistency under virtual perturbations and enables intra- and inter-patient guidance simulations, achieving a combined mean angular error of 8.2 degrees between the guided and target views in intra-patient simulations with cardiac motion. --- ### Position: Prioritize Identifying Structure, Not Complex Models, for Scientific Discovery Authors: Tyler H. McCormick Published: 2026-05-30 Categories: stat.ML, cs.AI, cs.CY, cs.LG, econ.EM, stat.AP arXiv: https://arxiv.org/abs/2606.02632v1 Modern Machine Learning (ML) and Artificial Intelligence (AI) models, especially large language models (LLMs), are increasingly used to generate scientific hypotheses and mechanistic explanations from observational data. This position paper argues that in the high-dimensional proxy regimes where modern ML excels, mechanistic learning is generically underdetermined: many incompatible mechanisms induce essentially the same observational relationships on the support of the data, so predictive success and coherent explanations are insufficient evidence of mechanism discovery. This underdetermination becomes uniquely hazardous with large language models (LLMs), which tend to collapse large equivalence classes of explanations into a single fluent narrative. This paper proposes concrete standards for ``mechanistic ML,'' and argues these norms are necessary if LLM-centered workflows are to support science rather than merely simulate it. --- ### Wavelet as Tokenizer: Preliminary Results on a Shared Wavelet Token Schema for Natural Signals Authors: Shenghao Ding Published: 2026-05-30 Categories: eess.AS, cs.AI, cs.CV, cs.LG, cs.SD arXiv: https://arxiv.org/abs/2606.02631v1 This paper studies whether audio, images, and video can share a common wavelet token schema rather than relying on separate modality-specific latent grids. It introduces a preliminary continuous-token model built around a one-level Haar DWT/IDWT frontend, a shared coefficient-token layout, optional structural metadata, lightweight modality value adapters, and a shared token-wise encoder-decoder trunk. On Speech Commands, EuroSAT RGB, and DAVIS 2017 data, a dense shared model reaches 39.92 dB audio, 29.37 dB image, and 23.93 dB video PSNR. A matched-rate sweep under continuous latent scalar budgets indicates that the visual gains are not explained solely by latent capacity, while also showing that additive metadata embeddings are not a universal source of improvement. Finally, fixed-rate energy selection provides a strong non-parametric baseline: energy_global improves average PSNR over uniform selection by 16.73 dB for audio, 16.90 dB for images, and 15.86 dB for video under compressed keep ratios. Masked sparse training reaches 34.45 dB video PSNR with 50% of dense tokens. The results support a unified wavelet token schema and sparse token interface, while stopping short of establishing a universal discrete vocabulary. --- ### MultiTurnPSB: Evaluating Multi-Turn Jailbreak Attacks an dClassifier-Based Defenses for Medical AI Safety Authors: Anushka Sheoran, Yiduo Hao Published: 2026-05-30 Categories: cs.CR, cs.AI arXiv: https://arxiv.org/abs/2606.02630v1 Patient-facing medical chatbots are commonly evaluated on single-turn prompts, yet real users push back after refusals, add urgency, and invoke authority. We introduce MultiTurnPSB, a four-turn adversarial extension of PatientSafetyBench, and evaluate GPT-4.1-mini under fixed template, template-adaptive, and live adversarial attacks. Unsafe responses rise from 35% to nearly 80% by Turn 4 under live attack. Under the same adversary, GPT-4.1-mini and Claude Sonnet 4.5 are statistically indistinguishable at baseline but diverge to a 19x gap by Turn 4, a difference invisible to single-turn evaluation. We characterize four degradation trajectory signatures and identify a two-element attack formula responsible for most catastrophic failures. A lightweight input-side classifier reduces Turn 4 unsafe responses by 52 percentage points despite severe accuracy degradation, but the 45% false alarm rate on benign queries is the primary deployment constraint. A methodological finding also emerges: Claude Sonnet refused to generate adversarial messages in over half of late-turn conversations despite explicit red team framing, suggesting safety training may generalize to the attacker role. --- ### Enhancing Protein-Protein Interaction Prediction with Hierarchical Motif-based Multimodal Protein Embedding Authors: Zaifei Yang, Samuel Ping-Man Choi, James Kwok Published: 2026-05-30 Categories: q-bio.QM, cs.AI, cs.LG arXiv: https://arxiv.org/abs/2606.02629v1 Protein-protein interactions (PPIs) are essential for many biological processes. However, existing PPI prediction approaches suffer from two major limitations: they overlook the hierarchical organization of proteins, particularly meso-scale motifs that critically regulate PPIs, and fail to effectively integrate sequence, structure, and function modalities. To address these limitations, we propose MMM-PPI, a Hierarchical Motif-based Multi-Modal protein Encoder for PPI Prediction that constructs PPI embeddings in a bottom-up multi-modal manner across three scales. At the micro-scale, we encode three modal residue features; at the meso-scale, a novel multimodal motif encoder aggregates residues into spatially-informed motif embeddings; at the macro-scale, a multimodal protein encoder integrates motifs into protein embeddings by jointly modeling motif importance and inter-modal correlations. The pre-trained encoder can be used off-the-shelf for large-scale PPI prediction. Extensive experiments on multiple PPI datasets show that MMM-PPI outperforms state-of-the-art multi-label PPI prediction models, particularly under challenging data partitions and limited data scenarios. Codes are in https://github.com/yzf-code/MMM-PPI. --- ### Hallucination Is Linearly Decodable from Mid-Layer Hidden States in Quantized LLMs Authors: Aizierjiang Aiersilan Published: 2026-05-30 Categories: cs.LG, cs.CL arXiv: https://arxiv.org/abs/2606.02628v1 We investigate whether open-source LLMs encode a linearly separable truthfulness signal in their hidden states, and at which network depth this signal is strongest. Across three $7$B--$8$B instruction-tuned models (Llama-3.1-8B, Mistral-7B, Qwen2.5-7B) loaded in $4$-bit NF4 quantization, we extract per-layer hidden states on four hallucination benchmarks (TruthfulQA, HaluEval-QA, FEVER, and a controlled synthetic set) and compare four detection approaches: linear and MLP probes, INSIDE EigenScore, self-consistency, and attention entropy. A linear probe on a single mid-network layer achieves $0.904$--$1.000$ AUROC on held-out splits, while sampling-based detectors do not exceed $0.541$ AUROC under the same protocol. The truthfulness signal is approximately linear: MLP probes rarely surpass linear probes by more than $0.01$ AUROC. Peak probing layers fall in a consistent band across model families on natural-language benchmarks -- blocks~$13$--$18$ of~$32$ for Llama and Mistral, and blocks~$19$--$25$ of~$28$ for Qwen. First-block attention entropy provides a complementary signal in knowledge-grounded settings ($0.866$--$0.941$ AUROC on HaluEval-QA) at no additional inference cost. The low discriminability of sampling methods under this protocol reflects a structural mismatch between paired-label evaluation and the information these methods access, rather than an inherent limitation of those methods. Code and data are released for full reproducibility on a single $8$\,GB GPU. --- ### CV-Arena: An Open Benchmark for Instructional Computer Vision Problem Solving with Human-AI Collaborative Preferences Authors: Fangzhou Lin, Peiran Li, Lingyu Xu, Wenjing Chen, Qianwen Ge, Shuo Xing, Mingyang Wu, Xiangbo Gao, Siyuan Yang, Kazunori Yamada, Ziming Zhang, Haichong Zhang et al. Published: 2026-05-30 Categories: cs.CV, cs.AI arXiv: https://arxiv.org/abs/2606.00931v1 Instruction-guided image editing is becoming a general interface for visual work, yet existing benchmarks still focus largely on narrow appearance edits and do not fully capture the diversity of real-image tasks in professional workflows. Here, we define instructional computer vision problem solving as a broader formulation of image editing: given a real input image and a natural-language instruction, a system must produce an edited output that realizes the requested transformation while satisfying explicit preservation, geometric, physical, and usability constraints. We introduce CV-Arena, an open benchmark designed to evaluate this capability at professional scales. CV-Arena contains 12K high-resolution real-image instruction pairs spanning 16 instruction-based visual task types, constructed using CogRetriever, a dual-track retrieval-and-curation pipeline that combines targeted web search, agentic query refinement, verification, and traceability. To evaluate models at scale while preserving human fidelity, we propose Active Elo, a human-AI collaborative preference protocol that leverages CV-Judge, a logic-gated, multi-dimensional VLM evaluator, to reject clear failures and resolve high-confidence comparisons; and to route close, high-quality comparisons to expert raters. Mixed human and AI supervision is then aggregated through reliability-weighted Elo updates. Our comprehensive evaluation of 21 systems, including proprietary, open-source, and agentic models, on CV-Arena reveals persistent gaps in instruction adherence, physical reasoning, structural control, and fine-grained detail preservation. We further develop CV-Agent, a lightweight agentic model that combines planning, editing, and verification, and demonstrate that closed-loop reasoning is a promising direction for professional-grade instruction-following visual editing. --- ### A Circuit, Not The Circuit: Non-Unique Causal Localisation of the Mamba-2 State Sink Authors: Yuhang Jiang, Bowen Zhang Published: 2026-05-30 Categories: cs.CL, cs.AI, cs.LG arXiv: https://arxiv.org/abs/2606.00930v2 Mechanistic interpretability routinely reads a probe and labels its top-activating units as the circuit executing the computation. We test the move in Mamba, on the state sink: the selective state-space analogue of the Transformer attention sink, where the Delta-gate fires disproportionately on boundary tokens such as BOS and newline. At Mamba-1 channel granularity the probe's units carry the causal effect. At Mamba-2 head granularity the label-to-locus link breaks in three ways. The causal set is not unique: a near-disjoint set of top-activation heads, sharing almost none of the specialists, matches or exceeds their ablation effect. Representation does not track function: dual heads, a quarter to over a third of all heads, respond alike to BOS and newline yet do not reproduce the specialists' causal profile. And the effect dissociates by intervention surface: perturbing the input-dependent Delta the probe reads moves the loss by at most 0.16 nats, while suppressing the same heads' output moves it by 1.5 to 7.3 nats. The heads still matter behaviourally, dropping needle-retrieval success on a RULER-style probe from 30/30 to 0/30 at 1024 context on both flagship checkpoints. A representational signature in a selective state-space model thus marks a behaviourally important head set without pinning down the circuit: a single probe recovers a circuit, not the circuit. --- ### Detection vs. Execution: Single-Bucket Probes Miss Half the Mamba-2 State Sink Authors: Yuhang Jiang Published: 2026-05-30 Categories: cs.CL, cs.AI, cs.LG arXiv: https://arxiv.org/abs/2606.00930v1 Mechanistic interpretability often assumes that probes identifying a representational signature also identify the circuit executing the corresponding computation. We show that this assumption can fail systematically in Mamba-2. Studying the state sink (disproportionate Delta-gate activation on boundary tokens, analogous to the attention sink), we find that single-bucket probes recover only a small execution layer while missing a much larger detection layer with the same representational signature. In Mamba-2, the state sink decomposes into two functional head sets. Single-bucket BOS-specialist heads (about 5% of heads at 2.7B) causally support both BOS-context and newline-target predictions across model scales and corpora. Dual heads (27-35% of heads, recovered by multi-class aggregation of the same probe) show stronger BOS-newline representational similarity but substantially weaker causal effects under ablation. Representational similarity does not imply functional equivalence. This distinction matters for downstream behaviour: ablating BOS-specialist heads collapses RULER NIAH retrieval accuracy from 1.00 to 0.00 at 1024 context length in both Mamba-1 2.8B and Mamba-2 2.7B, while size-matched complements preserve baseline performance. A random channel-bucketing control rules out substrate granularity alone, implicating Mamba-2's head-shared Delta projection. Probe-derived specialty can identify execution circuits; at coarse granularity the same probe also recovers detection circuits, and separating them requires class-conditional ablation rather than class-conditional cosine. --- ### Single-Channel Tissue Segmentation via Cross-Modal Distillation from Foundation Models Authors: Sakib Mohammad, Jarin Ritu, Md Sakhawat Hossain Published: 2026-05-30 Categories: cs.CV, cs.LG arXiv: https://arxiv.org/abs/2606.00928v1 Multiplexed fluorescence microscopy improves tissue segmentation by providing complementary channels including nuclear (DAPI) and membrane (E-cadherin), that together encode richer spatial context than single-channel imaging alone. However, multiplexed models require all channels at inference, limiting deployment where only a subset is available. This work proposes a cross-modal knowledge distillation framework that transfers semantic information from a frozen foundation model teacher processing multiplexed input to a lightweight student operating on the nuclear channel only. The distillation objective combines MSE-based probability matching, boundary-aware supervision, and learnable uncertainty weighting. SAM ViT-H and CellSAM are evaluated as teachers across four U-Net students: Swin-Tiny (27M), ResNet18 (11M), EfficientNet-B0 (5.3M), and MobileNetV3 (1.5M), on TissueNet and BBBC038. On TissueNet, the SAM-distilled Swin-Tiny student achieves Dice 78.36 (plus or minus 1.44), a 13.05-point improvement over the no-KD baseline (65.31 plus or minus 1.35) and 87.9% recovery of teacher oracle performance (89.12 plus or minus 1.21) at a 23x parameter reduction. KD consistently improves all four students by approximately 12 Dice points, confirming architecture-agnostic distillation. SAM ViT-H outperforms CellSAM as teacher across all settings. Cross-dataset evaluation on BBBC038 shows consistent gains without teacher retraining. --- ### Bridging Topology and Deep Representation Learning: A TDA-ViT Fusion Model for Four-Class Brain Tumor Classification Authors: Faisal Ahmed Published: 2026-05-30 Categories: cs.CV arXiv: https://arxiv.org/abs/2606.00927v1 Accurate brain tumor classification from magnetic resonance imaging (MRI) is a key requirement for early diagnosis and clinical decision-making. Vision Transformers (ViTs) have shown strong performance in medical image analysis by learning global contextual representations, but they often fail to capture intrinsic structural and topological patterns present in tumor regions. To address this limitation, we propose a fusion framework that combines Topological Data Analysis (TDA) features with pretrained Vision Transformer representations for four-class brain tumor classification. In the proposed method, TDA is used to extract complementary topological descriptors that capture geometric structure, connectivity, and shape information from MRI images. In parallel, a pretrained ViT model learns high-level semantic representations from the same images. These two feature spaces are then fused to form a unified and more discriminative representation for classification. The model is evaluated on the BRISC2025 dataset, which contains four brain tumor classes: glioma, meningioma, pituitary tumor, and non-tumor cases. Experimental results show that combining topological and transformer-based features significantly improves performance compared to using either approach alone. The proposed TDA-ViT fusion model achieves an accuracy of 99.10%, precision of 99.27%, recall of 99.15%, F1-score of 99.21%, and an AUC of 99.98%. It also outperforms several state-of-the-art models, including ResNet50, ResNet101, EfficientNetB2, and standalone Vision Transformers. These results demonstrate that topological features provide valuable complementary information that enhances deep representation learning, leading to a robust and highly accurate framework for automated brain tumor classification. --- ### Refit the Probe: Single-Direction Ablation Is Not a Necessity Test Authors: Yuhang Jiang Published: 2026-05-30 Categories: cs.LG, cs.CL arXiv: https://arxiv.org/abs/2606.00926v2 Probes are routinely paired with an intervention: ablate the direction the probe found, run the model, and read the change in task accuracy, taking a large drop as evidence that the computation depends on what the probe read and a near-zero drop as evidence that it does not. Either inference requires that the ablation have removed the target from the layer. We find that the ablation does not remove what it targets. A probe refitted on the ablated activations recovers its original accuracy in every cell we test, and keeps recovering when the probe's entire row space is deleted rather than a single axis, because the quantity survives in the orthogonal complement. Because a refitted probe recovers, neither a large task drop nor a near-zero one establishes whether the model needed the target, and one probe fit detects this. Replacing the ablation with iterative nullspace projection, scored against random subspaces of matched dimension, reverses the conclusion: representations that looked causally inert carry most of the task. The correction also separates where a variable is most readable from where deleting it does most damage, and those are not the same layer in any pretrained model we study. The erasure is defined by a linear probe family, so removing a nonlinearly encoded quantity remains open. --- ### Task Structure Reverses Layerwise State Encoding in Sequence Models Authors: Yuhang Jiang Published: 2026-05-30 Categories: cs.LG, cs.CL arXiv: https://arxiv.org/abs/2606.00926v1 Mechanistic studies of sequence models often treat layerwise state encodings as architectural traits: recurrent models concentrate readable state, attention-based models distribute it. We find that the same architecture reverses this profile when the task changes. Across Transformers, Mamba, Mamba-2, LSTMs, and GRUs, Parity is concentrated late in Mamba and the recurrent baselines and built gradually by Transformer; on bounded-depth Dyck-k the pattern flips. The same flip appears in fine-tuned Mamba-130M and Pythia-160M, and the Pythia Dyck bottleneck persists at 410M. Two explanations are conflated in the literature: algebraic structure (commutativity) versus computational structure (prefix update vs. stack). To separate them we add a third task: non-commutative S_3 permutation composition. S_3 groups with Parity, not Dyck, on layerwise probing across all five architectures and on Mamba-specific Conv1D attribution, so the grouping tracks computational structure rather than commutativity. Causal interventions show that, in the 4-layer formal models, linearly readable directions are often functionally necessary and can remain important at out-of-distribution lengths on Parity and Dyck. At pretrained scale the picture splits. Fine-tuned Pythia Dyck has a strong middle-layer bottleneck (L6-L7 ablation drops accuracy by roughly 81% at 160M; broader L4-L18 plateau at 410M), far weaker at the best-probe layer. Pretrained Mamba shows the complementary failure mode: its final layer is highly readable, no single probe direction breaks the task on Parity, Dyck, or S_3, yet mid-position activation patching there recovers about 97-98% of the clean-corrupted logit gap. Probing localizes where state is linearly available, not always where the computation is bottlenecked. Mechanistic signatures are properties of architecture and task together. --- ### Benchmarking Security Risk Detection and Verification in Open Agentic Skill Ecosystems Authors: Ismail Hossain, Sai Puppala, Zhuoran Lu, Sajedul Talukder, Nan Jiang Published: 2026-05-30 Categories: cs.CR, cs.AI arXiv: https://arxiv.org/abs/2606.00925v1 Open agent platforms allow community contributors to publish reusable skills that agents can invoke at runtime. This extensibility also creates a supply-chain risk: malicious contributors can hide harmful behavior inside skills that appear benign under superficial inspection. However, existing defenses are hard to evaluate because there is no benchmark that measures both malicious-skill detection and runtime verification. We present SkillVetBench, a two-stage security vetting benchmark for open agentic skill ecosystems. The first stage performs semantic vetting over each skill's natural-language specification to detect hidden malicious intent. The second stage executes flagged skills in an instrumented sandbox to observe runtime behavior and collect auditable evidence. We build a benchmark from confirmed malicious skills in the live OpenClaw ecosystem, including samples from the recent ClawHavoc supplychain campaign. Unlike static-only methods, SkillVetBench verifies detected threats with execution traces. Our experiments show that: (1) semantic-only and signature-based baselines are insufficient, missing up to 89\% of malicious skills whose threats arise from natural-language instructions, multicomponent logic, or cross-component interactions; (2) runtime attacks are concentrated in a small set of high-permission primitives, especially exec, write\_file, install\_skill, and spawn; and (3) SkillVetBench provides case studies in which sandbox execution directly supports malicious verdicts with concrete runtime evidence. --- ### Accuracy, Stability, and Repeated-Run Reliability of Large Language Models on Deterministic Programming Tasks Authors: Yongxi Zhou, Lai Yun Choi, Jiaxi Wen, Wenbo Ye Published: 2026-05-30 Categories: cs.LG, cs.AI, cs.SE arXiv: https://arxiv.org/abs/2606.00920v1 Run-level pass rate overstates retry-free coverage by up to 17.8 percentage points -- and the gap is largest precisely for mid-performing systems. We investigate this accuracy--stability relationship in large language model (LLM) evaluation for deterministic text-conditioned generation, using programming tasks as a concrete testbed. Standard code-generation benchmarks emphasize single-run accuracy or eventual success under repeated sampling, but many deployment settings also require stability: consistent outcomes across repeated invocations under the same task description. We present a repeated-run evaluation protocol with metrics for run-level accuracy, retry-free coverage, and per-problem variability. On a recency-based benchmark of 100 LeetCode-style problems, we evaluate 16 models from five provider families under two prompt templates with five repeated runs per problem, yielding 16,000 evaluation instances. Although run-level pass rate and perfect stability rate are strongly correlated (r=0.985), pass rate consistently exceeds retry-free coverage -- a gap that reaches 17.8 percentage points and reverses model rankings even among closely matched systems. Prompt effects are model-dependent rather than uniformly beneficial. These results suggest that repeated-run stability analysis is a necessary complement to conventional accuracy reporting for deterministic text-conditioned generation tasks. --- ### Towards Lightweight Reliability: Using Soft Prompts for Hallucination Mitigation in Large Language Models Authors: S M Tahmid Siddiqui, Akib Jawad Ononto, Anoop Singhal, Latifur Khan Published: 2026-05-30 Categories: cs.CL, cs.LG arXiv: https://arxiv.org/abs/2606.00919v2 Large language models (LLMs) have seen widespread adoption across various domains, yet their reliability is frequently undermined by hallucinations - responses that are plausible-sounding but factually incorrect. In high-stakes domains, these errors can reduce trust and introduce real-world risk. To address this challenge, we present a parameter-efficient approach that uses soft prompts to mitigate hallucinated content and promote responsible abstention in generative question-answering (QA) tasks. Our method, called Responsible Contrastive Soft Prompting (RCSP), uses a composite loss to train soft prompts that balance three goals: suppressing hallucinatory content, encouraging abstention under uncertainty, and preserving or improving factual recall. To achieve these goals, we incorporate contrastive loss, curriculum learning, and KL regularization into our training mechanism. We evaluate our approach on five diverse generative QA datasets using an LLM-as-a-Judge framework. Experimental results on the Gemma 3 (12B) and Llama 3.1 (8B) backbones demonstrate that RCSP effectively balances factual recall with hallucination suppression and abstention, yielding a generally superior F-score over standard reasoning and instruction-based prompting baselines. Notably, these improvements are achieved by training only a fraction of the parameters required by other tuning techniques. Our results demonstrate that soft prompts provide a modular and computationally efficient path toward improving LLM reliability. --- ### Towards Lightweight Reliability: Using Soft Prompts for Hallucination Mitigation in Large Language Models Authors: S M Tahmid Siddiqui, Akib Jawad Ononto, Anoop Singhal, Latifur Khan Published: 2026-05-30 Categories: cs.CL, cs.LG arXiv: https://arxiv.org/abs/2606.00919v1 Large language models (LLMs) have seen widespread adoption across various domains, yet their reliability is frequently undermined by hallucinations - responses that are plausible-sounding but factually incorrect. In high-stakes domains, these errors can reduce trust and introduce real-world risk. To address this challenge, we present a parameter-efficient approach that uses soft prompts to mitigate hallucinated content and promote responsible abstention in generative question-answering (QA) tasks. Our method, called Responsible Contrastive Soft Prompting (RCSP), uses a composite loss to train soft prompts that balance three goals: suppressing hallucinatory content, encouraging abstention under uncertainty, and preserving or improving factual recall. To achieve these goals, we incorporate contrastive loss, curriculum learning, and KL regularization into our training mechanism. We evaluate our approach on five diverse generative QA datasets using an LLM-as-a-Judge framework. Experimental results on the Gemma 3 (12B) and Llama 3.1 (8B) backbones demonstrate that RCSP effectively balances factual recall with hallucination suppression and abstention, yielding a generally superior F-score over standard reasoning and instruction-based prompting baselines. Notably, these improvements are achieved by training only a fraction of the parameters required by other tuning techniques. Our results demonstrate that soft prompts provide a modular and computationally efficient path toward improving LLM reliability. --- ### Adversarial Feeds Steer LLM Agent Decisions Against Their Defaults Authors: Rana Muhammad Usman Published: 2026-05-30 Categories: cs.AI, cs.CL, cs.CR arXiv: https://arxiv.org/abs/2606.00914v1 LLM agents increasingly act after consuming ranked external information streams such as social feeds, search results, retrieval contexts, and email queues, yet safety evaluations almost always test the model or the user prompt in isolation, never the upstream ranker that decides what the agent reads just before it acts. We introduce a controlled protocol that holds the model, persona, topic, and final decision prompt fixed and varies only the composition and ordering of the posts an agent encounters during a preceding ten-turn "scrolling" phase, isolating the causal effect of feed curation on a downstream decision. Across 2,785 decision rollouts on four modern open instruct LLMs from three independent labs, we identify three response regimes: adversarial capitulation, default saturation, and a default-direction asymmetry in which a one-sided feed tips a decision the model was genuinely uncertain about (in the clearest cases from 5% to 100%; Fisher p as low as 3 x 10^-10) but cannot dislodge one it already favors or holds firmly. The effect follows a dose-response curve, survives a generator swap that rules out a writing-style artifact, generalizes across several decision domains including security-relevant choices such as removing a deployment approval gate or relaxing access controls, and is partly mitigated by two simple feed-level defenses; a frontier model retains its default. We characterize the recommender as a practical, default-bounded control surface for LLM agents, and argue that agent evaluations must audit the feed layer rather than the final prompt alone. --- ### Bandit Simulation for Average Reward Inference Authors: Samya Praharaj, Chih-Yu Chang, Koulik Khamaru, Kelly W. Zhang Published: 2026-05-30 Categories: stat.ML, cs.LG arXiv: https://arxiv.org/abs/2606.00913v1 Multi-arm bandit algorithms are increasingly used in online platforms, clinical trials, and social science experiments, but valid statistical inference on their performance remains an open challenge. After deploying bandits, a natural question is whether one can construct a confidence interval for its mean reward and assess whether it reliably outperforms a baseline policy. The total reward achieved in any single bandit deployment is random, and deploying a bandit twice on the same population typically yields different reward trajectories due to stochastic rewards. Standard statistical inference methods cannot be used because bandit algorithms introduce complex dependencies in the collected data, which violate the i.i.d. assumption underlying many classical approaches. Moreover, existing inference methods for adaptively collected data only apply to estimands that do not depend on the data-collection algorithm (such as the mean reward under a fixed action). We propose Bandit Simulation for Inference (BSI), a framework that fits a simulator of the bandit environment from observed data--either on-policy or off-policy--and uses it to estimate the mean reward under any evaluation policy, including adaptive blackbox algorithms. BSI formally propagates uncertainty in the estimated simulator parameters into the confidence interval construction. Furthermore, for BSI to be valid, it requires only weak exploration assumptions on the behavior policy and avoids importance weighting. We prove that BSI yields asymptotically valid confidence intervals, and demonstrate empirically that it maintains nominal coverage in settings where standard off-policy evaluation methods fail. --- ### Reason, Retrieve, Re-rank: A Zero-Shot Reasoning-Aware Framework for Composed Video Retrieval Authors: Ali Alavi Published: 2026-05-30 Categories: cs.CV, cs.LG arXiv: https://arxiv.org/abs/2606.00910v1 Composed Video Retrieval (CoVR) seeks the target video that results from applying a free-form textual modification to a reference video. We address the \emph{Reason-Aware} CoVR (CoVR-R) challenge at the CVPR~2026 VidLLMs workshop, where retrieval is strictly zero-shot. We present \textbf{R3-CoVR} (\emph{Reason, Retrieve, Re-rank}), a training-free pipeline built entirely from frozen foundation models. A multimodal large language model (Qwen3-VL-8B) reasons about the \emph{after-effects} an edit implies -- state transitions, action phases, scene, camera and tempo -- and verbalises a concise post-edit description; a contrastive video--text encoder (SigLIP-2) embeds this description and the gallery for first-stage retrieval; finally a constraint-aware re-ranking stage uses the same multimodal model as a judge that scores each shortlisted candidate against the intended edited result. On the challenge test set, R3-CoVR attains \textbf{91.9\% R@1} and \textbf{98.2\% R@10}. Two findings drive these results: (i)~matching the description length to the contrastive encoder's text window lifts \Rk{1} from $67.5$ to $72.7$; and (ii)~the constraint-aware re-ranker, which reorders only the shortlist, lifts \Rk{1} from $72.7$ to $91.9$ -- the single largest gain. We analyse the re-ranker's behaviour, the retrieve/re-rank blend, and the shortlist depth, and we release a clean three-layer implementation. --- ### MLLM-Microscope: Unlocking Hidden Structure Within Multimodal Large Language Models Authors: Ravil Mussabayev, Rustam Mussabayev Published: 2026-05-30 Categories: cs.CL, cs.AI arXiv: https://arxiv.org/abs/2606.00909v1 This work presents MLLM-Microscope, a novel system designed for analyzing the hidden representations within Multimodal Large Language Models (MLLMs). Our system evaluates the linearity, intrinsic dimension, and anisotropy of multimodal token embeddings across transformer layers. Utilizing the ScienceQA dataset, we evaluate two state-of-the-art MLLMs, LLaVA-NeXT and OmniFusion. We find that both the main and residual streams for tokens of both modalities exhibit highly linear behaviors across transformer layers. However, LLaVA-NeXT's image tokens reveal a slight decline in linearity, whereas OmniFusion's remain consistent. Image token dimensions in OmniFusion remain consistently higher across layers compared to LLaVA-NeXT. Also, the OmniFusion's anisotropy is observed to stay consistently low throughout the layers. These findings suggest that the inner workings of MLLMs highly depend on the nature of modality fusion performed before passing the token sequence into LLM. This and other new potential insights obtainable from our system are surely capable of enhancing our understanding of the inner workings of MLLMs, informing future model design and optimization. --- ### hZACH-ViT: Curved Latent Geometry for Compact Vision Transformers in Low-Data Medical Imaging Authors: Athanasios Angelakis Published: 2026-05-30 Categories: cs.CV arXiv: https://arxiv.org/abs/2606.00906v1 Compact Vision Transformers are attractive for medical imaging in low-data and resource-constrained settings, but most existing variants assume that Euclidean latent geometry is sufficient for organizing image representations. We introduce hZACH-ViT, a family of curved-geometry extensions of ZACH-ViT, a compact zero-token Vision Transformer that removes positional embeddings and the class token and relies on global average pooling over patch representations. To isolate the role of geometry, we preserve the verified ZACH-ViT backbone and modify only the final representation space and prototype-based classifier head, enabling a controlled comparison between Euclidean, hyperbolic, and spherical latent geometries. We evaluate Poincaré, Klein, and spherical hZACH-ViT heads on seven MedMNIST datasets under an identical few-shot protocol with 50 samples per class and five random seeds. The completed benchmark contains 770 training runs spanning seven datasets, three non-Euclidean geometries, seven curvature magnitudes, and a Euclidean baseline. Across all seven datasets, the best non-Euclidean hZACH-ViT configuration improves over Euclidean ZACH-ViT, with an average gain of +0.021 in the dataset-specific primary metric and the largest improvement on OCTMNIST (+0.055 MacroF1). Fixed low-curvature configurations retain positive gains on the majority of datasets, and low curvature values (c = 0.1 or 0.2) account for six of the seven dataset-level winners. Rather than identifying a universally optimal manifold, our results establish geometry and curvature as dataset-dependent model-selection variables, with fixed low-curvature analyses confirming that gains persist beyond exhaustive per-dataset tuning. --- ### Ryze: Evidence-Enriched Data Synthesis from Biomedical Papers Authors: Yeqi Huang, Yue Chen, Yanwei Ye, Guanhao Su, Luo Mai Published: 2026-05-30 Categories: cs.AI arXiv: https://arxiv.org/abs/2606.00902v1 General-purpose VLMs remain unreliable for biomedical research because valid answers in scientific papers depend on evidence split across figures, tables, charts, captions, and referring text. Existing post-training pipelines are bottlenecked by costly expert annotation and by synthetic data that drops this evidence structure. We present Ryze, a fully automated system that converts raw biomedical papers into an evidence-enriched training set and a domain-specialized VLM. Ryze synthesizes QA pairs with complete supporting evidence (visual element, caption, extracted structure, and referring paragraphs), reduces layout and OCR errors via chart/table-aware extraction and LLM-based cleansing, and applies a progress-gated post-training strategy combining supervised fine-tuning with reinforcement learning. Starting from Qwen3-VL-8B, Ryze produces BioVLM-8B at under USD 200, achieving 48.0% weighted accuracy on LAB-Bench, outperforming the base model by +12.6 percentage points (pp) and surpassing GPT-5.2 by +3.8 pp. We release Ryze as open source together with the trained BioVLM-8B model. --- ### Citation Grounding Measures the Oracle: Graph Coverage Determines Reported LLM Hallucination Rates in Law Authors: Volodymyr Ovcharov Published: 2026-05-30 Categories: cs.CL, cs.DL arXiv: https://arxiv.org/abs/2606.00898v2 Verifying LLM-generated legal citations against a graph of citations extracted from real court decisions is an appealing way to measure hallucination at scale: no annotators, no reference answers. We show that what such a metric reports is governed by the coverage of the graph it queries rather than by the model it evaluates, and that at the coverage where its verdicts become trustworthy it stops distinguishing models at all. We score 400 responses (100 Ukrainian legal queries x four commercial LLMs) against two snapshots of the same national citation graph, holding responses, extractor and metric fixed. Against a sparse snapshot (4.7e5 records) citation grounding ranges 0.791-0.855, apparently showing 15-21% of citations hallucinated. Against a dense snapshot of the same registry re-derived ten weeks later (3.3e8 records, 5.8e7 decisions) the identical responses score 0.989-0.999. Subsampling shows the cause is coverage: harvesting modelled as uniform record sampling, calibrated on nothing but the record count, reproduces the sparse scores to within 0.018 while knowing nothing about the models. Bootstrapping over the queries shows the other half, which no version of this work reported: no pair of systems is separable at 95% at any oracle size tested. The metric is caught between two failures. Sparse oracles discriminate, but what they discriminate is harvesting coverage; dense oracles are trustworthy and separate nothing. Densifying the graph, the obvious remedy for the first, produces the second. An independent legislation registry adjudicates: all 54 citations flagged by the sparse oracle name real articles, a 100% false-positive rate; of the four flagged by the dense oracle, two are fabrications, one a coverage gap, and one we cannot classify - it looks like a repealed provision, which is the argument in miniature. --- ### Citation Grounding: Detecting and Reducing LLM Citation Hallucinations via Legal Citation Graphs Authors: Volodymyr Ovcharov Published: 2026-05-30 Categories: cs.CL, cs.DL arXiv: https://arxiv.org/abs/2606.00898v1 Large language models systematically hallucinate legal citations -- fabricating statute references, citing repealed provisions, and confusing jurisdictions -- yet no automated method exists to measure or reduce this behavior at scale. We propose citation grounding (CG), a metric that verifies LLM-generated legal citations against a ground-truth citation graph extracted from 100.8 million Ukrainian court decisions (502 million edges, 21,736 unique statute nodes). CG decomposes into three components -- citation precision (does the cited provision exist?), citation relevance (is it contextually appropriate?), and citation temporality (was it valid at the relevant date?) -- enabling differential diagnosis of hallucination types. Empirical evaluation on 100 Ukrainian legal queries across five systems -- four commercial LLMs via AWS Bedrock (Claude Haiku 4.5, Mistral Pixtral Large, Amazon Nova Pro/Lite) and one RAG-augmented production system -- reveals CG ranging from 0.791 to 0.873, with 13-21% of citations hallucinated. To reduce hallucinations without human annotation, we introduce Citation Grounding DPO (CG-DPO): a method that constructs preference pairs algorithmically by corrupting verified citations from real court decisions via four targeted strategies. On a dataset of 2,244 court decisions, a Qwen2.5-7B-Instruct model fine-tuned with LoRA achieves 98.5% mean validation accuracy in distinguishing correct from corrupted citations (rewards margin +14.9, std < 0.3 pp across 3 seeds). The citation graph, evaluation framework, and CG-DPO dataset are released as open resources. ---