# Research Papers — 2026-05 · bundle page 200 of 326 > Full-text markdown of papers #9951–10000 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/200 [← Previous bundle](https://aipod.dev/papers/archive/2026-05/page/199.md) [Next bundle →](https://aipod.dev/papers/archive/2026-05/page/201.md) ### Revisiting DAgger in the Era of LLM-Agents Authors: Changhao Li, Rushi Qiang, Jiawei Huang, Chenxiao Gao, Chao Zhang, Niao He, Bo Dai Published: 2026-05-13 Categories: cs.LG arXiv: https://arxiv.org/abs/2605.12913v1 Long-horizon LM agents learn from multi-turn interaction, where a single early mistake can alter the subsequent state distribution and derail the whole trajectory. Existing recipes fall short in complementary ways: supervised fine-tuning provides dense teacher supervision but suffers from covariate shift because it is trained on off-policy teacher trajectories; while reinforcement learning with verifiable rewards avoids this off-policy mismatch by learning from on-policy rollouts but with only sparse outcome feedback. We address this dilemma by revisiting Dataset Aggregation (DAgger) for multi-turn LM agents: the algorithm collects trajectories through a turn-level interpolation of student and teacher policies, and the student is then trained on these trajectories using supervised labels provided by the teacher. By directly interacting with environments, we expose the model to realistic states likely to be encountered during deployment, thereby effectively mitigating covariate shift. Besides, since the student is learned by mimicking the teacher's behavior, it receives rich feedback during learning. To demonstrate DAgger enjoys the benefits of both worlds, we tested the algorithm to train a software-engineering agent with 4B- and 8B-scale student models. On SWE-bench Verified, our DAgger-style training improves over the strongest post-training baseline by +3.9 points at 4B and +3.6 points at 8B. The resulting 4B agent reaches 27.3%, outperforming representative published 8B SWE-agent systems, while the 8B agent achieves 29.8%, surpassing SWE-Gym-32B and coming within 5 points of stronger 32B-scale agents. Together with consistent gains on the held-out SWE-Gym split, these results suggest the effectiveness of DAgger for modern long-horizon LM agents. --- ### The Mechanism of Weak-to-Strong Generalization: Feature Elicitation from Latent Knowledge Authors: Ryoya Awano, Taiji Suzuki Published: 2026-05-13 Categories: stat.ML, cs.LG arXiv: https://arxiv.org/abs/2605.12908v1 Weak-to-strong (W2S) generalization, in which a strong model is fine-tuned on outputs of a weaker, task-specialized model, has been proposed as an approach to aligning superhuman AI systems. Existing theoretical analyses either fix the student's representations or operate in restricted settings. Whether multi-step SGD can succeed in feature learning while preserving diverse pre-trained capabilities remains open. We study W2S in the setting of reward-model learning with two-layer neural networks. The strong model has pre-trained representations organized into low-dimensional subspaces $V_k$, and is fine-tuned under the supervision of a weak model specialized on task $κ$. We prove that the strong model efficiently learns task $κ$, eliciting its pre-trained knowledge while retaining general capabilities. This establishes W2S generalization in the feature-learning regime, in the sense that the strong model acquires the target feature direction through W2S training, rather than having it given a priori. Moreover, W2S preserves pre-trained off-target features, whereas standard supervised fine-tuning causes catastrophic forgetting when off-target feature directions are correlated with the target's. Numerical experiments on synthetic data confirm our theoretical results. --- ### Data Difficulty and the Generalization--Extrapolation Tradeoff in LLM Fine-Tuning Authors: Siyuan Liu, Tinghong Chen, Xinghan Li, Yifei Wang, Jingzhao Zhang Published: 2026-05-13 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2605.12906v2 Data selection during supervised fine-tuning (SFT) can critically change the behavior of large language models (LLMs). Although existing work has studied the effect of selecting data based on heuristics such as perplexity, difficulty, or length, the reported findings are often inconsistent or context-dependent. In this work, we systematically study the role of data difficulty in fine-tuning from both empirical and theoretical perspectives, and find that there is no universally optimal difficulty level; rather, its effectiveness depends on the dataset size. We show that for a fixed data budget, there exists an optimal data difficulty for SFT, and that this optimal difficulty shifts toward harder data as the data budget increases. To explain this phenomenon, we conduct controlled synthetic experiments that reveal a simple underlying mechanism: the interplay between the (in-distribution) generalization gap and the extrapolation gap. We further support this mechanism through a theoretical analysis using PAC-Bayesian generalization bounds. Overall, our results clarify how data size and difficulty jointly affect the trade-off between generalization and extrapolation in SFT, providing guidance for difficulty-based data selection under certain model and data conditions. --- ### VIP-COP: Context Optimization for Tabular Foundation Models Authors: Yilong Chen, Xueying Ding, Leman Akoglu Published: 2026-05-13 Categories: cs.LG arXiv: https://arxiv.org/abs/2605.12904v1 Tabular foundation models (TFMs) have emerged as a powerful paradigm for in-context learning on structured data, enabling direct prediction on new tabular tasks without task-specific training. However, their effectiveness is constrained by context length limits, restricting application to medium-scale data and degrading performance when inference-time data exceed pretraining size distributions. Our work introduces VIP-COP, estimating the Value of Importance for Prediction of training examples and features for hard Context OPtimization for TFMs. Its explicit selection mechanism suppresses noise and isolates influential data, enabling the model to also benefit from data augmentation by prioritizing high-value augmented samples and features. VIP-COP is (i) fast, boosting performance often within minutes of optimization, based on an online KernelSHAP-based regression with iterative refinement, value-guided context sampling, and multi-fidelity pruning; (ii) budget-aware and any-time, improving with additional test-time compute unlike heuristics that produce fixed contexts; (iii) model-aware yet fully black-box, requiring no access to model internals, making it compatible with both proprietary and open-source TFMs; (iv) interpretable, identifying discrete ``Very Important Predictors'' (samples and features) that maximize signal-to-noise, which makes it (v) robust, isolating high-value data from noise. In contrast, soft-prompt optimization requires model gradients, produces abstract latent tokens, and lacks explicit signal discrimination. Extensive experiments show that VIP-COP consistently outperforms heuristic and optimized baselines across large-scale high-dimensional testbeds, including data augmentation and data-noise settings, establishing a new state of the art in test-time context refinement for TFMs. --- ### Robust Sequential Experimental Design for A/B Testing Authors: Qianglin Wen, Xiangkun Wu, Chengchun Shi, Ting Li, Niansheng Tang, Yingying Zhang, Hongtu Zhu Published: 2026-05-13 Categories: stat.ML, cs.LG arXiv: https://arxiv.org/abs/2605.12899v1 Experimental design has emerged as a powerful approach for improving the sample efficiency of A/B testing, yet existing designs rely critically on correctly specified models. We study robust sequential experimental design under model misspecification and develop a unified framework that covers both contextual bandit and dynamic settings. Theoretically, we prove that our design bounds the worst-case mean squared error of the estimated treatment effect. Empirically, we demonstrate the effectiveness of the proposed approach using synthetic and real-world datasets from a leading technology company. --- ### When Do LLMs Generate Realistic Social Networks? A Multi-Dimensional Study of Culture, Language, Scale, and Method Authors: Sai Hemanth Kilaru, Sriram Theerdh Manikyala, Raghav Upadhyay, Sri Sai Kumar Ramavath, Srivika Nunavathu, Dalal Alharthi Published: 2026-05-13 Categories: cs.SI, cs.CL, cs.CY arXiv: https://arxiv.org/abs/2605.12898v1 Large language models (LLMs) are increasingly used as substitutes for human subjects in behavioral simulations, including synthetic social network generation. Yet it remains unclear how their relational outputs depend on prompt design, cultural framing, prompt language, and model scale. Building on homophily theory and structural balance theory, we formalize four LLM-based tie-formation mechanisms: sequential, global, local, and iterative, and treat them as distinct conditional distributions over edge sets. Using a fixed roster of 50 demographically grounded personas, we generate 192 verified directed networks across four cultural contexts, four prompt languages, three GPT-4.1 variants, and four prompting architectures, with two seeds per condition. We find that cultural framing shifts inbreeding homophily and largest-component connectivity. Political affiliation dominates tie formation under three methods, while the global method substitutes age, showing that prompt architecture functions as a substantive sociological variable. Model scale produces a stable divergence ranking, with the smallest variant behaving qualitatively differently rather than merely noisily. Prompt language alone sharply shifts religion homophily, especially under Hindi prompting, while leaving political homophily nearly invariant. LLM-generated networks match real social graphs on clustering and modularity better than standard graph baselines, yet encode demographic biases above empirical levels. These results show that prompt choices often treated as implementation details encode substantive sociological assumptions. --- ### RISED: A Pre-Deployment Evaluation Framework for High-Stakes AI Decision-Support Systems, with Application to Healthcare Authors: Rohith Reddy Bellibatlu, Manpreet Singh, Yash Jajoo, Shyamal Lakhanpal, Abhishek Israni Published: 2026-05-13 Categories: cs.LG, cs.AI, cs.CY, stat.AP arXiv: https://arxiv.org/abs/2605.12895v2 Clinical decision-support systems are expert systems whose recommendations clinicians act on directly, yet they are usually cleared on one aggregate accuracy number from a held-out test set. That number says nothing about input reliability under encoding shifts, subgroup gaps, threshold sensitivity, or operational feasibility. We present RISED, a pre-deployment evaluation framework operationalising five dimensions (Reliability, Inclusivity, Sensitivity, Equity, Deployability) through BCa bootstrap 95% confidence intervals, literature-grounded thresholds, and Holm-Bonferroni-corrected PASS / FAIL / INCONCLUSIVE verdicts; Equity is a proxy-dependence diagnostic rather than a gating test. Applied to seven cohorts spanning 35 years (n from 303 to 99,492), RISED surfaces failures invisible to AUROC: on Diabetes 130, Reliability passes by three orders of magnitude (PSS = 0.0004) while Inclusivity (AUC parity gap = 0.262) and Sensitivity (max threshold-flip rate 49.1%) fail decisively; both NHIS cohorts reproduce this. NHANES 2021-2023, with a complete feature profile, achieves INCONCLUSIVE verdicts; BRFSS 2024 produces the suite's most severe Sensitivity failure (max threshold-flip rate 64.2%) after instrument rotation removed hypertension and cholesterol. The pattern recurs on credit- and income-prediction cohorts, confirming domain-agnosticity; a multi-model check shows the failures are data-driven, not model-specific. RISED ships as an open-source Python package complementing TRIPOD+AI, FUTURE-AI, and Fairlearn with the structured numerical evidence those standards require but do not prescribe. --- ### Beyond Cooperative Simulators: Generating Realistic User Personas for Robust Evaluation of LLM Agents Authors: Harshita Chopra, Kshitish Ghate, Aylin Caliskan, Tadayoshi Kohno, Chirag Shah, Natasha Jaques Published: 2026-05-13 Categories: cs.AI, cs.CL arXiv: https://arxiv.org/abs/2605.12894v1 Large Language Model (LLM) agents are increasingly deployed in settings where they interact with a wide variety of people, including users who are unclear, impatient, or reluctant to share information. However, collecting real interaction data at scale remains expensive. The field has turned to LLM-based user simulators as stand-ins, but these simulators inherit the behavior of their underlying models: cooperative and homogeneous. As a result, agents that appear strong in simulation often fail under the unseen, diverse communication patterns of real users. To narrow this gap, we introduce Persona Policies (PPol), a plug-and-play control layer that induces realistic behavioral variation in user simulators while preserving the original task goals. Rather than hand-crafting personas, we cast persona generation as an LLM-driven evolutionary program search that optimizes a Python generator to discover behaviors and translate them into task-preserving roleplay policies. Candidate generators are guided by a multi-objective fitness score combining human-likeness with broad coverage of human behavioral patterns. Once optimized, the generator produces a diverse population of human-like personas for any task in the domain. Across tau^2-bench retail and airline domains, evolved PPol programs yield 33-62% absolute gains in fitness score over the baseline simulator. In a blinded evaluation, annotators rated PPol-conditioned users as human 80.4% of the time, close to real human traces and nearly twice as frequently as baseline simulators. Agents trained with PPol are more robust to challenging, out-of-distribution behaviors, improving task success by +17% relative to training only on existing simulated interactions. This offers a novel approach to strengthen simulator-based evaluation and training without changing tasks or rewards. --- ### Steer-to-Detect: Probing Hidden Representations for Detection of LLM-Generated Texts Authors: Luxu Liang, Xiang Li Published: 2026-05-13 Categories: stat.AP, cs.LG arXiv: https://arxiv.org/abs/2605.12890v1 The rapid advancement of large language models (LLMs) has made machine-generated text increasingly difficult to distinguish from human-written text. While recent studies explore leveraging internal representations of language models to uncover deeper detection signals, these raw features often exhibit substantial overlap between classes, limiting their discriminative power. To address this challenge, we propose Steer-to-Detect (\texttt{S2D}), a two-stage framework for detecting LLM-generated text. In the first stage, \texttt{S2D} learns a steering vector that is injected into the hidden states of a frozen observer LLM, producing representations with improved class separability. In the second stage, detection is performed via a hypothesis testing procedure based on the steered representations. We establish finite-sample, high-probability guarantees for Type I and Type II errors, providing a theoretical characterization of the procedure. Empirically, \texttt{S2D} achieves strong and consistent performance across a range of settings, including out-of-distribution scenarios and adversarial perturbations. --- ### EcoGEO: Trajectory-Aware Evidence Ecosystems for Web-Enabled LLM Search Agents Authors: Hengwei Ye, Jiasheng Mao, Zhenhan Guan, Zheng Tian Published: 2026-05-13 Categories: cs.IR, cs.AI arXiv: https://arxiv.org/abs/2605.12887v2 Web-enabled LLM agents are changing how online information influences search outcomes. Existing Generative Engine Optimization (GEO) studies mainly focus on individual webpages. However, agentic web search is not a single-document setting: an agent may issue queries, crawl pages, follow links, reformulate searches, and synthesize evidence across multiple browsing steps. Influence therefore depends not only on page content, but also on how pages are organized, connected, and encountered along the agent's browsing trajectory. We study this shift through Ecosystem Generative Engine Optimization (EcoGEO), which treats GEO as an environment-level influence problem for web-enabled LLM agents. To instantiate this perspective, we propose TRACE, a Trajectory-Aware Coordinated Evidence Ecosystem. Given a recommendation query and a fictional target product, our method builds a controlled evidence environment that coordinates an agent-facing navigation entry page with heterogeneous support pages. These pages use shared terminology, internal links, and consistent product attributes to introduce, verify, and reinforce the target product. We evaluate our method on OPR-Bench, a benchmark for open-ended product recommendation. Experiments show that it consistently outperforms page-level GEO baselines in final target recommendation. Trajectory-level metrics further show increased initial target-result crawls, target-specific follow-up searches, and internal-link crawls, suggesting that the gains come from shaping the agent's evidence-acquisition process rather than merely adding more target-related content. Overall, our findings support an ecosystem research paradigm for GEO, where web-enabled LLM agents are studied in relation to the broader evidence environments that guide search, browsing, and answer synthesis. --- ### CiteVQA: Benchmarking Evidence Attribution for Trustworthy Document Intelligence Authors: Dongsheng Ma, Jiayu Li, Zhengren Wang, Yijie Wang, Jiahao Kong, Weijun Zeng, Jutao Xiao, Jie Yang, Wentao Zhang, Bin Wang, Conghui He Published: 2026-05-13 Categories: cs.CL, cs.CV arXiv: https://arxiv.org/abs/2605.12882v1 Multimodal Large Language Models (MLLMs) have significantly advanced document understanding, yet current Doc-VQA evaluations score only the final answer and leave the supporting evidence unchecked. This answer-only approach masks a critical failure mode: a model can land on the correct answer while grounding it in the wrong passage -- a critical risk in high-stakes domains like law, finance, and medicine, where every conclusion must be traceable to a specific source region. To address this, we introduce CiteVQA, a benchmark that requires models to return element-level bounding-box citations alongside each answer, evaluating both jointly. CiteVQA comprises 1,897 questions across 711 PDFs spanning seven domains and two languages, averaging 40.6 pages per document. To ensure fidelity and scalability, the ground-truth citations are generated by an automated pipeline-which identifies crucial evidence via masking ablation-and are subsequently validated through expert review. At the core of our evaluation is Strict Attributed Accuracy (SAA), which credits a prediction only when the answer and the cited region are both correct. Auditing 20 MLLMs reveals a pervasive Attribution Hallucination: models frequently produce the right answer while citing the wrong region. The strongest system (Gemini-3.1-Pro-Preview) achieves an SAA of only 76.0, and the strongest open-source MLLM reaches just 22.5. Ultimately, towards trustworthy document intelligence, CiteVQA exposes a reliability gap that answer-only evaluations overlook, providing the instrumentation needed to close it. Our repository is available at https://github.com/opendatalab/CiteVQA. --- ### ASAP: Amortized Doubly-Stochastic Attention via Sliced Dual Projection Authors: Huy Tran, Max Milkert, David Hyde Published: 2026-05-13 Categories: cs.LG arXiv: https://arxiv.org/abs/2605.12879v1 Doubly-stochastic attention has emerged as a transport-based alternative to row-softmax attention, with recent Transformer variants using it to reduce attention sinks and rank collapse while improving performance. In this family, the standard approach is Sinkhorn scaling, which trains more efficiently but still repeats matrix scaling in every inference forward pass. Sliced-transport attention removes the online iteration, but its soft sorting approximation materializes dense tensors for each slice, requiring substantially more training resources than Sinkhorn attention. We introduce ASAP: Amortized Doubly-Stochastic Attention via Sliced Dual Projection, a train-then-compile method that trains the doubly-stochastic layer with Sinkhorn, then replaces the iterative scaling loop at inference with a fixed sliced-dual operator. It learns a lightweight parametric map from exact one-dimensional Kantorovich potentials to the Sinkhorn query-side dual, then reconstructs the attention plan with a two-sided entropic c-transform. Across language and vision benchmarks, ASAP keeps the cheaper training setup and remains highly competitive with recent baselines. In the main frozen-layer benchmark, ASAP is 5.3 faster than the trained Sinkhorn teacher while matching its accuracy; in downstream replacements, ASAP recovers most of the teacher performance without any retraining. --- ### Adam-SHANG: A Convergent Adam-Type Method for Stochastic Smooth Convex Optimization Authors: Yaxin Yu, Long Chen, Minfu Feng Published: 2026-05-13 Categories: math.OC, cs.LG arXiv: https://arxiv.org/abs/2605.12878v1 We propose Adam-SHANG, a Lyapunov-guided Adam-type method that couples momentum, adaptive preconditioning, and a curvature-aware correction through a more stable lagged-preconditioner update. For stochastic smooth convex optimization, we prove convergence in expectation under an admissible stepsize condition that can always be satisfied by a conservative spectral bound, without imposing global monotonicity on the second-moment sequence. To obtain a less conservative practical rule, we introduce a computable trace-ratio stepsize, motivated by a local coordinatewise alignment condition. The same structural update is also tested beyond the convex setting with simplified parameters. Experiments validate the predicted stochastic decay and show competitive training performance against Adam and AdamW on deep learning tasks. --- ### Certified Robustness under Heterogeneous Perturbations via Hybrid Randomized Smoothing Authors: Blaise Delattre, Hengyu Wu, Paul Caillon, Wei Yang Bryan Lim, Yang Cao Published: 2026-05-13 Categories: cs.LG arXiv: https://arxiv.org/abs/2605.12876v1 Randomized smoothing provides strong, model-agnostic robustness certificates, but existing guarantees are limited to single modalities, treating continuous and discrete inputs in isolation. This limitation becomes critical in multimodal models, where decisions depend on cross-modal semantics and adversaries can jointly perturb heterogeneous inputs, rendering unimodal certificates insufficient. We introduce a unified randomized smoothing framework for mixed discrete--continuous inputs based on an analytically tractable Neyman--Pearson formulation of the joint worst-case problem. By analyzing the joint likelihood ordering induced by factorized discrete and continuous noise, our approach yields a closed-form, one-dimensional certificate that strictly generalizes both Gaussian (image-only) and discrete (text-only) randomized smoothing. We validate the framework on multimodal safety filtering, providing, to our knowledge, the first model-agnostic Neyman--Pearson certificate for joint discrete-token and continuous-image perturbations in interaction-dependent text--image safety filtering. --- ### Descriptive Collision in Sparse Autoencoder Auto-Interpretability: When One Explanation Describes Many Features Authors: Jordan F. McCann Published: 2026-05-13 Categories: cs.LG arXiv: https://arxiv.org/abs/2605.12874v1 Sparse autoencoders (SAEs) are now standard tools for decomposing language model activations into interpretable features, and automated interpretability pipelines routinely assign each feature a short natural-language explanation. Existing critiques of this practice focus on polysemanticity -- one feature with many meanings -- or on whether explanations predict activations. We identify a complementary, structurally distinct problem we call descriptive collision: many distinct SAE features admit the same explanation. Reanalyzing the largest publicly-available dataset of human-annotated SAE features (Marks et al., 2025), comprising 722 annotated features across Gemma 2 2B and Pythia 70M, we find that the mean annotation string is reused across 3.07 features; 82.1% of features share their annotation with at least one other feature; and the single most common annotation string ("plural nouns") labels 101 distinct features spanning 18 layers and four model components. Information-theoretically, the average annotation resolves only 70% of feature identity. We formalize a property called discrimination, prove that current detection-style auto-interpretability scoring is invariant to collision, and propose two complementary corrective metrics -- collision-adjusted detection and discrimination scoring -- that explicitly penalize explanations that fail to distinguish a feature from its neighbors. The collision problem is independent of, and additive with, previously identified failure modes of auto-interpretability; ignoring it inflates reported feature interpretability by a quantity equal to roughly one-third of the bits required to identify a feature. --- ### SMA: Submodular Modality Aligner For Data Efficient Multimodal Learning Authors: Truong Pham, Anay Majee, Rishabh Iyer Published: 2026-05-13 Categories: cs.LG arXiv: https://arxiv.org/abs/2605.12872v1 Despite the recent success of Multimodal Foundation Models (FMs), their reliance on massive paired datasets limits their applicability in low-data and rare-scenario settings where aligned data is scarce and expensive. A key bottleneck is the adoption of an instance-level formulation, which learns alignment by maximizing correlation between individual image-text pairs while neglecting the underlying geometric structure across modalities resulting in a modality gap across input modalities. In this paper, we propose a combinatorial paradigm for multimodal alignment that moves beyond pairwise learning and introduce the \emph{Submodular Modality Aligner (SMA)}, which treats multiple augmentations and descriptions of an entity as a set, leveraging multiple descriptions of the data to capture richer cross-modal structure. We instantiate SMA using a principled objective based on Submodular Mutual Information (SMI), which jointly maximizes inter-modality mutual information while reducing cross-modal divergence. This formulation enables the model to effectively utilize multiple positive associations and extract significantly more information from limited data. We evaluate SMA on 14 zero-shot classification and retrieval tasks from the CLIP benchmark and demonstrate consistent gains in the low-data regime. Notably, SMA achieves strong multimodal generalization using only tens of thousands of samples. This is orders of magnitude fewer than standard approaches. Our results highlight the importance of set-based formulations and submodular objectives for data-efficient multimodal learning. --- ### Quantifying LLM Safety Degradation Under Repeated Attacks Using Survival Analysis Authors: Zvi Topol Published: 2026-05-13 Categories: cs.CR, cs.AI arXiv: https://arxiv.org/abs/2605.12869v1 Large language models (LLMs) are increasingly deployed in a wide range of applications, yet remain vulnerable to adversarial jailbreak attacks that circumvent their safety guardrails. Existing evaluation frameworks typically report binary success/failure metrics, failing to capture the temporal dynamics of how attacks succeed under persistent adversarial pressure. This preliminary work proposes a novel evaluation framework that applies survival analysis techniques to characterize LLM jailbreak vuln`erability. Our approach models the time-to-jailbreak as a survival outcome, enabling estimation of hazard functions, survival curves, and risk factors associated with successful attacks. We evaluate three LLMs against a subset of prompts from the HarmBench dataset spanning three attack categories. Our analysis reveals that models exhibit distinct vulnerability profiles: while one model demonstrates rapid degradation under iterative attacks, the two other models show consistent moderate vulnerability. Our framework provides actionable insights for model and LLM application developers and establishes survival analysis as a rigorous methodology for LLM safety evaluation. --- ### Language-Based Agent Control Authors: Timothy Zhou, Loris D'Antoni, Nadia Polikarpova Published: 2026-05-13 Categories: cs.PL, cs.AI, cs.CR arXiv: https://arxiv.org/abs/2605.12863v1 This paper introduces language-based agent control (LBAC), a new programming model for agentic applications that brings techniques from programming languages and language-based security to the problem of agent control. In conventional programming, combinations of static typing and runtime enforcement have long been used to guarantee that well-typed programs satisfy user-specified policies, including policies for access control, information flow, data provenance, and more. The key idea behind LBAC is to extend these guarantees to agentic applications by requiring agents to generate programs that are themselves well typed in the context of the surrounding scaffolding code. Unsafe programs are rejected by the type-checker before execution, allowing policies to apply uniformly across the entire application, including both agent-generated behavior and developer-written scaffolding. At the same time, LBAC preserves substantial expressiveness: agents may perform arbitrary side-effect-free computation and recursively invoke subagents, which retain full tool access subject to the same -- or potentially more restrictive -- policies. We demonstrate LBAC with three case studies: I/O sandboxing via filesystem capabilities, data provenance, and information-flow control. --- ### NeuroRisk: Physics-Informed Neural Optimization for Risk-Aware Traffic Engineering Authors: Yingming Mao, Ximeng Liu, Jingyi Cheng, Xiyuan Liu, Jiashuai Liu, Yike Liu, Zhen Yao, Yuzhou Zhou, Siyuan Feng, Qiaozhu Zhai, Shizhen Zhao Published: 2026-05-13 Categories: cs.NI, cs.LG arXiv: https://arxiv.org/abs/2605.12862v1 In production Wide-Area Networks (WANs), correlated failures dominate availability losses, forcing operators to reserve large safety margins that leave substantial capacity underutilized. Achieving high utilization under strict availability targets therefore requires risk-aware Traffic Engineering (TE) over dozens to hundreds of probabilistic failure scenarios-yet solving this problem at operational timescales remains elusive. We demonstrate that existing risk-aware formulations can be unified under an embedded Sort-and-Select structure, exposing a fundamental trade-off between expressiveness and tractability: classical optimizers either restrict scenario selection for efficiency or incur prohibitive decomposition costs. While deep learning appears promising, prior Deep TE methods mainly target maximum link utilization and rely on scaling-based feasibility, which fundamentally breaks under explicit capacity constraints and scenario-dependent risk. We present NeuroRisk, a physics-informed deep unrolled optimizer that exploits the structure of Sort-and-Select. NeuroRisk enforces feasibility via gated edge-local reservations and represents scenario sets through permutation-invariant, gradient-aligned cues. Evaluations on production-style WANs show that NeuroRisk achieves small optimality gaps relative to the solver with orders of magnitude speedup $(10^2- 10^5 \times)$ on risk objectives, while outperforming neural baselines on nominal throughput. --- ### ChipMATE: Multi-Agent Training via Reinforcement Learning for Enhanced RTL Generation Authors: Zhongkai Yu, Yichen Lin, Chenyang Zhou, Yuwei Zhang, Kun Zhou, Junxia Cui, Haotian Ye, Zhengding Hu, Zaifeng Pan, Ruiyi Wang, Yujie Zhao, Hejia Zhang et al. Published: 2026-05-13 Categories: cs.MA, cs.AI, cs.AR, cs.LG arXiv: https://arxiv.org/abs/2605.12857v1 Existing API-based agentic systems for RTL code generation are fundamentally misaligned with industrial practice: they assume a golden testbench is available at generation time, rely on closed-source APIs incompatible with chip vendors' air-gapped security requirements, and cannot be trained on vendors' proprietary RTL codebases, leaving valuable internal data unused. Recent self-trained models address the deployment constraint but remain single-turn generators that overlook the critical role of verification in real industrial flows. To bridge these gaps, we present ChipMATE, the first self-trained multi-agent framework for RTL generation. Inspired by industrial practice where correctness emerges from cross-comparison between independently written RTL modules and reference models, ChipMATE pairs a Verilog agent with a Python reference-model agent that mutually verify each other's outputs without any golden oracle. We design a backtrack-based inference workflow to prevent error propagation across turns, and a two-stage training pipeline that first trains each agent individually to saturate its code-generation capability, then trains the team jointly to collaborate effectively. To support the training, we further build a hybrid data-generation framework that produces 64.4K high-quality reference model training samples. ChipMATE achieves 75.0\% and 80.1\% pass@1 on VerilogEval V2 with 4B and 9B base models, outperforming all existing self-trained models and even DeepSeek V4 with 1600B parameters. Our code and model weights are publicly available in https://github.com/zhongkaiyu/ChipMATE. --- ### Moltbook Moderation: Uncovering Hidden Intent Through Multi-Turn Dialogue Authors: Ali Al-Lawati, Nafis Tripto, Abolfazl Ansari, Jason Lucas, Suhang Wang, Dongwon Lee Published: 2026-05-13 Categories: cs.AI, cs.SI arXiv: https://arxiv.org/abs/2605.12856v2 The emergence of multi-agent systems introduces novel moderation challenges that extend beyond content filtering. Agents with malicious intent may contribute harmful content that appears benign to evade content-based moderation, while compromising the system through exploitative and malicious behavior manifested across their overall interaction patterns within the community. To address this, we introduce BOT-MOD (BOT-MODeration), a moderation framework that grounds detection in agent intent rather than traditional content level signals. BOT-MOD identifies the underlying intent by engaging with the target agent in a multi-turn exchange guided by Gibbs-based sampling over candidate intent hypotheses. This progressively narrows the space of plausible agent objectives to identify the underlying behavior. To evaluate our approach, we construct a dataset derived from Moltbook that encompasses diverse benign and malicious behaviors based on actual community structures, posts, and comments. Results demonstrate that BOT-MOD reliably identifies agent intent across a range of adversarial configurations, while maintaining a low false positive rate on benign behaviors. This work advances the foundation for scalable, intent-aware moderation of agents in open multi-agent environments. --- ### Prediction of Rectal Cancer Regrowth from Longitudinal Endoscopy Authors: Jorge Tapias Gomez, Despoina Kanata, Aneesh Rangnekar, Christina Lee, Hannah Williams, Hannah Thompson, J. Joshua Smith, Francisco Sanchez-Vega, Mert R. Sabuncu, Julio Garcia-Aguilar, Harini Veeraraghavan Published: 2026-05-13 Categories: cs.CV arXiv: https://arxiv.org/abs/2605.12855v1 Clinical trial studies indicate benefit of watch-and-wait (WW) surveillance for patients with rectal cancer showing a complete or near clinical response (CR) directly after treatment (restaging). However, there are no objectively accurate methods to early detect local tumor regrowth (LR) in patients undergoing WW from follow-up exams. Hence, we developed Temporal Rectal Endoscopy Cross-attention (TREX), a longitudinal deep learning approach that combines pairs of images acquired at restaging and follow-up to distinguish CR from LR. TREX uses pretrained Swin Transformers in a siamese setting to extract features from longitudinal images and dual cross-attention to combine the features without spatial co-registration between image pairs. TREX and Swin-based baselines were trained under two settings: (a) detecting LR or CR at the last available follow-up and (b) early detection of LR at 3--6, 6--12, and 12--24 months before clinical confirmation. TREX achieved the highest accuracy in detecting LR with a high sensitivity of 97% $\pm$ 6% and a balanced accuracy of 90% $\pm$ 3%, and outperformed all baselines in early detection at both 3--6 (74% $\pm$ 1%) and 6--12 months (62% $\pm$ 4%) prior to clinical detection. Clinical validation via a surgeon survey showed that TREX matched attending-level overall accuracy (TREX: 86.21% vs.\ Clinicians: 87.84% $\pm$ 1.28%). Finally, we explored TREX's ability to predict treatment response by combining pre-treatment (pre-TNT) and restaging endoscopies, achieving a balanced accuracy of 73% $\pm$ 12%. These results show that longitudinal deep learning analysis of endoscopy may improve surveillance and enable earlier identification of rectal cancer regrowth. --- ### Multitask Multimodal Fusion with Tabular Foundation Models for Peak and Durability Prediction of Pertussis Booster Response Authors: Divya Sitani Published: 2026-05-13 Categories: cs.LG, q-bio.QM arXiv: https://arxiv.org/abs/2605.12852v1 Pertussis booster vaccination produces immune responses that vary widely across individuals in both peak magnitude and long-term durability. These two phases are governed by partly distinct biological compartments:peak reflects acute B-cell activation and antibody secretion, while durability reflects the establishment of long-term humoral memory. Yet most computational models target only one, missing the full boost-and-wane trajectory. Jointly predicting both is non-trivial because the two endpoints are biologically dissociated rather than redundant; samples are small, modalities are heterogeneous with structured missingness, and the two tasks rely on different measurement windows. We propose a multi-task contrastive multimodal fusion architecture combining frozen TabPFN-v2 per-modality encoders, a dual-label supervised contrastive loss that treats two subjects as a positive pair if they agree on the Task 1 label or the Task 2 label, modality dropout calibrated to empirical missingness, and missingness-masked attention fusion. Applied to a curated subset of the CMI-PB pertussis booster dataset (n = 158 subjects, four modalities, 44.9% with at least one modality missing; Spearman r = -0.58 between peak and durability, n = 96), the model achieves test AUROC 0.797 (95% CI (0.621, 0.948)) for peak response and 0.755 (95% CI (0.519, 0.945)) for durability, with both significant under joint label permutation (N = 1000; p = 0.002 and p = 0.045). Across logistic regression, XGBoost, and MLP baselines on raw features and on TabPFN embeddings, the proposed model is the only one whose 95% CIs lie above chance on both tasks simultaneously. Per-modality contribution analyses recover task-specific modality contributions consistent with the underlying immunology: peak prediction is carried by cytokine signatures, while durability is carried by baseline antibody features. --- ### PRISM: Perinuclear Ring-based Image Segmentation Method for Acute Lymphoblastic Leukemia Classification Authors: Larissa Ferreira Rodrigues Moreira, Leonardo Gabriel Ferreira Rodrigues, Rodrigo Moreira, André Ricardo Backes Published: 2026-05-13 Categories: cs.CV, cs.AI arXiv: https://arxiv.org/abs/2605.12851v1 Automated analysis of peripheral blood smears for Acute Lymphoblastic Leukemia (ALL) is hindered by low contrast and substantial variability in cytoplasmic appearance, which complicate conventional membrane-based segmentation. We found that many recent approaches rely on heavy neural architectures and extensive training, but still struggle to generalize across staining and acquisition variability. To address these limitations, we propose the Perinuclear Ring-based Image Segmentation Method (PRISM), which replaces explicit cytoplasmic delineation with adaptive concentric zones constructed around the nucleus. These perinuclear regions enable the extraction of robust cytoplasmic descriptors by integrating color information with texture statistics derived from grey-level co-occurrence patterns, without requiring accurate cell-boundary detection. A calibrated stacking ensemble of traditional classifiers leverages these descriptors to achieve a high performance, with an accuracy of 98.46% and a precision-recall AUC of 0.9937. --- ### Persona-Model Collapse in Emergent Misalignment Authors: Davi Bastos Costa, Renato Vicente Published: 2026-05-13 Categories: cs.CL, cs.AI, cs.CR, cs.LG arXiv: https://arxiv.org/abs/2605.12850v2 Fine-tuning large language models on narrow data with harmful content produces broadly misaligned behavior on unrelated prompts, a phenomenon known as emergent misalignment. We propose that emergent misalignment involves persona-model collapse: deterioration of the model's internal capacity to simulate, differentiate, and maintain consistent characters. We test this hypothesis behaviorally using two metrics: moral susceptibility (S) and moral robustness (R), computed from the across- and within-persona variability of models' Moral Foundations Questionnaire responses under persona role-play. These metrics formalize the model's ability to differentiate characters (S) and its consistency when simulating a given one (R). We evaluate four frontier models (DeepSeek-V3.1, GPT-4.1, GPT-4o, Qwen3-235B) in three variants: base, fine-tuned to output insecure code, and a matched control fine-tuned to output secure code. Across the four models, insecure fine-tuning produces an average $55\%$ increase in S, pushing all four insecure variants beyond the band observed across 13 frontier models benchmarked in prior work -- with GPT-4o reaching more than twice the band's upper end -- signaling dysregulated differentiation. It also causes an average $65\%$ decrease in R, equivalent to a $304\%$ increase in 1/R. By contrast, the matched secure control preserves S near the base and induces only a partial R loss, showing that these effects are largely misalignment-specific. Complementing these metric shifts, insecure variants' unconditioned responses converge toward saturation near the scale ceiling, departing markedly from both base models' structured responses and those elicited when base models role-play toxic personas. Taken together, these metrics provide a sensitive diagnostic for emergent misalignment and serve as behavioral evidence that it involves persona-model collapse. --- ### AssemblyBench: Physics-Aware Assembly of Complex Industrial Objects Authors: Danrui Li, Jiahao Zhang, Bernhard Egger, Moitreya Chatterjee, Suhas Lohit, Tim K. Marks, Anoop Cherian Published: 2026-05-13 Categories: cs.CV, cs.AI arXiv: https://arxiv.org/abs/2605.12845v1 Assembling objects from parts requires understanding multimodal instructions, linking them to 3D components, and predicting physically plausible 6-DoF motions for each assembly step. Existing datasets focus on simplified scenarios, overlooking shape complexities and assembly trajectories in industrial assemblies. We introduce AssemblyBench, a synthetic dataset of 2,789 industrial objects with multimodal instruction manuals, corresponding 3D part models, and part assembly trajectories. We also propose a transformer-based model, AssemblyDyno, which uses the instructional manual and the 3D shape of each part to jointly predict assembly order and part assembly trajectories. AssemblyDyno outperforms prior works in both assembly pose estimation and trajectory feasibility, where the latter is evaluated by our physics-based simulations. --- ### Bayesian Model Merging Authors: Kaiyang Li, Shaobo Han, Qing Su, Shihao Ji Published: 2026-05-13 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2605.12843v1 Model merging aims to combine multiple task-specific expert models into a single model without joint retraining, offering a practical alternative to multi-task learning when data access or computational budget is limited. Existing methods, however, face two key limitations: (1) they overlook the valuable inductive bias of strong anchor models and estimate the merged weights from scratch, and (2) they rely on a shared hyperparameter setting across different modules of the network, lacking a global optimization strategy. This paper introduces Bayesian Model Merging (BMM), a plug-and-play bi-level optimization framework, where the inner level formulates the model merging as an activation-based Bayesian regression under a strong prior induced by an anchor model, yielding an efficient closed-form solution; and the outer level leverages a Bayesian optimization procedure to search module-specific hyperparameters globally based on a small validation set. Furthermore, we reveal a key alignment between activation statistics and task vectors, enabling us to derive a data-free variant of BMM that estimates the Gram matrix for regression without any auxiliary data. Across extensive benchmarks, including up to 20-task merging in vision and 5-task merging in language, BMM consistently outperforms all plug-and-play anchor baselines (e.g., TA, WUDI-Merging, and TSV). In particular, on the ViT-L/14 benchmark for 8-task merging, a single merged model reaches 95.1, closely matching the average performance of eight task-specific experts (95.8). --- ### Decision Support for Marketplace Policies under Incomplete Evidence: From Replay to Launch Readiness Authors: Prashant Shekhar, Caroline Howard Published: 2026-05-13 Categories: stat.AP, cs.LG arXiv: https://arxiv.org/abs/2605.12840v1 Marketplace platforms routinely evaluate pricing and allocation policies using logged observational data, yet strong offline performance does not imply that a policy is safe to deploy. In real-time bidding (RTB) marketplaces, reserve-price and floor-policy changes affect not only revenue but also fill, advertiser value, budget pacing, and competition across auctions, creating feedback and interference. The central problem is therefore not to estimate whether a policy improves an offline metric, but to determine whether the available evidence justifies direct launch or only further validation. In this regard, we propose a support-aware decision-support system (DSS) that distinguishes promising from actionable evidence. The framework integrates replay, support-aware off-policy evaluation (OPE), conservative lower-bound ranking, multi-sided guardrails, out-of-time validation, sensitivity analysis, and interference-aware validation design into a claim-preserving pipeline that outputs a launch-readiness classification rather than a single performance estimate. Applying the framework to iPinYou-style RTB logs, we identify a margin-gated floor policy as the leading candidate, with a 47.7% replay yield lift, a 45.8% conservative lower-tail lift, and stable out-of-time performance. However, the framework does not recommend direct launch. A decision-rule ablation shows that simplified pipelines select the same policy but incorrectly recommend deployment, leaving key causal assumptions unresolved. In contrast, the proposed DSS selects the same policy but changes the action to online validation, reflecting missing evidence on propensities, bidder response, and interference. Overall, the contribution is a reproducible DSS protocol that prevents decision overclaim under partial identification and converts offline evaluation into an auditable, action-oriented recommendation. --- ### Multimodal Hidden Markov Models for Persistent Emotional State Tracking Authors: Anamika Ragu, Aneesh Jonelagadda Published: 2026-05-13 Categories: cs.AI arXiv: https://arxiv.org/abs/2605.12838v1 Tracking an interpretable emotional arc of a conversation via the sentiment of individual utterances processed as a whole is central to both understanding and guiding communication in applied, especially clinical, conversational contexts. Existing approaches to emotion recognition operate at the utterance level, obscuring the persistent phases that characterize real conversational dynamics. We propose a lightweight framework that models conversational emotion as a sequence of latent emotional regimes using sticky factorial HDP-HMMs over multimodal valence-arousal representations derived from simultaneous video, audio and textual input. We evaluate the quality of regime prediction using LLM-as-a-Judge, geometric, and temporal consistency metrics, demonstrating that the sticky HDP-HMM produces more interpretable regime sequences than the baseline Gaussian HMM at a fraction of the computational cost of LLM-based dialogue state tracking methods. In addition, Question-Answer experiments in a clinical dataset suggest that meaningful emotional phases can reliably be recovered from multimodal valence-arousal trajectories and used to improve the quality of LLM responses in unstable affective regimes via context augmentation. This framework thus opens a path toward interpretable, lightweight, and actionable analysis of conversational emotion dynamics at scale. --- ### Discrete Stochastic Localization for Non-autoregressive Generation Authors: Yunshu Wu, Jiayi Cheng, Longxuan Yu, Partha Thakuria, Rob Brekelmans, Evangelos E. Papalexakis, Greg Ver Steeg Published: 2026-05-13 Categories: cs.LG arXiv: https://arxiv.org/abs/2605.12836v2 Continuous diffusion is a natural framework for non-autoregressive generation but has generally lagged behind masked discrete diffusion models (MDMs) on discrete sequence generation. We argue that the bottleneck is not continuity itself, but a representation in which denoising depends on timestep-indexed noise regimes. We introduce \emph{Discrete Stochastic Localization} (DSL), a continuous-state framework with unit-sphere token embeddings whose Bayes-optimal denoiser is invariant to the nominal signal-to-noise ratio (SNR) under the localization channel. One trained network then supports an entire family of per-token SNR paths, with endpoint masked-diffusion paths as a special case. Fine-tuning a pretrained MDLM checkpoint with DSL substantially improves distributional faithfulness (MAUVE) on OpenWebText across all step budgets from $T{=}128$ to $T{=}1024$, and the same checkpoint supports random-order autoregressive sampling, as well as a hybrid continuous-then-discrete sampler using as few as T=48 total steps -- without distillation or retraining. --- ### PROMETHEUS: Automating Deep Causal Research Integrating Text, Data and Models Authors: Sridhar Mahadevan Published: 2026-05-13 Categories: cs.AI arXiv: https://arxiv.org/abs/2605.12835v1 Large language models can extract local causal claims from text, but those claims become more useful when organized as persistent, navigable world models rather than as flat summaries. We introduce PROMETHEUS, a framework that turns retrieved literature, filings, reviews, reports, agent traces, source data, code, simulations, and scientific models into causal atlases: sheaf-like families of local causal predictive-state models over an explicit cover of a research substrate. Each local region contains causal episodes, structured claim tables, predictive tests, support statistics, and provenance; restriction maps compare overlapping regions; gluing diagnostics expose agreement, drift, contradiction, and underdetermination. The resulting Topos World Model is not a single universal graph. It is a research instrument for navigating what a corpus says, where it says it, how strongly it is supported, and where local claims fail to assemble into a coherent global view. Three literature-atlas case studies -- ocean-temperature impacts on marine populations, GLP-1 weight-loss evidence, and resveratrol/red-wine health-benefit claims -- illustrate deep causal research from text with explicit locality, evidence, persistent state, and gluing tension. Four grounded-counterfactual case studies -- a Nature Climate Change microplastics forcing paper, an Indus Valley hydrology paper with VIC-derived figure data and model code, the canonical Sachs protein-signaling study with single-cell perturbation data, and a Nature singing-mouse study with MAPseq projection matrices -- show a stronger mode: when a paper ships source data, simulation outputs, or code, PROMETHEUS can evaluate a counterfactual against that scientific substrate and then rebuild the sheaf world model around the --- ### EduRiskX: A Neuro-Symbolic Framework with F-Logic Reasoning for Early Academic Risk Prediction Authors: Yu Fu, Yongqi Kang, Yong Zhao, Rongfang Bie Published: 2026-05-12 Categories: cs.AI arXiv: https://arxiv.org/abs/2608.26107v1 Predicting students' academic risk in online education is crucial for enabling timely interventions that can improve retention and learning outcomes. However, existing models often suffer from limited early detection capability and insufficient interpretability, leading to a "black-box" trust crisis that hinders their adoption in real-world pedagogical settings. To address these challenges, we propose EduRiskX, a neuro-symbolic framework that integrates a temporal Transformer-based predictor with F-Logic symbolic reasoning. The neural component models longitudinal student activity sequences using temporal attention, class-weighted loss, and dynamic weekly truncation. Acting as a data-driven expert system, an F-Logic rule base -- grounded in established educational theories (Engagement Theory and Student Integration Model) to mimic the diagnostic logic of human educators -- is constructed exclusively from the training data. The neural risk probability and the symbolic confidence score are then combined through a logistic regression-based fusion mechanism that learns the relative contribution of each signal. Experiments on the Open University Learning Analytics Dataset (OULAD) using a strict 80/10/10 student-level split show that EduRiskX achieves an accuracy of 0.900 and an F1-score of 0.894 at the end of the semester (Week 38), with an average early detection week of 9.32 and a detection rate of 94.30 percent. Compared with state-of-the-art time-series models (PatchTST, iTransformer) and common deep learning baselines (LSTM, CNN), EduRiskX yields improved recall and earlier risk identification under identical conditions. Beyond predictive performance, the F-Logic module provides structured rule-based explanations linking predictions to observable behavioral patterns and educational theories. --- ### Confidently Deceptive: On the Relationship Between Confidence and Deception in LLMs Authors: Ali Asad, Stephen Obadinma, Anshul Pattoo, Wenxuan Zhang, Xiaodan Zhu Published: 2026-05-12 Categories: cs.CL, cs.AI arXiv: https://arxiv.org/abs/2607.20444v2 The increasing capabilities of large language models (LLMs) are being accompanied by deep-rooted risks of deceptive behaviours that cause models to produce misleading outputs in service of a contextually or experimentally induced goal. The harm posed by such behaviours depends not only on the content of deceptive outputs but also how confidently models deliver them, since confidence has a major impact on how persuasive the communication is to end users. In this paper, we provide a comprehensive study on the crucial relationship between confidence and deception across existing deception benchmarks and different model families, while covering both verbalized numerical and logit-based aggregated confidence. Through this, we reveal how confidently models behave when being deceptive. We demonstrate that when producing deceptive rather than honest responses, models exhibit a gap between their belief (how likely they think a claim is to be true) and their commitment (how firmly they assert and would defend that claim). LLMs produce persuasive deceptive claims while reporting low belief in their factual correctness. Their reported commitment to deceptive responses can easily be increased through further prompting and preference fine-tuning, with smaller and condition-dependent changes in reported belief. However, we show that low reported belief remains comparatively invariant and provides a strong signal for detecting deception in the evaluated settings. Using only an API call, our approach achieves detection scores of up to 0.99 for induced deception and 0.89 for emergent deception. This ultimately shows how confidence can be a practical tool for detecting and diagnosing deceptive behaviour in LLMs. --- ### Confidently Deceptive: How Confidence Amplifies the Risk of LLM Deception Authors: Ali Asad, Stephen Obadinma, Anshul Pattoo, Wenxuan Zhang, Xiaodan Zhu Published: 2026-05-12 Categories: cs.CL, cs.AI arXiv: https://arxiv.org/abs/2607.20444v1 Large language models (LLMs) can produce deceptive responses: outputs that mislead users in service of a contextually or experimentally induced goal. Yet it remains unclear how confidently models deceive and whether higher confidence makes deceptive responses more persuasive to end users. In this paper, we study these basic questions in various models and different deception datasets. We provide a comprehensive study measuring confidence through both verbalized self-reports and a range of logit-based estimators. We show that LLMs deliver deceptive responses with substantial verbalized confidence and that human annotators prefer the higher-confidence deceptive response 78% of the time in paired comparisons. Misalignment fine-tuning amplifies the problem. Confidence in deceptive responses rises across all three benchmarks, increasing the resulting potential risk, with effects generalizing beyond the training distribution. Strikingly, models classify their own deceptive outputs as deceptive at high rates (82.7% under misalignment) while still predicting they would produce them - recognition without avoidance. We argue that confident deception is a distinct alignment risk requiring evaluations that jointly measure deception, confidence, and awareness. --- ### GLAN-QnA-KR: A Seedless Taxonomy-Driven Korean Instruction Corpus Authors: Daekeun Kim Published: 2026-05-12 Categories: cs.CL, cs.LG arXiv: https://arxiv.org/abs/2607.20443v1 We release GLAN-QnA-KR, a 303,581-row openly redistributable Korean instruction-QA corpus produced via the seedless taxonomy-driven GLAN synthesis pipeline with Microsoft's Phi-3.5-MoE-instruct as the producer model (generation: 2024-12; release: 2024-12; licence: OpenRAIL). The corpus spans a flat taxonomy of 1,084 English-labelled disciplines paired with Korean question/answer text, a 100-900 difficulty scale, and a median of 313 question characters and 1,098 answer characters per record. Two properties are atypical for synthetic instruction data at this scale: (i) exact duplicate questions number only 1 in 303,581 rows and character-trigram near-duplicate clusters at Jaccard >= 0.9 number zero in a 5,000-sample probe, and (ii) a two-layer contamination audit against KMMLU, KoBEST (five sub-tasks), and HAE-RAE-Bench shows a maximum test-vs-corpus question-level character-trigram Jaccard of 0.163 with zero test items at Jaccard >= 0.7, and a maximum multilingual-E5 cosine of 0.901 with a single test item at cosine >= 0.90 and zero at >= 0.95, across 20,000 sampled GLAN questions and seven evaluation sets. At the time of release, this is, to our knowledge, the largest single-pipeline synthetic Korean instruction corpus verifiable on the Hugging Face Hub and the only Korean >=100k-row corpus built under a seedless taxonomy-driven protocol. This note documents the generation protocol, corpus statistics, the contamination audit, and the licensing boundary in a form suitable for downstream citation. --- ### Naver-News-KO: A Korean News Summarization Dataset for Open-Source Fine-Tuning of Summarization Models Authors: Daekeun Kim Published: 2026-05-12 Categories: cs.CL, cs.LG arXiv: https://arxiv.org/abs/2607.20442v1 We release Naver-News-KO, a Korean news summarization dataset of 27,400 (document, summary) pairs collected from Naver News over a ten-day window in July 2022 across two categories (Economy and IT/Science; 77/23 split), with train/validation/test partitions of 22,194 / 2,466 / 2,740 and a mean per-record document-to-summary character-compression ratio of 6.03x. The dataset has been publicly hosted on the Hugging Face Hub since January 2023 and, as of May 2026, receives approximately 33,000 downloads per month; community-maintained Korean summarization models fine-tuned on it include Gemma-2B-ko and Gemma2-9B variants. This technical report (i) documents the collection protocol, the column schema, and the split construction, (ii) reports corpus-level statistics (length distributions, compression ratio, and a measured 16.8% near-duplicate title-Jaccard overlap between test and train that users should be aware of), (iii) positions the resource against other open Korean summarization corpora, (iv) provides a Lead-3 extractive reference point (ROUGE-1 55.1, ROUGE-L 50.6) and two reproducible fine-tuned baselines -- KoBART (R-1 56.6, BERTScore-F1 81.5) and Gemma-2B-ko with LoRA (R-1 55.3, BERTScore-F1 78.3) -- with release-time training scripts, and (v) clarifies the licensing and intended-use scope of the resource. The goal is to provide a citable reference for downstream work that already uses this dataset, not to propose a new benchmark. --- ### Belief Propagation in LLM World Models: Measuring Strategic Information Bias with Prediction Markets Authors: Mykola Khandoga, Yevhen Kostiuk, Anton Polishko, Yurii Filipchuk, Kostiantyn Kozlov, Dmytro Zamriy, Artur Kiulian Published: 2026-05-12 Categories: cs.CL, cs.LG arXiv: https://arxiv.org/abs/2607.20441v1 Every information ecosystem produces beliefs that shape strategic decisions. Both human analysts and AI systems inherit the blind spots of their information sources. We show that LLMs, combined with prediction markets, function as a calibrated instrument for measuring how far ecosystem-induced beliefs deviate from an external reference: LLMs extract the beliefs a text corpus implies, and prediction market price trajectories, anchored at resolution by realised outcomes, provide the calibration reference against which to quantify the deviation. We isolate the bias contribution of specific text through ablation: varying information context while holding the model fixed, with a contaminated model that knows actual outcomes as control. Applied to 111 Ukraine-related prediction markets, comprising approximately 93,000 predictions across four models, we find that English news context systematically biases territorial predictions, wrong 64 to 72 percent of the time when it pushes predictions toward territorial capture. A contaminated model that knows actual outcomes shows the same error rate, indicating that the bias originates primarily in the text. Supplementing with Ukrainian military-analytical sources reduces the bias for all clean models, while absolute-error gains are partial and model-dependent. We show that the distortion originates primarily in the sources, not the models. Consistent across four architectures, it will persist in any system that processes them and propagate into downstream decisions. --- ### Answer-then-Edit: Reasoning Skeleton Editing for Anti-Distillation with Preserved Utility Authors: Fan Li, Mengting Pan, Sijia Xu, Xiaoyang Wang, Chen Chen, Wenjie Zhang Published: 2026-05-12 Categories: cs.CL, cs.AI arXiv: https://arxiv.org/abs/2607.20440v1 Proprietary large language models (LLMs) entail substantial intellectual and financial investment, making them valuable intellectual property (IP). However, even when deployed via black-box APIs, these models remain vulnerable to unauthorized knowledge distillation, which allows adversaries to cheaply extract and replicate model capabilities. To address this issue, anti-distillation (AD) has been proposed to generate defensive outputs that hinder distillation effectiveness, overcoming the limitation of watermarking-based approaches that rely on post-hoc verification. However, existing AD methods based on internal model perturbations struggle to balance anti-distillability and utility (e.g., answer accuracy and naturalness) of reasoning traces, with stronger defenses often causing significant utility loss. To fill this gap, we propose \textbf{\underline{S}}keleton-\textbf{\underline{G}}uided \textbf{\underline{R}}easoning \textbf{\underline{E}}diting (SGRE), an \textit{Answer-then-Edit} framework that performs post-hoc trace modification for anti-distillation. In the answer stage, the teacher model first generates clean reasoning traces, preserving the original reasoning accuracy while enabling more flexible control over trace naturalness. In the editing stage, we draw inspiration from Cognitive Load Theory (CLT) and introduce a three-stage strategy consisting of reasoning skeleton extraction, skeleton graph coarsening, and skeleton verbalization. These operations jointly perturb reasoning structures and augment textual complexity to amplify extraneous load on student models, hindering their acquisition of underlying reasoning patterns. Extensive experiments across diverse LLMs demonstrate that SGRE achieves state-of-the-art performance in reducing distillation effectiveness, while maintaining lossless reasoning accuracy and superior trace naturalness. --- ### AsymVerify at SemEval-2026 Task 6: Asymmetric Confidence-Gated Verification for Political Evasion Detection Authors: Sebastien Kawada Published: 2026-05-12 Categories: cs.CL arXiv: https://arxiv.org/abs/2607.20439v1 Political evasion is difficult to detect because evasive answers often appear cooperative while avoiding concrete commitment. We present AsymVerify, a confidence-gated verification system for SemEval-2026 Task 6, a three-way classification of Clear Reply, Ambivalent, and Clear Non-Reply responses. AsymVerify scored 0.85 Macro F1 on the evaluation split (D_eval, n=237), placing 2nd out of 41 teams on the official leaderboard. The system first classifies each question-answer pair, then selectively applies downgrade verification (CR/CNR -> AMB) or upgrade verification (AMB -> CR) to low-confidence predictions. Development analysis shows that errors concentrate at the Ambivalent boundary in both directions, motivating this asymmetric two-verifier design while confidence gating keeps additional inference cost low. On D_dev (n=308), AsymVerify with GLM-4.7 gains +17.1 Macro F1 over single-pass classification at 1.48 calls/example, and the upgrade verifier alone improves every tested LLM backend on D_dev by +6.8 to +15.2 Macro F1 over its single-pass baseline. Code is available at https://github.com/kaons-research/AsymVerify-ACL. --- ### Preference Tuning as Spectral Update Reorganization Authors: Peiyan Zhang, Haibo Jin, Liying Kang, Haohan Wang Published: 2026-05-12 Categories: cs.CL, cs.AI, cs.LG arXiv: https://arxiv.org/abs/2607.20438v1 Preference-based post-training is usually understood through endpoint behavior, yet the learned update that produces this behavior remains largely opaque. We study RLHF and related preference optimization through the spectral structure of their induced parameter updates. By decomposing effective LoRA updates and reloading their spectral components as plug-in modules, we turn preference-induced updates into objects that can be isolated, recomposed, and directly intervened on. Across model families, optimization algorithms, and supervision regimes, these updates consistently develop a spectral head--tail organization. A compact head emerges early and carries the dominant endpoint shift, while a heterogeneous residual tail remains. The split is functional rather than merely descriptive. Plug-in intervention shows that the head accounts for the visible behavioral departure from the base model, while the tail is weak in isolation. Cross-run recomposition further shows that mixed adapters follow the source of the head, indicating that the head carries run-level solver bias. This endpoint dominance does not imply learning sufficiency. Head-only learning is non-vacuous but fails to recover the full solution, especially on out-of-distribution behavior. Tail-only learning yields little visible gain, yet the full solution is not recovered without the tail. These findings recast preference post-training as structured update reorganization rather than a monolithic behavioral correction, and suggest that alignment gain and coverage loss are tied to how the learned update itself is organized. --- ### MILP-Evo: Closed-Loop Fully Automatic Design of MILP Solvers Authors: Jinbiao Nie, Kewei Feng, Xiaoyuan Zhang, Shan Yin, Zizhuo Wang, Bin Dong Published: 2026-05-12 Categories: cs.AI, cs.NE arXiv: https://arxiv.org/abs/2607.18252v1 Machine learning methods have shown that data-driven policies can accelerate mixed-integer linear programming (MILP) solvers, but many such approaches remain difficult to inspect, adapt, and deploy because the learned policy is represented as an external predictor or other opaque model. By contrast, explicit solver logic is easier to understand and integrate, but is usually hand-designed rather than learned from solver feedback. We study whether the automatic design of MILP solver logic can instead be cast as LLM-guided closed-loop search over executable white-box components evaluated directly by end-to-end solver behavior. To this end, we propose a closed-loop program evolution framework for MILP solver auto-design, implemented through PySCIPOpt, and instantiate it on the joint design of a cut selector and a branching rule. Candidate programs are iteratively generated, loaded into SCIP, and evaluated by direct execution on MILP instances, with the resulting feedback guiding performance-based selection, targeted repair, diagnostic reflection, and diversity-aware population maintenance. The method outputs explicit solver components that can be inspected, modified, and deployed within standard solver workflows. Across four benchmark families, we find that LLM-guided program evolution can discover competitive domain-specialized policies in several settings. --- ### Semantic Register Compression in Multi-Agent LLM Cascades Authors: Manuele Tele Junior Fernandez Published: 2026-05-12 Categories: cs.CL arXiv: https://arxiv.org/abs/2607.14119v2 Multi-agent LLM systems commonly decompose complex tasks into specialized roles. However, this modularity introduces a representational risk: when intermediate agents transform text across linguistic registers, they can systematically compress the semantic distinctions needed for accurate downstream decisions. We term this phenomenon semantic register compression and characterize it as an observable failure mode in multi-agent cascades. Using a three-agent pipeline (Collector-Evaluator-Decider), we quantify compression via inter-label separation in sentence-transformer embedding space. Across political fact-checking (LIAR), sentiment analysis (SST-5), and medical triage (Triagegeist), critical evaluation reduces label separability at the Evaluator stage, while identity passthrough preserves it nearly fully. Five controlled variants show that geometric change depends on the specific intermediate transformation rather than on the mere presence of an additional cascade stage. A credibility-seeking variant expands rather than compresses inter-label separation, while shifting outputs toward mostly-true, demonstrating that transformation valence controls both the direction and the sign of geometric change independently of compression magnitude. Compression generalizes across the three domains with domain-dependent intensity (10.3% in fact-checking, 28.2% in sentiment, 9.1% in triage). A 20-level prompt gradient reveals a non-monotonic compression profile: balanced evaluative prompts produce the strongest compression, while extreme critical prompts show irregular moderate compression. These results demonstrate that semantic register compression is a measurable and generalizable phenomenon in multi-agent LLM systems, with implications for safety evaluation in high-stakes domains. --- ### Knowledge Graphs Meet Graph Neural Networks: A Comprehensive Survey Authors: Chengcheng Sun, Jiayun Tian, Cheng Zhai, Zhixiao Wang, Yajie Song, Xiaobin Rui, Jian Zhang, Philip S. Yu Published: 2026-05-12 Categories: cs.LG, cs.AI, cs.SI arXiv: https://arxiv.org/abs/2607.09666v1 Graph Neural Networks (GNNs) have emerged as a powerful paradigm in Knowledge Graphs (KGs) due to their intrinsic ability to model graph-structured data. However, there remains a lack of a systematic review about GNN-based methodologies across the entire knowledge graph technologies pipeline. To address this gap, we first propose a novel two-level taxonomy framework for GNN-based knowledge graph technologies: the KG technologies pipeline and GNN-based perspective. Specifically, the knowledge graph technologies pipeline covers knowledge graph construction, knowledge graph embedding, knowledge reasoning and knowledge graph applications. Meanwhile, the GNN-based perspective provides a new categorization of knowledge graph technologies with GNN models, such as GCN, GAT, and HGNN. Then, we analyze the advantages of GNN technology based on the characteristics of different tasks in the knowledge graph lifecycle. Furthermore, we detailed review various GNN-based models for knowledge graph following the proposed taxonomy, and summarize strengths and limitations. Finally, we discuss unresolved challenges and outline promising directions for future research. --- ### SemChunk-C: Semantic Segmentation for C Code Authors: Boris Nazarov, Darya Frolova, Shaked Leibzirer, Pavel Kisilev Published: 2026-05-12 Categories: cs.SE, cs.AI, cs.PL arXiv: https://arxiv.org/abs/2606.23697v1 Semantic segmentation of code written in a C-family language remains a challenging problem, due to the language's complex syntax, macro expansion, and irregular structural patterns. Existing chunking methods, such as fixed-sized windows, heuristic splitting, and syntax-based tools, often fail to capture meaningful functional units, limiting the efficacy of retrieval and other downstream LLM driven tasks. In this paper, we address the problem of chunking in C-related languages. First, we define a set of code chunk categories. Second, we train an LLM-based classifier to a) identify chunk boundaries, and b) assign each chunk a descriptive functional attribute (a category), which can be useful for downstream tasks. By leveraging the LLM's ability to capture semantic context within the code, we assume flexible chunk boundaries, allowing to adapt to the specific structure and context of each instance. Third, we introduce SemChunk-C, a family of lightweight language models for semantic chunking of C-related files (.c, .cpp, .h, .cs, etc.). These models are based on the first four Ettin encoders (1) with 17M, 32M, 68M, and 150M parameters. Despite their relatively small size, they are capable of identifying cohesive code units, such as data structures, interface blocks, and other components. Furthermore, we demonstrate the robustness of our approach on real-world code, including challenging constructs such as nested definitions and macros. We test our approach on various datasets, and show that it achieves high boundary accuracy and semantic coherence, matching or outperforming chunkers that are based on much larger code-oriented LLMs. We also validate the improved performance of the downstream tasks on a few curated benchmarks. --- ### Physical-AI: From Channel Awareness to Environmental Intelligence in 6G Wireless Networks Authors: Farooque Hassan Kumbhar, Kapal Dev, Sunder Ali Khowaja, Alexandros-Apostolos A. Boulogeorgos, Mehdi Bennis, Yuanwei Liu Published: 2026-05-12 Categories: cs.NI, cs.AI arXiv: https://arxiv.org/abs/2606.20583v1 Conventional wireless networks rely on instantaneous channel state information (CSI) and react to channel variations without explicitly modeling the physical environment, limiting their ability to handle blockage, mobility, and interference in dynamic deployments. Paradigms such as Integrated Sensing and Communication (ISAC) add sensing capabilities but lack explicit environment modeling and decision-making. In this article, we propose Physical-AI: a new architecture for environment-aware wireless networking, where radio signals enable sensing, modeling, and interaction with the environment in addition to data transmission. The framework proposes a self-supervised spatiotemporal radio foundation model for transforming distributed radio observations into a shared latent environmental representation. Multiple inference heads operate on this representation to estimate key environmental properties, including blockage, user distribution, mobility dynamics, and interference structure. A task-specific neural decision layer maps this representation to proactive, context-aware control actions. By integrating perception, world modeling, and decision-making in a closed loop, the proposed framework goes beyond ISAC and establishes Physical-AI as a promising architecture for intelligent 6G systems. Simulation results show that the proposed predictive framework reduces outage probability and blockage-response latency, particularly under increasing beam-switching delays. --- ### How Well Do Large Language Models Capture Human Personality? Authors: Aanisha Bhattacharyya, Yaman Kumar Singla, Rajiv Ratn Shah, Changyou Chen, Jitendra Ajmera Published: 2026-05-12 Categories: cs.HC, cs.AI arXiv: https://arxiv.org/abs/2606.18263v1 Large language models (LLMs) are increasingly used to simulate human populations via persona prompting, often under the assumptions that richer persona descriptions improve behavioral fidelity, similarly sized attribute combinations are equally simulatable, and persona definitions generalize across tasks. In this work, we formalize these assumptions and systematically evaluate them across multiple architectures, scales, and simulation settings. We identify a fundamental limitation we term persona manifold collapse, where increasingly expressive persona specifications lead to systematic contraction of representational and behavioral diversity. Across models, increasing persona complexity consistently reduces inter-persona separation in latent space and weakens behavioral differentiation in downstream simulation tasks. These effects persist across multiple analyses as richer personas fail to preserve human subgroup disagreement, performance varies across attribute combinations of similar size, and adding descriptive detail often degrades rather than improves simulation fidelity. Surprisingly, simple Age-Gender personas consistently outperform richly specified Ideal Customer Profiles (ICPs) across industries, achieving substantially higher downstream prediction accuracy. We find that collapse is not uniform across attributes. Certain combinations remain behaviorally stable and preserve stronger alignment with human responses, forming localized regions we term alignment bridges. Together, our results provide empirical and conceptual foundations for understanding the limits of persona-conditioned simulation, highlighting the need for representation-aware persona construction rather than increasing persona expressivity alone. --- ### Two Wrongs, No Right: Auditing Social-Desirability Bias in LLM Annotators for Computational Social Science Authors: Varun Kotte Published: 2026-05-12 Categories: cs.CY, cs.CL, cs.LG arXiv: https://arxiv.org/abs/2606.12426v1 LLM annotators are increasingly used in computational social science (CSS), but it is unclear whether their alignment-shaped errors preserve the empirical conclusions a researcher would report. We audit three open-source 7B instruction-tuned models (Zephyr, Mistral-Instruct, Qwen2.5-Instruct) across six TweetEval tasks under four prompt conditions (72 cells) and find that social-desirability failures do not run in a single direction. Zephyr exhibits leniency bias, systematically under-applying harmful labels (offensive language: false benign rate 0.729, false alarm rate 0.031). Mistral and Qwen exhibit overcorrection, over-applying the same labels (Mistral hate-speech FAR = 0.604). All three models exhibit neutrality bias on abortion stance, underestimating opposition prevalence by 24 to 40 percentage points and inflating the neutral label. None of the four prompting interventions we test (neutral, safety framing, depersonalized, chain-of-thought) corrects these failures across models; safety framing can worsen stance distortion. Strikingly, Zephyr's hate-speech prevalence estimate matches the gold rate exactly while its class-conditional errors are large in both directions, an accidental cancellation that misleads aggregate validation. We translate these patterns into a three-part taxonomy with diagnostic FBR/FAR signatures and a lightweight gold-sample validation protocol. The headline for trustworthy CSS: a model that looks calibrated on aggregate metrics can still flip the substantive empirical conclusion a researcher would report. --- ### An Explainable AI Assistant for Introductory Programming Education: Improving Feedback Reliability with Instructor-AI Collaboration Authors: Muntasir Hoq, Griffin Pitts, Bradford Mott, Seung Lee, Jessica Vandenberg, Shuyin Jiao, Narges Norouzi, James Lester, Bita Akram Published: 2026-05-12 Categories: cs.CY, cs.AI, cs.ET, cs.HC, cs.LG arXiv: https://arxiv.org/abs/2606.12425v1 Active learning is widely recognized as an effective approach for improving learning outcomes in introductory programming courses. However, insufficient instructional support often limits students' access to timely, personalized feedback, which is crucial for mastering foundational programming concepts. Although recent advances in AI, particularly large language models, offer scalable opportunities for feedback, concerns about explainability and reliability remain. In this paper, we present an AI-driven classroom assistant that leverages an explainable AI model to analyze student code, map logical errors to instructor-identified misconceptions, and deliver instructor-authored feedback, thereby grounding reliability in instructor-defined pedagogical knowledge. To evaluate the effectiveness of our framework, we conducted an expert evaluation to examine its alignment with instructor-verified feedback and deployed the system in a classroom setting to assess students' perceptions of its usability. Results indicate that the assistant can provide accurate, instructor-verified feedback to students while fostering a positive experience. --- ### SynIB: Informational Bottleneck for Maximizing Synergy in Multimodal Learning Authors: Konstantinos Kontras, Teodora Gagaleska, Thomas Strypsteen, Christos Chatzichristos, Matthew Blaschko, Maarten De Vos, Paul Pu Liang Published: 2026-05-12 Categories: cs.LG, cs.IT arXiv: https://arxiv.org/abs/2606.09853v1 A central objective in multimodal learning is to capture synergy: task-relevant information that arises only from the joint use of multiple modalities, and is not available from any single modality alone. While most approaches operate at the architectural level through larger or more complex fusion models, we propose a complementary axis: shaping the training objective itself. Standard training often emphasizes unimodal or redundant information, falling short on examples that require cross-modal reasoning. We formalize multimodal synergy through information theory and introduce the Synergistic Information Bottleneck (SynIB), a scalable objective that targets synergy directly. To prioritize learning synergy, SynIB motivates the model to predict accurately from all modalities while penalizing confidence when information from any modality is withheld. Alongside the standard task loss, the model runs forward passes with one modality masked at a time and is penalized for remaining confident, which would indicate reliance on unimodal cues rather than cross-modal interactions. We validate SynIB in two regimes. On synthetic XOR tasks where the ground-truth synergy is known by construction, standard training fails to recover it while SynIB does. On five real-world benchmarks, including three MultiBench affective tasks, Hateful Memes with CLIP-ViT and DeBERTa backbones, and a controllable irony extension of CREMA-D we introduce, SynIB improves accuracy on synergy-dependent examples by up to 7.8% and overall accuracy by up to 3.8%. --- ### Zero-shot Quantum Neural Architecture Search Authors: Tung Dao, Son N. Tran, Huynh Thi Thanh Binh Published: 2026-05-12 Categories: quant-ph, cs.LG, cs.NE arXiv: https://arxiv.org/abs/2605.27410v2 Variational Quantum Algorithms (VQAs) are a leading approach to exploiting near-term quantum hardware, leveraging parameterized quantum circuits and classical optimization to achieve advantage. Despite their promise, the practical deployment of VQAs is challenged by the difficulty of designing quantum circuit architectures that balance expressivity, trainability, and hardware constraints. Existing evolutionary-based quantum neural architecture search methods address these challenges but suffer from high computational costs due to repeated training of candidate circuits. In this work, we identify a setting in which the Gram matrix of the Quantum Neural Tangent Kernel converges. Building on this observation, we design a zero-shot surrogate model to estimate candidate performance without full training, significantly accelerating the architecture search process. Using this surrogate, we propose MZeQAS, a Monte Carlo Tree Search (MCTS)-based Zero-Shot Quantum Neural Architecture Search framework for VQAs. By integrating proxy-based performance estimation with MCTS exploration, MZeQAS efficiently discovers high-performing architectures. Experimental results demonstrate that MZeQAS outperforms existing approaches in terms of both search efficiency and solution quality, providing a scalable and effective framework for advancing VQA deployment on noisy intermediate-scale quantum devices. ---