arXiv subjects

Wei Ma

Publications and source records attributed to Wei Ma.

At least 19 recordsLinked to original sources

arXiv2026

Better Understanding, Better Fixes? A Study of Hallucination in LLM-based Automated Program Repair

Large language models (LLMs) have significantly advanced automated program repair (APR), yet existing evaluations remain largely result-centric and provide limited insight into hallucination during repair. In APR, hallucination may arise not only in final patches but also in the intermediate artifacts that guide patch generation. To address this gap, we perform a multi-layered analysis of hallucination throughout the APR process. Specifically, we characterize hallucination as the production of patches or intermediate artifacts that are not faithfully grounded in the available repair evidence. We examine repair hallucination in final patches and understanding hallucination in intermediate artifacts through three tasks, namely triggering testcase identification, line coverage prediction, and additional testcase generation.We then evaluate three representative LLMs on 832 Defects4J bugs through automatic evaluation and manual analysis. Our results show that both repair and understanding hallucinations remain prevalent. Across models and settings, only 21.0%-55.9% of generated patches pass the developer-written test suite. Moreover, although more accurate intermediate artifacts are generally associated with successful repairs, this relationship does not always hold. Manual analysis of 812 sampled repairs identifies repair hallucinations in 72.7% of cases, including patches that pass all available tests; incorrect causal localization and incorrect repair strategies account for 45.9% and 18.5% of these hallucinations, respectively. Meanwhile, models frequently misidentify triggering testcases, mispredict line coverage involving branching control flow, and generate additional testcases with missing bug-triggering conditions or incorrect expected behavior.

cs.SE↗

arXiv2026

SOMTab: Set-Order Mamba for Efficient Tabular In-Context Learning

Tabular foundation models based on in-context learning have recently emerged as strong alternatives to task-specific model fitting. However, the current performance frontier remains dominated by attention-heavy architectures, where attention is used throughout the modeling pipeline. This raises a natural question: is attention necessary at every stage of tabular in-context learning? We introduce SOMTab, a Set-Order Mamba architecture for efficient tabular in-context learning. SOMTab separates representation construction from query-conditioned retrieval. For row and column representations, it maps unordered table tokens into stable latent slots and applies Mamba-based state-space mixing to construct compact representations. For final prediction, it retains attention-based in-context learning to preserve query-conditioned retrieval from labeled context examples. We further introduce DCH-TailMix, a synthetic prior that combines degree-corrected graph heterogeneity with mixed heavy-tailed regimes to diversify synthetic dependency structures. Across tabular benchmarks, SOMTab approaches the performance of strong Transformer-based tabular foundation models while achieving faster inference and lower GPU memory usage, yielding a favorable efficiency--accuracy trade-off.

cs.LG↗

arXiv2026

Diffusion Bootstrap for High-Dimensional Linear Models

Classical bootstrap methods can behave poorly in high-dimensional linear models: the pairs bootstrap often yields overly conservative inference, whereas the residual bootstrap can be anti-conservative, reflecting systematic failures in variance calibration. We propose a diffusion-based pairs bootstrap that replaces the empirical joint distribution with a learned generative law. We establish variance consistency under a score approximation assumption, using complementary SDE and PDE arguments. Counterexamples show that terminal $W_4$ convergence alone is insufficient for variance consistency. Experiments indicate that diffusion pairs bootstrap improves variance calibration and generally improves Type~I error calibration, including in settings not covered by our theory.

stat.ME↗

arXiv2026

World Models as Adversaries: Multi-Agent Self-Play Fine-Tuning for Robust Motion Planning

Robust motion planning in dense traffic requires autonomous vehicles to interact in rare and safety-critical scenarios that are underrepresented in naturalistic driving data. Although adversarial training offers a feasible solution, existing methods often rely on external scenario generators, heuristic perturbations, or simulator-heavy rollouts, which makes them difficult to integrate with modern autoregressive planners. Here, we cast adversarially robust planner learning as a constrained min-max game and propose Adversarial World Modeling (AWM), a theoretically grounded multi-agent self-play fine-tuning framework. Since solving the exact game is intractable, AWM introduces a principled decoupled solver. In the inner minimization, the planner's predictive world model is converted into a role-conditioned adversary that learns sparse, scene-adaptive attack coalitions via counterfactual credit assignment. In the outer maximization, the ego planner optimizes a regret-aware robust best response against the frozen AWM, utilizing tail-risk weighting and reference-anchored trust regions to improve hard-case recovery while preserving nominal driving behavior. Experiments on the nuPlan and InterPlan benchmarks demonstrate that our method generates transferable adversarial interactions and yields a robust planner that achieves competitive closed-loop performance in both nominal and highly interactive long-tail scenarios. Theoretical analysis justifies the decoupled solver and the main optimization components.

cs.RO↗

arXiv2026

Discounted Expected Utility: A Revealed Preference Analysis

We present a revealed preference characterization of the discounted expected utility model with a concave utility function. The characterization offers a nonparametric test of the model. We apply the test to an experimental data set in the literature and find that the model is almost always rejected even when all payments involved are subject to risk.

econ.TH↗

arXiv2026

From Attacks to Curricula: Learnability-Guided Adversarial Training for Safe Autonomous Driving

Closed-loop adversarial training improves autonomous driving safety by exposing policies to rare safety-critical scenarios. Standard pipelines first generate adversarial scenarios and then sample them for policy optimization. However, most existing frameworks remain attack-oriented: collision-driven generators often synthesize unsolvable extreme situations, which can degrade learning, while heuristic samplers ignore the evolving capability of the driving policy, causing sample inefficiency and delayed convergence. We propose AlignADV, a learnability-guided closed-loop adversarial training framework that converts adversarial scenarios into resolvable and capability-aligned curricula. First, we reformulate adversarial scenario generation as a preference alignment problem and employ direct preference optimization to guide the generator toward critical yet resolvable scenarios. Second, we introduce behavioral fingerprints to capture the intrinsic characteristics of the evolving policy and construct a multi-modal capability prediction model that estimates policy performance without expensive closed-loop simulations. By combining resolvability-aligned scenarios with capability predictions, AlignADV develops a dynamic curriculum sampling mechanism that prioritizes scenarios targeting the current policy's vulnerabilities. Experiments on the Waymo Open Motion Dataset demonstrate that AlignADV improves convergence efficiency and final performance, reducing training steps by up to 40.6 percent compared with baseline methods while lowering collision rate and improving route completion under both normal and adversarial traffic conditions. These results highlight a shift from attack-oriented scenario generation to learnability-guided policy improvement, offering a principled direction for safer and more efficient autonomous driving training. Project page: https://meiyuewen.github.io/AlignADV/.

cs.RO↗

arXiv2026

UXBench: Benchmarking User Experience in AI Assistants

As AI assistants serve millions of users daily, evaluating user experience (UX) beyond general model capability has become increasingly important. We present UXBench, the first user-centric benchmark grounded in real user feedback signals for evaluating preference alignment and dialogue generation. The benchmark consists of three interconnected tasks, UX Judge, UX Eval, and UX Recovery, with 7,400 test instances extracted from over 70K interaction logs of a mainstream Chinese AI assistant. The dataset closely reflects real user distributions, covering 8 scenarios, 83 domains, and diverse failure patterns that pose severe challenges. Extensive experiments on 26 frontier language models provide novel insights into how well models perceive user experience and how improvements in model capability contribute to better dialogue engagement. Through comprehensive analysis of model behavior and performance gaps, we show that user feedback prediction is a learnable capability, where a reward model trained from in-the-wild feedback signals can achieve well-calibrated accuracy. We further document the systematic biases of LLM-as-a-judge evaluation protocols and compare typical response strategies that directly affect user experience. UXBench establishes a new evaluation landscape and calls for greater attention to tailored UX optimization, contributing to a user-centric scaling law that shapes the success of AI assistants.

cs.CL↗

arXiv2026

EvoDrive: Pareto Evolution for Safety-Critical Autonomous Driving via Self-Improving LLM Agents

Generating safety-critical scenarios is essential for validating and improving autonomous driving systems, yet it inherently requires maximizing adversariality to expose failures while preserving realism. Existing methods usually manage this trade-off with handcrafted heuristics, confining generation to known priors and overlooking underexplored patterns. While recent open-ended agentic evolution can push this limit, unconstrained general agents lack strict simulator grounding and tend to collapse the multi-objective tension into single-scalar maximization. Here we present EvoDrive, the first automated, LLM-based agentic evolution framework for multi-objective scenario generation. EvoDrive employs a simulator-grounded actor-critic architecture where a memory-driven actor iteratively proposes improvements to the generators and critics filter out implausible candidates, and a self-evolving world evaluator routes promising proposals to optimize simulation budgets. EvoDrive further maintains a Pareto archive of evaluated candidates to preserve diverse attack-realism trade-offs and guide future evolution via simulation feedback. Benchmark results on MetaDrive and CARLA show that EvoDrive not only significantly expands the Pareto frontier across various generators, but also produces valuable scenarios for policy training.

cs.AI↗

arXiv2026

MobEvolve: An Agentic Self-Evolving Heuristic System for Interpretable Human Mobility Generation

Human mobility generation aims to synthesize realistic trip chains for target populations based on individual features. Existing paradigms, including deep generative models, LLM-based methods, and traditional heuristics, struggle to satisfy the complex demands of this task while simultaneously maintaining interpretability, behavioral plausibility, population-level distributional alignment, and inference efficiency. To bridge this gap, we introduce MobEvolve, the first agentic self-evolving heuristic framework for human mobility generation. MobEvolve initializes a behavior-inspired heuristic system and employs an LLM agent to iteratively evolve its internal logic. By diagnosing empirical misalignments and failure cases on a validation set, the agent proposes targeted updates and accumulates evolution memory for cumulative self-improvement. Extensive evaluations on the Singapore and Montreal benchmarks demonstrate that MobEvolve significantly outperforms state-of-the-art deep generative and LLM-based methods in individual trajectory fidelity, population-level distribution alignment, and behavioral plausibility, while preserving interpretability and high inference efficiency.

cs.AI↗

arXiv2026

Reasoning-preserved Efficient Distillation of Large Language Models via Activation-aware Initialization

Efficient Distillation (EDistill) compresses large language models (LLMs) by structured pruning parameters and tuning lightweight modules with high training efficiency. Although these EDistilled LLMs achieve state-of-the-art (SOTA) performance on general ability benchmarks relative to similarly sized LLMs, we identify a severe degradation in their multi-step reasoning ability, which we term reasoning collapse. We systematically analyze the geometric origins of reasoning collapse and show that the SOTA EDistill method based on width-reducing projection matrices suffers from eRank collapse, in which the effective rank (eRank) of hidden representations drops. We theoretically explain how singular values of randomly initialized projection matrices become unevenly distributed, leading to eRank collapse and thus token indistinguishability. To address this issue, we propose RED (Reasoning-preserved Efficient Distillation) for LLMs, which introduces activation-aware initialization to initialize projection matrices as channel-selection matrices, thus theoretically mitigating eRank collapse. Experiments on Llama and Qwen series demonstrate that RED substantially recovers reasoning while maintaining high training efficiency and SOTA general ability.

cs.CL↗

arXiv2026

Same Signal, Different Semantics: A Cross-Framework Behavioral Analysis of Software Engineering Agents

Behavioral studies of LLM-based software engineering agents extract operational rules about which trajectory shapes correlate with higher resolution rates: that a test step follows a code modification, that error cascades are short, or that trajectories are compact. Each rule is typically derived from a single framework, and whether it transfers, in sign as well as magnitude, to structurally different agent designs has not been directly tested. We address this at ecosystem scale: 64,380 SWE-bench runs from 126 agent configurations spanning 43 frameworks, where each configuration pairs an LLM with a framework (e.g., SWE-Agent, OpenHands) that supplies its tools and workflow. We separate framework effects from LLM effects by holding each layer fixed in turn, then measure one behavior-outcome effect per configuration and examine how those effects agree or disagree. Swapping the framework while the LLM is held fixed produces large behavioral differences in every action feature. On most signals, configurations disagree not merely in magnitude but in direction. Error rate is the cleanest case: 47 configurations resolve more issues when their error rate is lower, while 48 resolve more when it is higher. Five other continuous features and three of seven binary patterns from prior SE literature show similar directional disagreement. Framework identity accounts for more of this variation than LLM family: for mean turns, framework explains 64% of the between-configuration variance against the LLM's 10%. The implication is that the same observable behavioral signal can carry opposite meaning for different agent configurations. Behavioral findings from any single framework therefore warrant cross-configuration validation before being claimed as general.

cs.SE↗

arXiv2026

CBARA: Covariate-Balanced-and-Adjusted Response-Adaptive Randomization

We propose the covariate-balanced-and-adjusted response-adaptive randomization (CBARA) procedure for adaptive design in clinical trials, which integrates the complementary strengths of covariate-adjusted response-adaptive randomization (CARA) and covariate-adaptive randomization (CAR). The CBARA procedure updates the target allocation ratio according to observed responses and patient covariate profiles without requiring a correctly specified model, thereby retaining CARA's ethical and efficiency considerations while improving robustness. In addition, the CBARA procedure extends the CAR principle from fixed target allocation ratios to covariate-adjusted adaptive target allocation ratios, yet still pursues balance in treatment allocation with respect to covariate features. This integration is enabled by a newly defined imbalance vector and three interrelated components: the allocation function, parameter estimation and update mechanism. We establish the asymptotic properties of covariate imbalance and the estimators under the CBARA procedure. The results demonstrate that the CBARA procedure can improve balance for both observed and unobserved covariates while preserving the consistency of the allocation ratio. The theoretical analysis is developed through a pseudo-Markov chain framework, where a new discrepancy measure for transition kernels is introduced to handle the continuity of Poisson equation solutions with respect to parameters.

stat.ME↗

arXiv2026

Spin-Axis-Layer Locking for Intrinsic Bipolar Altermagnetic Semiconductors: Proof-of-Concept in Bilayer CuBr2

Electrical control of spin and magnetic sublattice degrees of freedom is essential for multifunctional and low-power spintronic devices. Bipolar altermagnetic semiconductors (BAMSs)-characterized by opposite spin polarizations at the valence and conduction band edges-offer such control, yet known systems require external strain and sizable valley polarization for gate-tunable switching. Here, we propose a universal spin-axis-layer locking (SALL) paradigm to overcome these limitations. By stacking two quasi-1D ferromagnetic monolayers with a 90 degrees twist, the bilayer reconstructs altermagnetic symmetry, yielding an intrinsic BAMS state where carrier spin is locked to specific layers and transport directions. Using synthesized CuBr2 monolayers as proof-of-concept, we demonstrate via first-principles calculations a robust BAMS state. Electrostatic gating enables simultaneous, reversible switching of carrier type, spin, and active layer, generating fully spin-polarized axial charge currents and directionally controllable pure spin currents with near-unity charge-to-spin conversion efficiency. This SALL model establishes a versatile, strain-independent strategy for advanced all-electrical altermagnetic devices.

cond-mat.mtrl-sci↗

arXiv2026

Orbital angular momentum radiation and polarization of relativistic electrons in magnetic fields

While spin polarization from synchrotron radiation is well established, the polarization of orbital angular momentum (OAM) in such radiative processes remains elusive. We study radiation and polarization of relativistic electrons in a uniform magnetic field, focusing on OAM polarization radiation for vortex electrons which carry intrinsic OAM. The results illustrate that transition rates are asymmetric in the low-photon-energy regime, favoring OAM decrease, analogous to the spin-flip asymmetry in the Sokolov-Ternov effect. Under these conditions, synchrotron radiation can polarize the OAM. The characteristic relaxation time and stationary-state OAM distribution are obtained analytically. The polarization of spin about $\mathcal{P}_{\text{spin}}$ reaches $92.38\%$, while that of $\mathcal{P}_{\text{OAM}}$ can even approach almost unity for a large OAM; however, their polarization behaviors are different. For typical storage ring parameters, the OAM polarization time is orders of magnitude shorter than the spin polarization time. Thus, synchrotron radiation offers a mechanism for controlling vortex electron beams which carry OAM for high-energy accelerator applications.

physics.acc-ph↗

arXiv2026

Validating Computational Markers of Depressive Behavior: Cross-Linguistic Speech-Based Depression Detection with Neurophysiological Validation

Speech-based depression detection has shown promise as an objective diagnostic tool, yet the cross-linguistic robustness of acoustic markers and their neurobiological underpinnings remain underexplored. This study extends Cross-Data Multilevel Attention (CDMA) framework, initially validated on Italian, to investigate these dimensions using a Chinese Mandarin dataset with Electroencephalography (EEG) recordings. We systematically fuse read speech with spontaneous speech across different emotional valences (positive, neutral, negative) to investigate whether emotional arousal is a more critical factor than valence polarity in enhancing detection performance in speech. Additionally, we establish the first neurophysiological validation for a speech-based depression model by correlating its predictions with neural oscillatory patterns during emotional face processing. Our results demonstrate strong cross-linguistic generalizability of the CDMA framework, achieving state-of-the-art performance (F1-score up to 89.6%) on the Chinese dataset, which is comparable to the previous Italian validation. Critically, emotionally valenced speech (both positive and negative) significantly outperformed neutral speech. This comparable performance between positive and negative tasks supports the emotional arousal hypothesis. Most importantly, EEG analysis revealed significant correlations between the model's speech-derived depression estimates and neural oscillatory patterns (theta and alpha bands), demonstrating alignment with established neural markers of emotional dysregulation in depression. This alignment, combined with the model's cross-linguistic robustness, not only supports that the CDMA framework's approach is a universally applicable and neurobiologically validated strategy but also establishes a novel paradigm for the neurophysiological validation of computational mental health models.

eess.AS↗

arXiv2026

Hidden Reliability Risks in Large Language Models: Systematic Identification of Precision-Induced Output Disagreements

Large language models (LLMs) are increasingly deployed under diverse numerical precision configurations, including standard floating-point formats (e.g., bfloat16 and float16) and quantized integer formats (e.g., int16 and int8), to meet efficiency and resource constraints. However, minor inconsistencies between LLMs of different precisions are difficult to detect and are often overlooked by existing evaluation methods. In this paper, we present PrecisionDiff, an automated differential testing framework for systematically detecting precision-induced behavioral disagreements in LLMs. PrecisionDiff generates precision-sensitive test inputs and performs cross-precision comparative analysis to uncover subtle divergences that remain hidden under conventional testing strategies. To demonstrate its practical significance, we instantiate PrecisionDiff on the alignment verification task, where precision-induced disagreements manifest as jailbreak divergence-inputs that are rejected under one precision may produce harmful responses under another. Experimental results show that such behavioral disagreements are widespread across multiple open-source aligned LLMs and precision settings, and that PrecisionDiff significantly outperforms vanilla testing methods in detecting these issues. Our work enables automated precision-sensitive test generation, facilitating effective pre-deployment evaluation and improving precision robustness during training.

cs.AI↗

arXiv2026

ADV-0: Closed-Loop Min-Max Adversarial Training for Long-Tail Robustness in Autonomous Driving

Deploying autonomous driving systems requires robustness against long-tail scenarios that are rare but safety-critical. While adversarial training offers a promising solution, existing methods typically decouple scenario generation from policy optimization and rely on heuristic surrogates. This leads to objective misalignment and fails to capture the shifting failure modes of evolving policies. This paper presents ADV-0, a closed-loop min-max optimization framework that treats the interaction between driving policy (defender) and adversarial agent (attacker) as a zero-sum Markov game. By aligning the attacker's utility directly with the defender's objective, we reveal the optimal adversary distribution. To make this tractable, we cast dynamic adversary evolution as iterative preference learning, efficiently approximating this optimum and offering an algorithm-agnostic solution to the game. Theoretically, ADV-0 converges to a Nash Equilibrium and maximizes a certified lower bound on real-world performance. Experiments indicate that it effectively exposes diverse safety-critical failures and greatly enhances the generalizability of both learned policies and motion planners against unseen long-tail risks.

cs.LG↗

arXiv2026

SUBTA: A Framework for Supported User-Guided Bimanual Teleoperation in Structured Assembly

In human-robot collaboration, shared autonomy enhances human performance through precise, intuitive support. Effective robotic assistance requires accurately inferring human intentions and understanding task structures to determine optimal support timing and methods. In this paper, we present SUBTA, a supported teleoperation system for bimanual assembly that couples learned intention estimation, scene-graph task planning, and context-dependent motion assists. We validate our approach through a user study (N=12) comparing standard teleoperation, motion-support only, and SUBTA. Linear mixed-effects analysis revealed that SUBTA significantly outperformed standard teleoperation in position accuracy (p<0.001, d=1.18) and orientation accuracy (p<0.001, d=1.75), while reducing mental demand (p=0.002, d=1.34). Post-experiment ratings indicate clearer, more trustworthy visual feedback and predictable interventions in SUBTA. The results demonstrate that SUBTA greatly improves both effectiveness and user experience in teleoperation.

cs.RO↗