Searcharxiv⌕ Search

arXiv subjects

Zhongyu Wei

Publications and source records attributed to Zhongyu Wei.

At least 37 records · Page 2Linked to original sources
arXiv2026

LIBERO-Occ: Evaluating and Improving Vision-Language-Action Models under Scene-Induced Occlusion via Viewpoint Imagination

Vision-Language-Action (VLA) models achieve strong performance on standard manipulation benchmarks, but most evaluations assume that task-relevant objects are fully visible. This assumption often fails in realistic settings, where occlusion makes manipulation partially observable. In this paper, we study \textit{scene-induced occlusion} as a fundamental challenge for VLA models and introduce \textbf{LIBERO-Occ}, an occlusion-oriented extension of LIBERO. Experiments show that state-of-the-art VLAs suffer substantial performance degradation under occlusion. To address this issue, we propose \textbf{Viewpoint Imagination (VIM)}, which generates a complementary view from an occluded primary observation and conditions action prediction on both observed and imagined evidence. VIM improves robustness across task suites, occlusion types, and severity levels without requiring additional cameras at deployment time, suggesting that viewpoint imagination is an promising mechanism for perception completion in partially observable manipulation. Our benchmark and corresponding code are available at: \href{https://github.com/litsh/Libero-Occ}{https://github.com/litsh/Libero-Occ}.

cs.CV↗
arXiv2026

Can Agents Read the Room? Benchmarking Visual Social Intelligence in Multimodal Simulation

Social interaction depends on both language and visible social signals, such as facial expressions, posture, gaze, and emotional shifts. Yet existing social-agent benchmarks are largely text-based and rarely test whether multimodal agents can use visual cues to guide interaction. We introduce \textsc{\benchmarkname{}}, a benchmark evaluating visual social intelligence in multimodal social simulation. It contains 240 scenarios, 585 role instances, and 2,340 role-task instances, combining aligned textual-visual evidence, structured role profiles, and four role-level tasks: expression task, characteristic task, interaction regulation task, and interaction outcome task. Evaluating seven recent MLLMs under verbalized-vision and direct-vision reveals a clear gap between local role enactment and interaction management: role-specific expression and conflict handling are near saturation, whereas interaction regulation and visually grounded outcome achievement remain substantially more difficult. The code is released at https://github.com/JunsWan/AgentViSS, and the dataset is available at https://huggingface.co/datasets/JunsWan/AgentViSS.

cs.CL↗
arXiv2026

From Knowing to Acting: Benchmarking Self-Awareness Capability of LLM Agents

The integration of external tools has transitioned LLM agents from passive responders to autonomous systems. However, current benchmarks prioritize execution success, neglecting self-awareness capability, the ability to discern whether a problem requires necessary external resources or can be solved via internal parametric knowledge. To address this, we introduce KAPRO (Knowing-Acting Quadrant PRObe), a framework that evaluates cognitive-behavioral alignment by decoupling an agent's metacognitive judgment (Knowing) from its spontaneous execution (Acting). We further construct KAware, a dataset rigorously partitioning tasks into external, internal, and hybrid subspaces to systematically probe these epistemic boundaries. Extensive experiments across diverse agent architectures show that self-awareness capability is strongly correlated with task success but degrades sharply in internal-capability settings. Moreover, open-source and instruction-following models exhibit stronger tool overuse due to shallow pattern matching, while proprietary and reasoning-oriented models demonstrate more reliable cognitive gating. Benchmark and codes are available at https://github.com/AI-Santiago/KAware.

cs.AI↗
arXiv2026

CURP: Codebook-based Continuous User Representation for Personalized Generation with LLMs

User modeling characterizes individuals through their preferences and behavioral patterns to enable personalized simulation and generation with Large Language Models (LLMs) in contemporary approaches. However, existing methods, whether prompt-based or training-based methods, face challenges in balancing personalization quality against computational and data efficiency. We propose a novel framework CURP, which employs a bidirectional user encoder and a discrete prototype codebook to extract multi-dimensional user traits. This design enables plug-and-play personalization with a small number of trainable parameters (about 20M parameters, about 0.2\% of the total model size). Through extensive experiments on variant generation tasks, we show that CURP achieves superior performance and generalization compared to strong baselines, while offering better interpretability and scalability. The code are available at https://github.com/RaidonWong/CURP_code

cs.CL↗
arXiv2026

Beyond Isolated Behaviors: Hierarchical User Modeling for LLM Personalization

Large Language Models (LLMs) have demonstrated remarkable capabilities across diverse domains, yet personalizing their outputs to individual users remains an open challenge. Existing approaches predominantly adopt a flat behavioral paradigm, aggregating user behaviors without an explicit account of how they are organized into deeper behavioral structures. In this work, we draw on Pierre Bourdieu's Theory of Practice to propose PHF (Practice-Habitus-Field), a sociologically grounded framework that reconceptualizes LLM personalization through three hierarchical levels: individual behaviors as practices, their temporal accumulation into stable dispositions as habitus, and shared regularities across similar users as fields. We instantiate PHF through $\mathrm{PHF}_{\text{Compass}}$, a lightweight and model-agnostic implementation based on a frozen LLM. Experiments on the Language Model Personalization (LaMP) benchmark demonstrate consistent improvements across diverse tasks, while further analyses validate the interpretability and extensibility of the learned behavioral structures.

cs.CL↗
arXiv2026

HyLaT: Efficient Multi-Agent Communication via Hybrid Latent-Text Protocol

Communication protocol design is a central challenge in large language model-based multi-agent systems. Existing single-channel approaches face an inherent communication trilemma: text-based methods are interpretable but verbose, while latent-space methods are efficient but opaque and limited to unidirectional workflows. Inspired by multi-channel communication theory, we propose HyLaT, a hybrid latent-text communication protocol that transmits elaborate cognitive signals through a latent channel for efficiency, while expressing concise critical signals in natural language to preserve interpretability and precision. We introduce a two-stage training framework combining single-agent hybrid generation learning and multi-agent interactive co-training, enabling agents to generate and interpret hybrid messages across multiple rounds of interaction. Experiments demonstrate that HyLaT reduces communication overhead significantly while maintaining competitive task performance, with strong generalization and robustness across diverse settings.

cs.CL↗
arXiv2026

ACE: Self-Evolving LLM Coding Framework via Adversarial Unit Test Generation and Preference Optimization

Large Language Models (LLMs) excel at code generation but remain heavily reliant on large-scale annotated solutions and verification-based supervision, which constrains scalability and hinders sustained self-improvement. Recent solver--verifier frameworks exploit program execution as an automatic supervision signal, but their effectiveness degrades as solvers become moderately strong: verifier-generated tests increasingly confirm semantic correctness rather than exposing the remaining failure modes. We propose \textbf{ACE}, a self-evolving code generation framework based on a solver--adversary architecture that prioritizes active failure discovery through execution-centric supervision. A single LLM alternates between generating candidate programs and producing adversarial unit test inputs optimized to induce execution-level failures, such as runtime errors, exceptions, or non-termination. Supervision is derived solely from execution outcomes: robust programs are selected for supervised fine-tuning, while adversarial tests are optimized via Kahneman--Tversky Optimization using execution-derived preferences. Notably, the entire training loop requires no ground-truth code or external reward models. Experiments on CodeContests, MBPP, and LiveCodeBench demonstrate that ACE consistently outperforms strong solver--verifier baselines, achieving 3--7\% absolute gains in pass@1, with larger improvements on out-of-distribution benchmarks, while maintaining competitive or improved inference efficiency.

cs.SE↗
arXiv2026

PersonaDual: Balancing Personalization and Objectivity via Adaptive Reasoning

As users increasingly expect LLMs to align with their preferences, personalized information becomes valuable. However, personalized information can be a double-edged sword: it can improve interaction but may compromise objectivity and factual correctness, especially when it is misaligned with the question. To alleviate this problem, we propose PersonaDual, a framework that supports both general-purpose objective reasoning and personalized reasoning in a single model, and adaptively switches modes based on context. PersonaDual is first trained with SFT to learn two reasoning patterns, and then further optimized via reinforcement learning with our proposed DualGRPO to improve mode selection. Experiments on objective and personalized benchmarks show that PersonaDual preserves the benefits of personalization while reducing interference, achieving near interference-free performance and better leveraging helpful personalized signals to improve objective problem-solving.

cs.AI↗
arXiv2026

Mixture-of-Visual-Thoughts: Exploring Context-Adaptive Reasoning Mode Selection for General Visual Reasoning

Current visual reasoning methods mainly focus on exploring specific reasoning modes. Although improvements can be achieved in particular domains, they struggle to develop general reasoning capabilities. Inspired by this, we propose a novel adaptive reasoning paradigm, Mixture-of-Visual-Thoughts (MoVT), which unifies different reasoning modes within a single model and guides it to select the appropriate mode based on context. To achieve this, we introduce AdaVaR, a two-stage Adaptive Visual Reasoning learning framework: different modes are unified and learned during the supervised cold-start stage, and the mode selection capability is induced via an RL process with a carefully designed AdaGRPO algorithm. Extensive experiments show that AdaVaR effectively guides the model to learn and differentiate multiple modes and perform context-adaptive mode selection, achieving consistent improvement across various scenarios, highlighting MoVT as an effective solution for building general visual reasoning models.

cs.AI↗
arXiv2026

CT-FineBench: A Diagnostic Fidelity Benchmark for Fine-Grained Evaluation of CT Report Generation

The evaluation of generated reports remains a critical challenge in Computed Tomography (CT) report generation, due to the large volume of text, the diversity and complexity of findings, and the presence of fine-grained, disease-oriented attributes. Conventional evaluation metrics offer only coarse measures of lexical overlap or entity matching and fail to reflect the granular diagnostic accuracy required for clinical use. To address this gap, we propose CT-FineBench, a benchmark built from CT-RATE and Merlin to evaluate the fine-grained factual consistency of CT reports, constructed from CT-RATE and Merlin. Our benchmark is constructed through a meticulous, Question-Answering (QA) based process: first, we identify and structure key, finding-specific clinical attributes (like location, size, margin). Second, we systematically transform these attributes into a QA dataset, where questions probe for specific clinical details grounded in gold-standard reports. The evaluation protocol for CT-FineBench involves using this QA dataset to query a machine-generated report and scoring the correctness of the answers. This allows for a comprehensive, interpretable, and clinically-relevant assessment, moving beyond superficial lexical overlap to pinpoint specific clinical errors. Experiments show that CT-FineBench correlates better with expert clinical assessment and is substantially more sensitive to fine-grained factual errors than prior metrics.

cs.AI↗
arXiv2026

SpeechMedAssist: Efficiently and Effectively Adapting Speech Language Models for Medical Consultation

Medical consultations are intrinsically speech-centric. However, most prior works focus on long-text-based interactions, which are cumbersome and patient-unfriendly. Recent advances in speech language models (SpeechLMs) have enabled more natural speech-based interaction, yet the scarcity of medical speech data and the inefficiency of directly fine-tuning on speech data jointly hinder the adoption of SpeechLMs in medical consultation. In this paper, we propose SpeechMedAssist, a SpeechLM natively capable of conducting speech-based multi-turn interactions with patients. By exploiting the architectural properties of SpeechLMs, we decouple the conventional one-stage training into a two-stage paradigm consisting of (1) Knowledge & Capability Injection via Text and (2) Modality Re-alignment with Limited Speech Data, thereby reducing the requirement for medical speech data to only 10k synthesized samples. To evaluate SpeechLMs for medical consultation scenarios, we design a benchmark comprising both single-turn question answering and multi-turn simulated interactions. Experimental results show that our model outperforms all baselines in both effectiveness and robustness in most evaluation settings.

cs.CL↗
arXiv2026

Listen, Pause, and Reason: Toward Perception-Grounded Hybrid Reasoning for Audio Understanding

Recent Large Audio Language Models have demonstrated impressive capabilities in audio understanding. However, they often suffer from perceptual errors, while reliable audio reasoning is unattainable without first grounding the model's perception in structured auditory scenes. Inspired by Auditory Scene Analysis, we first introduce a Perception-Aware Question Answering (PAQA) dataset. PAQA implements a hierarchical decoupling strategy that separates speech from environmental sound and distinguishes multiple speakers, providing explicit perceptual reasoning for training. Building on this, we propose HyPeR, a two-stage Hybrid Perception-Reasoning framework. In Stage I, we finetune the model on PAQA to perceive acoustic attributes in complex audio. In Stage II, we leverage GRPO to refine the model's internal deliberation. We also introduce PAUSE tokens to facilitate latent computation during acoustically ambiguous phases and design perceptual consistency reward to align reasoning rationales with raw audio. Experiments across benchmarks demonstrate that HyPeR achieves absolute improvements over the base model, with performance comparable to large-scale models, stressing the effectiveness of hybrid perception-grounded reasoning for robust and multi-speaker audio understanding.

cs.SD↗
arXiv2026

MedRCube: A Multidimensional Framework for Fine-Grained and In-Depth Evaluation of MLLMs in Medical Imaging

The potential of Multimodal Large Language Models (MLLMs) in domain of medical imaging raise the demands of systematic and rigorous evaluation frameworks that are aligned with the real-world medical imaging practice. Existing practices that report single or coarse-grained metrics are lack the granularity required for specialized clinical support and fail to assess the reliability of reasoning mechanisms. To address this, we propose a paradigm shift toward multidimensional, fine-grained and in-depth evaluation. Based on a two-stage systematic construction pipeline designed for this paradigm, we instantiate it with MedRCube. We benchmark 33 MLLMs, \textit{Lingshu-32B} achieve top-tier performance. Crucially, MedRCube exposes a series of pronounced insights inaccessible under prior evaluation settings. Furthermore, we introduce a credibility evaluation subset to quantify reasoning credibility, uncover a highly significant positive association between shortcut behavior and diagnostic task performance, raising concerns for clinically trustworthy deployment. The resources of this work can be found at https://github.com/F1mc/MedRCube.

cs.CL↗
arXiv2026

SpatialAnt: Autonomous Zero-Shot Robot Navigation via Active Scene Reconstruction and Visual Anticipation

Vision-and-Language Navigation (VLN) has recently benefited from Multimodal Large Language Models (MLLMs), enabling zero-shot navigation. While recent exploration-based zero-shot methods have shown promising results by leveraging global scene priors, they rely on high-quality human-crafted scene reconstructions, which are impractical for real-world robot deployment. When encountering an unseen environment, a robot should build its own priors through pre-exploration. However, these self-built reconstructions are inevitably incomplete and noisy, which severely degrade methods that depend on high-quality scene reconstructions. To address these issues, we propose SpatialAnt, a zero-shot navigation framework designed to bridge the gap between imperfect self-reconstructions and robust execution. SpatialAnt introduces a physical grounding strategy to recover the absolute metric scale for monocular-based reconstructions. Furthermore, rather than treating the noisy self-reconstructed scenes as absolute spatial references, we propose a novel visual anticipation mechanism. This mechanism leverages the noisy point clouds to render future observations, enabling the agent to perform counterfactual reasoning and prune paths that contradict human instructions. Extensive experiments in both simulated and real-world environments demonstrate that SpatialAnt significantly outperforms existing zero-shot methods. We achieve a 66% Success Rate (SR) on R2R-CE and 50.8% SR on RxR-CE benchmarks. Physical deployment on a Hello Robot further confirms the efficiency and efficacy of our framework, achieving a 52% SR in challenging real-world settings.

cs.RO↗
arXiv2026

ProcureGym: A Multi-Agent Markov Game Framework for Modeling National Volume-based Drug Procurement

In this paper, we introduce ProcureGym, an data-driven multi-agent simulation platform that models China's National Volume-Based drug Procurement (NVBP) as a Markov Game. Based on real-world data from 7 rounds of NVBP (covering 325 drugs and 2,267 firms), the platform establishes a high-fidelity simulation environment. Within this framework, we evaluate diverse agent models, including Reinforcement Learning (RL), Large Language Model (LLM), and Rule-based algorithms. Experimental results demonstrate that RL agents achieve superior winner alignment and profits. Further analyses show that maximum valid bidding price and procurement volume dominate strategic outcomes. ProcureGym thus serves as a rigorous instrument for assessing policy impacts and formulating future procurement strategies.

cs.SI↗
arXiv2026

LifeSim: Long-Horizon User Life Simulator for Personalized Assistant Evaluation

The rapid advancement of large language models (LLMs) has accelerated progress toward universal AI assistants. However, existing benchmarks for personalized assistants remain misaligned with real-world user-assistant interactions, failing to capture the complexity of external contexts and users' cognitive states. To bridge this gap, we propose LifeSim, a user simulator that models user cognition through the Belief-Desire-Intention (BDI) model within physical environments for coherent life trajectories generation, and simulates intention-driven user interactive behaviors. Based on LifeSim, we introduce LifeSim-Eval, a comprehensive benchmark for multi-scenario, long-horizon personalized assistance. LifeSim-Eval covers 8 life domains and 1,200 diverse scenarios, and adopts a multi-turn interactive method to assess models' abilities to complete explicit and implicit intentions, recover user profiles, and produce high-quality responses. Under both single-scenario and long-horizon settings, our experiments reveal that current LLMs face significant limitations in handling implicit intention and long-term user preference modeling.

cs.CL↗
arXiv2026

Rethinking the Efficiency and Effectiveness of Reinforcement Learning for Radiology Report Generation

Radiologists highly desire fully automated AI for radiology report generation (R2G), yet existing approaches fall short in clinical utility. Reinforcement learning (RL) holds potential to address these shortcomings, but its adoption in this task remains underexplored. In this paper, we revisit RL in terms of data efficiency and optimization effectiveness for R2G tasks. First, we explore the impact of data quantity and quality on the performance of RL in medical contexts, revealing that data quality plays a more critical role than quantity. To this end, we propose a diagnostic diversity-based data sampling strategy that enables comparable performance with fewer samples. Second, we observe that the majority of tokens in radiology reports are template-like and diagnostically uninformative, whereas the low frequency of clinically critical tokens heightens the risk of being overlooked during optimization. To tackle this, we introduce Diagnostic Token-weighted Policy Optimization (DiTPO), which directly optimizes for clinical accuracy by using a diagnostic F1 score as the reward signal. Unlike standard RL approaches that treat all tokens equally, DiTPO explicitly models the varying importance of different tokens through rule- or gradient-based mechanisms to prioritize clinically relevant content. Extensive experiments on the MIMIC-CXR, IU-Xray, and CheXpert Plus datasets demonstrate that our framework achieves state-of-the-art (SOTA) performance while requiring substantially fewer training samples in RL. Notably, on MIMIC-CXR, our framework attains an F1 score of 0.516 using only 20% of the RL training samples.

cs.CV↗
arXiv2026

Not All Models Suit Expert Offloading: On Local Routing Consistency of Mixture-of-Expert Models

Mixture-of-Experts (MoE) enables efficient scaling of large language models (LLMs) with sparsely activated experts during inference. To effectively deploy large MoE models on memory-constrained devices, many systems introduce *expert offloading* that caches a subset of experts in fast memory, leaving others on slow memory to run on CPU or load on demand. While some research has exploited the locality of expert activations, where consecutive tokens activate similar experts, the degree of this **local routing consistency** varies across models and remains understudied. In this paper, we propose two metrics to measure local routing consistency of MoE models: (1) **Segment Routing Best Performance (SRP)**, which evaluates how well a fixed group of experts can cover the needs of a segment of tokens, and (2) **Segment Cache Best Hit Rate (SCH)**, which measures the hit rate of an expert cache utilizing a length of future information under a cache limit. We analyze 20 MoE LLMs with diverse sizes and architectures and use toy models to verify key factors related to local routing consistency. We find a strong trade-off between local routing consistency and *local* load balance, while showing that *global* load balance can coexist with local routing consistency. Meanwhile, settings like shared experts that decrease expert combination space can lead to low local routing consistency. We further reveal that domain-specialized experts contribute more to routing consistency than vocabulary-specialized ones, and that most models balance between cache effectiveness and efficiency with cache sizes approximately twice the active experts. These findings pave the way for memory-efficient MoE design and deployment without compromising inference speed. We publish the code for replicating experiments at https://github.com/ljcleo/moe-lrc .

cs.LG↗