Searcharxiv⌕ Search

arXiv subjects

Wentao Shi

Publications and source records attributed to Wentao Shi.

At least 19 recordsLinked to original sources
arXiv2026

CAST: Game Solvers as Turn-Level Teachers for LLM Agents

Training large language models (LLMs) to act in long-horizon games is a promising step toward generalist decision-making, yet reinforcement learning with verifiable rewards (RLVR) relies on sparse final rewards that reveal little about which decisions determine success. Denser process signals could supply this missing turn-level credit, but existing sources are hard to keep both cheap and accurate. We observe that changes in a game solver's state value reveal whether an action advances the state toward success. Building on this insight, we propose CAST (Credit Assignment from Solver Teachers), which converts these value changes into solver advantages and injects them into RLVR as turn-level signals. We further show that, under a soft-optimal solver assumption, maximizing the solver advantage is equivalent to on-policy distillation from the solver, requiring only scalar values rather than teacher logits. Across Sokoban, Minesweeper, and Rush Hour, CAST outperforms all trained baselines on every game under both in-domain and unseen-difficulty evaluation and achieves the highest average zero-shot performance on ALFWorld and WebShop. Our code is available at https://github.com/Wloner0809/CAST.

cs.CL↗
arXiv2026

A Clinician-Centered Pipeline for Annotation and Evaluation in Ultrasound AI Studies

Clinician-centered evaluation is critical for validating medical AI systems, especially in ultrasound imaging where quantitative metrics do not always capture clinical usability. Existing medical image platforms primarily focus on dataset labeling. They lack integrated support for blinded model comparison and reproducible evaluation workflows. We present a clinician-centered pipeline for remote annotation and evaluation in ultrasound AI studies. The proposed pipeline uses a centralized server and lightweight browser interfaces to enable clinicians to perform annotation, blinded ranking, and review without local dataset downloads. The pipeline also supports multi-rater participation, centralized result aggregation, and automated statistical analysis. We validate the pipeline in a fetal ultrasound segmentation study with six raters spanning expert, generalist, and non-expert experience levels. The system automatically generated Spearman correlation, Kendall's $\tau$, and top-1 selection statistics. Results indicated moderate to strong agreement across experts and other groups. The blinded evaluation results showed a tendency for later active learning models to be preferred. These outcomes suggest that the pipeline can support clinician-centered annotation and reproducible human-\ac{AI} evaluation studies in ultrasound imaging. The proposed pipeline is available on \href{https://github.com/13204942/SonoRate}{GitHub}.

cs.HC↗
arXiv2026

Look Before You Leap: Autonomous Exploration for LLM Agents

Large language model based agents often fail in unfamiliar environments due to premature exploitation: a tendency to act on prior knowledge before acquiring sufficient environment-specific information. We identify autonomous exploration as a critical yet underexplored capability for building adaptive agents. To formalize and quantify this capability, we introduce Exploration Checkpoint Coverage, a verifiable metric that measures how broadly an agent discovers key states, objects, and affordances. Our systematic evaluation reveals that agents trained with standard task-oriented reinforcement learning consistently exhibit narrow and repetitive behaviors that impede downstream performance. To address this limitation, we develop a training strategy that interleaves task-execution rollouts and exploration rollouts, with each type of rollout optimized by its corresponding verifiable reward. Building on this training strategy, we propose the Explore-then-Act paradigm, which decouples information-gathering from task execution: agents first utilize an interaction budget to acquire grounded environmental knowledge, then leverage it for task resolution. Our results demonstrate that learning to systematically explore is imperative for building generalizable and real-world-ready agents.

cs.AI↗
arXiv2026

A General Framework for Multimodal LLM-Based Multimedia Understanding in Large-Scale Recommendation Systems

Conventional recommendation systems frequently fail to fully exploit the high-dimensional semantic signals inherent in multimedia content, thereby limiting the fidelity of user preference modeling. While Multimodal Large Language Models (MM-LLMs) offer robust mechanisms for interpreting such complex data, their integration into latency-constrained, industrial-scale architectures remains a significant challenge. To address this, we propose a generalized framework for MM-LLM-driven multimedia understanding. Our methodology employs a tripartite architecture encompassing content interpretation, representation extraction, and systematic pipeline integration, instantiated via a LLaMA2-based model that generates descriptive captions subsequently ingested as tokenized categorical features. Empirical evaluation demonstrates the efficacy of this approach, yielding a $0.35\%$ increase in offline AUC and a $0.02\%$ improvement in online metrics at scale, substantiating the practical viability of leveraging MM-LLMs to enhance large-scale recommendation performance.

cs.IR↗
arXiv2026

Objective Shaping with Hard Negatives: Windowed Partial AUC Optimization for RL-based LLM Recommenders

Reinforcement learning (RL) effectively optimizes Large Language Model (LLM)-based recommenders by contrasting positive and negative items. Empirically, training with beam-search negatives consistently outperforms random negatives, yet the mechanism is not well understood. We address this gap by analyzing the induced optimization objective and show that: (i) Under binary reward feedback, optimizing LLM recommenders with Group Relative Policy Optimization (GRPO) is theoretically equivalent to maximizing the Area Under the ROC Curve (AUC), which is often misaligned with Top-$K$ recommendation; and (ii) Replacing random negatives with beam-search negatives reshapes the objective toward partial AUC, improving alignment with Top-$K$ metrics. Motivated by this perspective, we introduce Windowed Partial AUC (WPAUC), which constrains the false positive rate (FPR) to a window [$\alpha,\alpha+d$] to more directly align with Top-$K$ metrics. We further propose an efficient Threshold-Adjusted Windowed reweighting (TAWin) RL method for its optimization, enabling explicit control over the targeted Top-$K$ performance. Experiments on four real-world datasets validate the theory and deliver consistent state-of-the-art performance.

cs.IR↗
arXiv2026

AJ-Bench: Benchmarking Agent-as-a-Judge for Environment-Aware Evaluation

As reinforcement learning continues to scale the training of large language model-based agents, reliably verifying agent behaviors in complex environments has become increasingly challenging. Existing approaches rely on rule-based verifiers or LLM-as-a-Judge models, which struggle to generalize beyond narrow domains. Agent-as-a-Judge addresses this limitation by actively interacting with environments and tools to acquire verifiable evidence, yet its capabilities remain underexplored. We introduce a benchmark AJ-Bench to systematically evaluate Agent-as-a-Judge across three domains-search, data systems, and graphical user interfaces-comprising 155 tasks and 516 annotated trajectories. The benchmark comprehensively assesses judge agents' abilities in information acquisition, state verification, and process verification. Experiments demonstrate consistent performance gains over LLM-as-a-Judge baselines, while also revealing substantial open challenges in agent-based verification. Our data and code are available at https://aj-bench.github.io/.

cs.AI↗
arXiv2026

LongCat-Flash-Thinking-2601 Technical Report

We introduce LongCat-Flash-Thinking-2601, a 560-billion-parameter open-source Mixture-of-Experts (MoE) reasoning model with superior agentic reasoning capability. LongCat-Flash-Thinking-2601 achieves state-of-the-art performance among open-source models on a wide range of agentic benchmarks, including agentic search, agentic tool use, and tool-integrated reasoning. Beyond benchmark performance, the model demonstrates strong generalization to complex tool interactions and robust behavior under noisy real-world environments. Its advanced capability stems from a unified training framework that combines domain-parallel expert training with subsequent fusion, together with an end-to-end co-design of data construction, environments, algorithms, and infrastructure spanning from pre-training to post-training. In particular, the model's strong generalization capability in complex tool-use are driven by our in-depth exploration of environment scaling and principled task construction. To optimize long-tailed, skewed generation and multi-turn agentic interactions, and to enable stable training across over 10,000 environments spanning more than 20 domains, we systematically extend our asynchronous reinforcement learning framework, DORA, for stable and efficient large-scale multi-environment training. Furthermore, recognizing that real-world tasks are inherently noisy, we conduct a systematic analysis and decomposition of real-world noise patterns, and design targeted training procedures to explicitly incorporate such imperfections into the training process, resulting in improved robustness for real-world applications. To further enhance performance on complex reasoning tasks, we introduce a Heavy Thinking mode that enables effective test-time scaling by jointly expanding reasoning depth and width through intensive parallel thinking.

cs.AI↗
arXiv2026

FMVP: Masked Flow Matching for Adversarial Video Purification

Video recognition models remain vulnerable to adversarial attacks, while existing diffusion-based purification methods suffer from inefficient sampling and curved trajectories. Directly regressing clean videos from adversarial inputs often fails to recover faithful content due to the subtle nature of perturbations; this necessitates physically shattering the adversarial structure. Therefore, we propose Flow Matching for Adversarial Video Purification FMVP. FMVP physically shatters global adversarial structures via a masking strategy and reconstructs clean video dynamics using Conditional Flow Matching (CFM) with an inpainting objective. To further decouple semantic content from adversarial noise, we design a Frequency-Gated Loss (FGL) that explicitly suppresses high-frequency adversarial residuals while preserving low-frequency fidelity. We design Attack-Aware and Generalist training paradigms to handle known and unknown threats, respectively. Extensive experiments on UCF-101 and HMDB-51 demonstrate that FMVP outperforms state-of-the-art methods (DiffPure, Defense Patterns (DP), Temporal Shuffling (TS) and FlowPure), achieving robust accuracy exceeding 87% against PGD and 89% against CW attacks. Furthermore, FMVP demonstrates superior robustness against adaptive attacks (DiffHammer) and functions as a zero-shot adversarial detector, attaining AUC-ROC scores of 0.98 for PGD and 0.79 for highly imperceptible CW attacks.

cs.CV↗
arXiv2025

TROJail: Trajectory-Level Optimization for Multi-Turn Large Language Model Jailbreaks with Process Rewards

Large language models have seen widespread adoption, yet they remain vulnerable to multi-turn jailbreak attacks, threatening their safe deployment. This has led to the task of training automated multi-turn attackers to probe model safety vulnerabilities. However, existing approaches typically rely on turn-level optimization, which is insufficient for learning long-term attack strategies. To bridge this gap, we formulate this task as a multi-turn reinforcement learning problem, directly optimizing the harmfulness of the final-turn response as the outcome reward. To address the sparse supervision of the outcome reward, we introduce TROJail, which employs two process rewards to evaluate the utility of intermediate prompts and integrate them into advantage estimation. These rewards (1) penalize overly harmful prompts that trigger the model's refusal mechanism, and (2) encourage steering the semantic relevance of responses toward the targeted harmful content. Experimental results show improved attack success rates across multiple models and benchmarks, highlighting the effectiveness of our approach. The code is available at https://github.com/xxiqiao/TROJail. Warning: This paper contains examples of harmful content.

cs.AI↗
arXiv2025

MGFRec: Towards Reinforced Reasoning Recommendation with Multiple Groundings and Feedback

The powerful reasoning and generative capabilities of large language models (LLMs) have inspired researchers to apply them to reasoning-based recommendation tasks, which require in-depth reasoning about user interests and the generation of recommended items. However, previous reasoning-based recommendation methods have typically performed inference within the language space alone, without incorporating the actual item space. This has led to over-interpreting user interests and deviating from real items. Towards this research gap, we propose performing multiple rounds of grounding during inference to help the LLM better understand the actual item space, which could ensure that its reasoning remains aligned with real items. Furthermore, we introduce a user agent that provides feedback during each grounding step, enabling the LLM to better recognize and adapt to user interests. Comprehensive experiments conducted on three Amazon review datasets demonstrate the effectiveness of incorporating multiple groundings and feedback. These findings underscore the critical importance of reasoning within the actual item space, rather than being confined to the language space, for recommendation tasks.

cs.IR↗
arXiv2025

AppAgent-Pro: A Proactive GUI Agent System for Multidomain Information Integration and User Assistance

Large language model (LLM)-based agents have demonstrated remarkable capabilities in addressing complex tasks, thereby enabling more advanced information retrieval and supporting deeper, more sophisticated human information-seeking behaviors. However, most existing agents operate in a purely reactive manner, responding passively to user instructions, which significantly constrains their effectiveness and efficiency as general-purpose platforms for information acquisition. To overcome this limitation, this paper proposes AppAgent-Pro, a proactive GUI agent system that actively integrates multi-domain information based on user instructions. This approach enables the system to proactively anticipate users' underlying needs and conduct in-depth multi-domain information mining, thereby facilitating the acquisition of more comprehensive and intelligent information. AppAgent-Pro has the potential to fundamentally redefine information acquisition in daily life, leading to a profound impact on human society. Our code is available at: https://github.com/LaoKuiZe/AppAgent-Pro. The demonstration video could be found at: https://www.dropbox.com/scl/fi/hvzqo5vnusg66srydzixo/AppAgent-Pro-demo-video.mp4?rlkey=o2nlfqgq6ihl125mcqg7bpgqu&st=d29vrzii&dl=0.

cs.AI↗
arXiv2025

Exploring Stiffness Gradient Effects in Magnetically Induced Metamorphic Materials via Continuum Simulation and Validation

Magnetic soft continuum robots are capable of bending with remote control in confined space environments, and they have been applied in various bioengineering contexts. As one type of ferromagnetic soft continuums, the Magnetically Induced Metamorphic Materials (MIMMs)-based continuum (MC) exhibits similar bending behaviors. Based on the characteristics of its base material, MC is flexible in modifying unit stiffness and convenient in molding fabrication. However, recent studies on magnetic continuum robots have primarily focused on one or two design parameters, limiting the development of a comprehensive magnetic continuum bending model. In this work, we constructed graded-stiffness MCs (GMCs) and developed a numerical model for GMCs' bending performance, incorporating four key parameters that determine their performance. The simulated bending results were validated with real bending experiments in four different categories: varying magnetic field, cross-section, unit stiffness, and unit length. The graded-stiffness design strategy applied to GMCs prevents sharp bending at the fixed end and results in a more circular curvature. We also trained an expansion model for GMCs' bending performance that is highly efficient and accurate compared to the simulation process. An extensive library of bending prediction for GMCs was built using the trained model.

cs.RO↗
arXiv2025

VisualTrap: A Stealthy Backdoor Attack on GUI Agents via Visual Grounding Manipulation

Graphical User Interface (GUI) agents powered by Large Vision-Language Models (LVLMs) have emerged as a revolutionary approach to automating human-machine interactions, capable of autonomously operating personal devices (e.g., mobile phones) or applications within the device to perform complex real-world tasks in a human-like manner. However, their close integration with personal devices raises significant security concerns, with many threats, including backdoor attacks, remaining largely unexplored. This work reveals that the visual grounding of GUI agent-mapping textual plans to GUI elements-can introduce vulnerabilities, enabling new types of backdoor attacks. With backdoor attack targeting visual grounding, the agent's behavior can be compromised even when given correct task-solving plans. To validate this vulnerability, we propose VisualTrap, a method that can hijack the grounding by misleading the agent to locate textual plans to trigger locations instead of the intended targets. VisualTrap uses the common method of injecting poisoned data for attacks, and does so during the pre-training of visual grounding to ensure practical feasibility of attacking. Empirical results show that VisualTrap can effectively hijack visual grounding with as little as 5% poisoned data and highly stealthy visual triggers (invisible to the human eye); and the attack can be generalized to downstream tasks, even after clean fine-tuning. Moreover, the injected trigger can remain effective across different GUI environments, e.g., being trained on mobile/web and generalizing to desktop environments. These findings underscore the urgent need for further research on backdoor attack risks in GUI agents.

cs.CL↗
arXiv2025

Reinforcement Fine-Tuning for Reasoning towards Multi-Step Multi-Source Search in Large Language Models

Large language models (LLMs) can face factual limitations when responding to time-sensitive queries about recent events that arise after their knowledge thresholds in the training corpus. Existing search-augmented approaches fall into two categories, each with distinct limitations: multi-agent search frameworks incur substantial computational overhead by separating search planning and response synthesis across multiple LLMs, while single-LLM tool-calling methods restrict themselves to sequential planned, single-query searches from sole search sources. We present Reasoning-Search (R-Search), a single-LLM search framework that unifies multi-step planning, multi-source search execution, and answer synthesis within one coherent inference process. Innovatively, it structure the output into four explicitly defined components, including reasoning steps that guide the search process ( ), a natural-language directed acyclic graph that represents the search plans with respect to diverse sources ( ), retrieved results from executing the search plans ( ), and synthesized final answers ( ). To enable effective generation of these structured outputs, we propose a specialized Reinforcement Fine-Tuning (ReFT) method based on GRPO, together with a multi-component reward function that optimizes LLM's answer correctness, structural validity of the generated DAG, and adherence to the defined output format. Experimental evaluation on FinSearchBench-24, SearchExpertBench-25, and seven Q and A benchmarks demonstrates that R-Search outperforms state-of-the-art methods, while achieving substantial efficiency gains through 70% reduction in context token usage and approximately 50% decrease in execution latency. Code is available at https://github.com/wentao0429/Reasoning-search.

cs.IR↗
arXiv2025

K-order Ranking Preference Optimization for Large Language Models

To adapt large language models (LLMs) to ranking tasks, existing list-wise methods, represented by list-wise Direct Preference Optimization (DPO), focus on optimizing partial-order or full-order list ranking consistency for LLMs to enhance their ranking abilities. However, we argue that optimizing top-K ranking consistency could be more appropriate for real-world applications. There are two main reasons: (1) users are typically concerned with only the top-K results, making top-K ranking more important, and (2) tail items often lack precise feedback, making top-K ranking more reliable. Based on this, we propose K-order Ranking Preference Optimization (KPO) by extending the DPO's Plackett-Luce model to accommodate top-K rankings. Additionally, recognizing that the number of important items can vary across queries, we extend KPO to dynamically determine appropriate K for different samples and introduce a curriculum learning strategy to boost training efficiency. Extensive experiments demonstrate the effectiveness of KPO, highlighting its high sample efficiency and robustness to noise. The code is available at https://github.com/Lanyu0303/KPO.

cs.IR↗
arXiv2025

Fine-grained List-wise Alignment for Generative Medication Recommendation

Accurate and safe medication recommendations are critical for effective clinical decision-making, especially in multimorbidity cases. However, existing systems rely on point-wise prediction paradigms that overlook synergistic drug effects and potential adverse drug-drug interactions (DDIs). We propose FLAME, a fine-grained list-wise alignment framework for large language models (LLMs), enabling drug-by-drug generation of drug lists. FLAME formulates recommendation as a sequential decision process, where each step adds or removes a single drug. To provide fine-grained learning signals, we devise step-wise Group Relative Policy Optimization (GRPO) with potential-based reward shaping, which explicitly models DDIs and optimizes the contribution of each drug to the overall prescription. Furthermore, FLAME enhances patient modeling by integrating structured clinical knowledge and collaborative information into the representation space of LLMs. Experiments on benchmark datasets demonstrate that FLAME achieves state-of-the-art performance, delivering superior accuracy, controllable safety-accuracy trade-offs, and strong generalization across diverse clinical scenarios. Our code is available at https://github.com/cxfann/Flame.

cs.LG↗
arXiv2025

Process-Supervised LLM Recommenders via Flow-guided Tuning

While large language models (LLMs) are increasingly adapted for recommendation systems via supervised fine-tuning (SFT), this approach amplifies popularity bias due to its likelihood maximization objective, compromising recommendation diversity and fairness. To address this, we present Flow-guided fine-tuning recommender (Flower), which replaces SFT with a Generative Flow Network (GFlowNet) framework that enacts process supervision through token-level reward propagation. Flower's key innovation lies in decomposing item-level rewards into constituent token rewards, enabling direct alignment between token generation probabilities and their reward signals. This mechanism achieves three critical advancements: (1) popularity bias mitigation and fairness enhancement through empirical distribution matching, (2) preservation of diversity through GFlowNet's proportional sampling, and (3) flexible integration of personalized preferences via adaptable token rewards. Experiments demonstrate Flower's superior distribution-fitting capability and its significant advantages over traditional SFT in terms of accuracy, fairness, and diversity, highlighting its potential to improve LLM-based recommendation systems. The implementation is available via https://github.com/MrPeach0301/Flower

cs.IR↗
arXiv2025

Self-Improvement Towards Pareto Optimality: Mitigating Preference Conflicts in Multi-Objective Alignment

Multi-Objective Alignment (MOA) aims to align LLMs' responses with multiple human preference objectives, with Direct Preference Optimization (DPO) emerging as a prominent approach. However, we find that DPO-based MOA approaches suffer from widespread preference conflicts in the data, where different objectives favor different responses. This results in conflicting optimization directions, hindering the optimization on the Pareto Front. To address this, we propose to construct Pareto-optimal responses to resolve preference conflicts. To efficiently obtain and utilize such responses, we propose a self-improving DPO framework that enables LLMs to self-generate and select Pareto-optimal responses for self-supervised preference alignment. Extensive experiments on two datasets demonstrate the superior Pareto Front achieved by our framework compared to various baselines. Code is available at https://github.com/zyttt-coder/SIPO.

cs.LG↗