Searcharxiv⌕ Search

arXiv subjects

Chengyu Wang

Publications and source records attributed to Chengyu Wang.

At least 19 recordsLinked to original sources
arXiv2026

Bidirectional Context Self-Distillation for Reinforcement Learning of Skill-Based LLM Agents

External natural-language skills provide large language model (LLM) agents with reusable and editable guidance for solving complex tasks. Yet their effectiveness depends not only on skill quality, but also on whether the policy can translate the provided guidance into appropriate actions. However, methods specifically designed to improve this skill-utilization ability remain largely underexplored. In practice, skill-based agents are commonly trained with reinforcement learning objectives centered on task-level rewards, which offer limited supervision and struggle to capture subtle differences in how effectively the policy uses the provided skills. We propose BCSD (Bidirectional Context Self-Distillation), a framework that combines self-distillation with reinforcement learning to train LLM agents to use external skills more effectively. Unlike prior self-distillation methods that rely on a single privileged context, BCSD evaluates each trajectory from two complementary skill-context views. The augmented view introduces higher-level Meta-Skill guidance, while the reduced view prunes general guidance to highlight task-specific skills. Their complementary token-level signals are combined to rescale the RL advantage. Experiments on ALFWorld and WebShop demonstrate that BCSD achieves the strongest overall performance across model scales, enabling agents to utilize external skills more effectively. Ablation studies further verify the complementary contributions of the augmented and reduced context views. Code will be released to ensure full reproducibility.

cs.AI↗
arXiv2026

M$^3$Prune: Hierarchical Collaborative Pruning for Efficient Multi-Modal Multi-Agent Retrieval-Augmented Generation

Recent advances in multi-modal retrieval-augmented generation (mRAG), which augments multi-modal large language models (MLLMs) with external knowledge, have shown that collective intelligence from multiple agents can outperform a single model through effective communication. Despite their strong performance, existing multi-agent systems incur substantial token overhead and computational cost, posing challenges for large-scale deployment. To address these issues, we propose a Multi-Modal Multi-agent hierarchical communication graph PRUNING framework, termed M3Prune. M3Prune eliminates redundant communication edges both across and within modalities, improving the trade-off between task performance and token overhead. Specifically, M3Prune first performs intra-modal graph sparsification in the textual and visual modalities to identify task-critical communication links. It then constructs an inter-modal communication graph and sparsifies cross-modal connections while encouraging consistent cross-modal reasoning through a modality alignment score. Finally, it progressively prunes redundant edges to obtain an efficient hierarchical topology. Extensive experiments on both general-domain and domain-specific mRAG benchmarks show that M3Prune consistently outperforms single-agent and strong multi-agent mRAG systems while signifi- cantly improving token efficiency.

cs.MM↗
arXiv2026

Beyond Solution-Centric Search: Adaptive Inquiry and Knowledge Revision for Autonomous ML Engineering

Long-horizon autonomous research tasks such as machine learning engineering require systems to make interdependent decisions under a limited budget. Existing LLM-based agents typically organize candidate-solution improvement through tree, graph, or chain structures, meaning that the search process determines how information is acquired and managed. We call this design solution-centric search and propose instead the information paradigm, in which an evolving information state represents the system's understanding of the task and guides solution improvement. We instantiate this paradigm in Iris, an inquiry-revision loop. For information acquisition, Iris generates local action plans from the current information state and uses epistemic actions to probe decision-critical unknowns without modifying the retained solution. For information management, Iris synthesizes observations across experiments into task knowledge composed of revisable claims with explicit scope and status. It updates this knowledge as new evidence arrives and constructs each decision context from raw evidence, structured summaries, or task knowledge at the required level of detail. On MLE-Bench, Iris attains a 64.9% any-medal rate under a 12-hour budget, the highest among compared systems. Across four tasks spanning harness engineering and model post-training, Iris also demonstrates cross-domain generalization.

cs.AI↗
arXiv2026

MerchantBench: Benchmarking LLM Agents for Long-Term Coherence in E-Commerce Operations

Large language model agents are increasingly evaluated as autonomous tool users, yet most benchmarks focus on bounded tasks with immediate success criteria. Real-world deployments often require Long-Term Coherence, the capacity to preserve purposeful behavior across extended horizons while adapting decisions to accumulated evidence. Evaluating this capacity requires a persistent environment in which actions constrain future choices, feedback arrives at heterogeneous delays, and incoherent behavior produces measurable cumulative effects. Seller-side e-commerce provides a suitable setting for this evaluation through recurrent and interdependent decisions over Product Sourcing, Listing and Pricing Control, Cash-Flow Management, and Mixed-Latency Feedback Adaptation. We introduce MerchantBench, a 365-day order-level simulation grounded in 98,843 real e-commerce product records and equipped with 26 tools for agent interaction. MerchantBench couples promptly observable Upstream Supplier Events with delayed Downstream Order Outcomes, requiring agents to follow individual order lifecycles and revisit earlier decisions. We evaluate eight LLMs under two agent frameworks in 48 runs, each spanning 365 simulated days. Our results reveal a substantial gap between even the latest LLMs and human participants, with the best LLM configuration attaining only 27.3\% of the mean final net assets achieved by human participants.

cs.AI↗
arXiv2026

Observation of even-denominator fractional quantum Hall states at $\nu = 3/4$ and 5/4 in the lowest Landau level

Two-dimensional electron systems (2DESs) confined to wide GaAs quantum wells provide a unique platform to study exotic fractional quantum Hall states (FQHSs) because the 2DES has a bilayer charge distribution with significant interlayer tunneling. Precise control over the 2DES density allows the tuning of the interlayer tunneling over a wide range. Here, we present our discovery of new even-denominator FQHSs in the lowest Landau level (orbital index \textit{N} = 0) at filling factors $\nu = 3/4$ and 5/4 in an ultrahigh-quality 2DES confined to a 72.5-nm-wide GaAs quantum well. The ground states at $\nu = 3/4$ and 5/4 both evolve from composite fermion Fermi seas to FQHSs as the density is raised so that interlayer tunneling is sufficiently reduced and the 2DES becomes two-component, signaled by the behavior of the FQHSs flanking $\nu = 3/4$ and 5/4. The two-component nature of the $\nu=3/4$ and 5/4 FQHSs is also evident from their extreme sensitivity to the bilayer charge distribution symmetry: both states disappear quickly when the charge distribution is made asymmetric by only $\simeq 2\%$. We find a natural explanation for the 3/4 and 5/4 FQHSs in terms of two states linked by particle-hole symmetry, and using the Scarola-Jain bilayer composite fermion framework which is a generalization of the well-known, two-component, Halperin state ($\Psi_{331}$ state). Our observations elucidate the crucial role of competing energy and length scales in wide quantum wells in stabilizing new ground states.

cond-mat.mes-hall↗
arXiv2026

Taming "Zombie'' Agents: A Markov State-Aware Framework for Resilient Multi-Agent Evolution

Recent advancements in LLM-based multi-agent systems have demonstrated remarkable collaborative capabilities across complex tasks. To improve overall efficiency, existing methods often rely on aggressive graph evolution among agents (e.g., node or edge pruning), which risks prematurely discarding valuable agents due to transient issues such as hallucinations or temporary knowledge gaps. However, such hard pruning overlooks the potential for ``zombie'' agents to recover and contribute in subsequent discussion rounds. In this paper, we propose AgentRevive, a Markov state-aware framework for resilient multi-agent evolution. Our approach dynamically manages agent collaboration through soft state transitions, implemented via two key components: (1) State-Aware Policy Learning: Agent states are divided into ``Active'', ``Standby'', and ``Terminated'' states, selectively propagating messages based on agent memory. The policy employs a risk estimator to optimize agent state transitions by assessing hallucination risk, minimizing the influence of unreliable nodes while safeguarding valuable ones. (2) State-Aware Edge Optimization: Subgraph edges are pruned according to states learned from the policy, permanently removing ``Terminated'' nodes and retaining ``Standby'' nodes for subsequent rounds to assess their potential future contributions. Extensive experiments on general reasoning, domain-specific, and hallucination challenge tasks show that our method consistently outperforms strong baselines and significantly reduces token consumption through state-aware agent scheduling.

cs.CL↗
arXiv2026

AMATA: Adaptive Multi-Agent Trajectory Alignment for Knowledge-Intensive Question Answering

Despite substantial advances in large language models (LLMs), generating factually consistent responses for knowledge-intensive question answering remains challenging. These difficulties are primarily due to hallucinations and the limitations of LLMs in bridging long-tail knowledge gaps. To address this, we propose AMATA, an Adaptive Multi-Agent Trajectory Alignment framework that dynamically integrates external knowledge to improve response interpretability and factual grounding. Our architecture leverages six specialized agents that collaboratively perform structured actions for complex question reasoning. We formalize multi-agent collaboration with external tools as a trajectory preference alignment problem, incorporating question-aware agent customization and inter-agent preference harmonization. AMATA introduces two principal innovations: (1) Intra-Trajectory Preference Learning, which learns objective-oriented preferences to prioritize critical agents, and (2) Inter-Agent Dependency Learning, which captures cross-agent tool dependencies through a novel dependency-aware direct preference optimization technique. Empirical results show that AMATA consistently outperforms baseline approaches, knowledge-augmented frameworks, and LLM-based trajectory systems on five established knowledge-intensive QA benchmarks. Further analysis demonstrates the efficiency of our method in reducing token consumption.

cs.CL↗
arXiv2026

Learning Transferable Topology Priors for Multi-Agent LLM Collaboration Across Domains

Large language model (LLM)-based multi-agent systems have shown strong potential for complex reasoning by coordinating specialized agents through structured communication. However, existing topology-evolution methods typically construct or optimize a collaboration topology for each query from scratch, leading to substantial online search overhead, high inference-time token consumption, and limited scalability in multi-domain settings. We propose TopoPrior, a framework for learning transferable topology priors for multi-agent LLM collaboration across domains. Rather than repeatedly searching for effective collaboration structures online, TopoPrior learns reusable topology priors from reference collaboration graphs collected offline from multiple domains and uses them to generate query-conditioned initial collaboration graphs for downstream refinement. By shifting part of topology search from per-query online optimization to offline prior learning, TopoPrior amortizes search cost while remaining compatible with existing topology-evolution backbones. Technically, TopoPrior contains two key components. First, a transferable topology prior learning module employs a conditional variational graph framework to capture reusable structural regularities across domains in a latent space. Second, a query-conditioned latent adaptation module introduces adversarial alignment to reduce unnecessary domain discrepancy while preserving query-relevant structural variation. Experiments on multi-domain reasoning benchmarks show that TopoPrior consistently improves several heterogeneous topology-evolution backbones while reducing online inference-time token usage, with only modest additional trainable parameters. These results suggest that transferable topology initialization is an effective and lightweight mechanism for improving the efficiency of multi-agent LLM collaboration across domains.

cs.CL↗
arXiv2026

OmniThoughtVis: A Scalable Distillation Pipeline for Deployable Multimodal Reasoning Models

Recent multimodal large language models (MLLMs) have shown strong chain-of-thought (CoT) reasoning ability on vision-language tasks, but their direct deployment in real-world systems is often limited by latency and resource constraints. In practice, smaller MLLMs are preferred for online serving, yet their reasoning performance is bottlenecked by the lack of large-scale, high-quality multimodal CoT supervision. In this paper, we present OmniThoughtVis, a scalable data curation and distillation pipeline for transferring multimodal reasoning capabilities from high-capacity teacher models to smaller, deployment-oriented MLLMs. Starting from a diverse open-source seed pool, our pipeline generates structured CoT traces and performs joint annotation of reasoning difficulty, answer quality, and semantic task tags. To maintain data quality at scale, we combine rule-based filtering, difficulty-aware selection, and tag-based diversity sampling, resulting in a curated corpus of 1.8M samples that supports controllable subset construction for downstream training. We use OmniThoughtVis to distill Qwen3-VL models from 2B to 8B parameters and evaluate them on nine multimodal reasoning benchmarks. The resulting distilled models show consistent gains across model scales, including improvements of up to +16.8 points on MathVerse and +5.6 points on MMMU-Pro for the 4B model. Notably, the distilled 4B model matches or surpasses the undistilled 8B baseline on several tasks, highlighting the practical value of scalable reasoning distillation for deployment-oriented MLLMs.

cs.CL↗
arXiv2026

AgenticQwen: Training Small Agentic Language Models with Dual Data Flywheels for Industrial-Scale Tool Use

Modern industrial applications increasingly demand language models that act as agents, capable of multi-step reasoning and tool use in real-world settings. These tasks are typically performed under strict cost and latency constraints, making small agentic models highly desirable. In this paper, we introduce the AgenticQwen family of models, trained via multi-round reinforcement learning (RL) on synthetic data and a limited amount of open-source data. Our training framework combines reasoning RL and agentic RL with dual data flywheels that automatically generate increasingly challenging tasks. The reasoning flywheel increases task difficulty by learning from errors, while the agentic flywheel expands linear workflows into multi-branch behavior trees that better reflect the decision complexity of real-world applications. We validate AgenticQwen on public benchmarks and in an industrial agent system. The models achieve strong performance on multiple agentic benchmarks, and in our industrial agent system, close the gap with much larger models on search and data analysis tasks. Model checkpoints and part of the synthetic data: https://huggingface.co/collections/alibaba-pai/agenticqwen. Data synthesis and RL training code: https://github.com/haruhi-sudo/data_synth_and_rl. The data synthesis pipeline is also integrated into EasyDistill: https://github.com/modelscope/easydistill.

cs.CL↗
arXiv2026

Mock Worlds, Real Skills: Building Small Agentic Language Models with Synthetic Tasks, Simulated Environments, and Rubric-Based Rewards

Small LLMs often struggle to match the agentic capabilities of large, costly models. While reinforcement learning can help, progress has been limited by two structural bottlenecks: existing open-source agentic training data are narrow in task variety and easily solved; real-world APIs lack diversity and are unstable for large-scale reinforcement learning rollout processes. We address these challenges with SYNTHAGENT, a framework that jointly synthesizes diverse tool-use training data and simulates complete environments. Specifically, a strong teacher model creates novel tasks and tool ecosystems, then rewrites them into intentionally underspecified instructions. This compels agents to actively query users for missing details. When handling synthetic tasks, an LLM-based user simulator provides user-private information, while a mock tool system delivers stable tool responses. For rewards, task-level rubrics are constructed based on required subgoals, user-agent interactions, and forbidden behaviors. Across 14 challenging datasets in math, search, and tool use, models trained on our synthetic data achieve substantial gains, with small models outperforming larger baselines.

cs.CL↗
arXiv2026

VTC-R1: Vision-Text Compression for Efficient Long-Context Reasoning

Long-context reasoning has significantly empowered large language models (LLMs) to tackle complex tasks, yet it introduces severe efficiency bottlenecks due to the computational complexity. Existing efficient approaches often rely on complex additional training or external models for compression, which limits scalability and discards critical fine-grained information. In this paper, we propose VTC-R1, a new efficient reasoning paradigm that integrates vision-text compression into the reasoning process. Instead of processing lengthy textual traces, VTC-R1 renders intermediate reasoning segments into compact images, which are iteratively fed back into vision-language models as "optical memory." We construct a training dataset based on OpenR1-Math-220K achieving 3.4x token compression and fine-tune representative VLMs-Glyph and Qwen3-VL. Extensive experiments on benchmarks such as MATH500, AIME25, AMC23 and GPQA-D demonstrate that VTC-R1 consistently outperforms standard long-context reasoning. Furthermore, our approach significantly improves inference efficiency, achieving 2.7x speedup in end-to-end latency, highlighting its potential as a scalable solution for reasoning-intensive applications. Our code is available at https://github.com/w-yibo/VTC-R1.

cs.CL↗
arXiv2026

Where Does Vision Meet Language? Understanding and Refining Visual Fusion in MLLMs via Contrastive Attention

Multimodal Large Language Models (MLLMs) have achieved remarkable progress in vision-language understanding, yet how they internally integrate visual and textual information remains poorly understood. To bridge this gap, we perform a systematic layer-wise masking analysis across multiple architectures, revealing how visual-text fusion evolves within MLLMs. The results show that fusion emerges at several specific layers rather than being uniformly distributed across the network, and certain models exhibit a late-stage "review" phenomenon where visual signals are reactivated before output generation. Besides, we further analyze layer-wise attention evolution and observe persistent high-attention noise on irrelevant regions, along with gradually increasing attention on text-aligned areas. Guided by these insights, we introduce a training-free contrastive attention framework that models the transformation between early fusion and final layers to highlight meaningful attention shifts. Extensive experiments across various MLLMs and benchmarks validate our analysis and demonstrate that the proposed approach improves multimodal reasoning performance.

cs.CV↗
arXiv2025

An Information-Theoretic Framework for Robust Large Language Model Editing

Large Language Models (LLMs) have become indispensable tools in science, technology, and society, enabling transformative advances across diverse fields. However, errors or outdated information within these models can undermine their accuracy and restrict their safe deployment. Developing efficient strategies for updating model knowledge without the expense and disruption of full retraining remains a critical challenge. Current model editing techniques frequently struggle to generalize corrections beyond narrow domains, leading to unintended consequences and limiting their practical impact. Here, we introduce a novel framework for editing LLMs, grounded in information bottleneck theory. This approach precisely compresses and isolates the essential information required for generalizable knowledge correction while minimizing disruption to unrelated model behaviors. Building upon this foundation, we present the Information Bottleneck Knowledge Editor (IBKE), which leverages compact latent representations to guide gradient-based updates, enabling robust and broadly applicable model editing. We validate IBKE's effectiveness across multiple LLM architectures and standard benchmark tasks, demonstrating state-of-the-art accuracy and improved generality and specificity of edits. These findings establish a theoretically principled and practical paradigm for open-domain knowledge editing, advancing the utility and trustworthiness of LLMs in real-world applications.

cs.CL↗
arXiv2025

Observation of a Stripe/nematic Phase of Composite Fermions

Electronic stripe/nematic phases are fascinating strongly-correlated states characterized by spontaneous rotational symmetry breaking. In the quantum Hall regime, such phases typically emerge at half-filled, high-orbital-index ($N\geq2$) Landau levels (LLs) where the short-range Coulomb interaction is softened by the nodes of electron wave functions. In the lowest ($N=0$) LLs, these phases are not expected. Instead, composite fermion (CF) liquids and fractional quantum Hall states, which are well explained in the picture of weakly interacting CF quasiparticles, are favored. Here we report the observation of an unexpected stripe/nematic phase in the \textit{lowest} LL at filling factor $\nu=5/8$ in ultrahigh-quality GaAs two-dimensional \textit{hole} systems, evinced by a pronounced in-plane transport anisotropy. Remarkably, $\nu=5/8$ can be mapped to a half-filled, high-index CF LL ($N_{\text{CF}}=2$), analogous to the $N=2$ hole LL. Our finding signals a novel stripe/nematic phase of CFs, driven by the residual long-range interaction among these emergent quasiparticles. This phase is surprisingly robust, surviving up to $\sim$100 mK. Its absence in electron-type systems suggests that severe LL mixing stemming from the large hole effective mass and non-linear LL fan diagram plays a crucial role in modifying the CF-CF interaction.

cond-mat.mes-hall↗
arXiv2025

M$^3$Prune: Hierarchical Communication Graph Pruning for Efficient Multi-Modal Multi-Agent Retrieval-Augmented Generation

Recent advancements in multi-modal retrieval-augmented generation (mRAG), which enhance multi-modal large language models (MLLMs) with external knowledge, have demonstrated that the collective intelligence of multiple agents can significantly outperform a single model through effective communication. Despite impressive performance, existing multi-agent systems inherently incur substantial token overhead and increased computational costs, posing challenges for large-scale deployment. To address these issues, we propose a novel Multi-Modal Multi-agent hierarchical communication graph PRUNING framework, termed M$^3$Prune. Our framework eliminates redundant edges across different modalities, achieving an optimal balance between task performance and token overhead. Specifically, M$^3$Prune first applies intra-modal graph sparsification to textual and visual modalities, identifying the edges most critical for solving the task. Subsequently, we construct a dynamic communication topology using these key edges for inter-modal graph sparsification. Finally, we progressively prune redundant edges to obtain a more efficient and hierarchical topology. Extensive experiments on both general and domain-specific mRAG benchmarks demonstrate that our method consistently outperforms both single-agent and robust multi-agent mRAG systems while significantly reducing token consumption.

cs.AI↗
arXiv2025

Thinking with DistilQwen: A Tale of Four Distilled Reasoning and Reward Model Series

Recently, the demand for small and efficient reasoning models to support real-world applications has driven the development of knowledge distillation techniques that balance reasoning performance and inference speed. In this paper, we further extend the DistilQwen model family, initialized from the Qwen models, by introducing four model series specifically designed to meet industrial requirements. The distilled model collection comprises: (1) slow-thinking models, optimized for reasoning tasks that require high accuracy; (2) two series of adaptive-thinking models, which dynamically adjust reasoning strategies based on input tasks to maximize efficiency across diverse scenarios; and (3) distilled reward models, which enable further reinforcement learning of reasoning models using distilled knowledge. Comprehensive evaluations across multiple benchmarks demonstrate both high inference efficiency and strong reasoning performance for these models, as well as the practical utility of distilled reward models. We further show that these models support industry practitioners by providing scalable training and inference functionalities on the Alibaba Cloud PAI (Platform for Artificial Intelligence) platform.

cs.CL↗
arXiv2025

Fractional quantum Hall state at $\nu = 1/2$ with energy gap up to 6 K, and possible transition from one- to two-component state

The fractional quantum Hall state (FQHS) observed in the lowest Landau level at filling factor $\nu=1/2$ in wide quantum wells has been enigmatic for decades because the two-dimensional electron system (2DES) has a bilayer charge distribution but with significant interlayer tunneling. Of particular interest is whether the 1/2 FQHS in this system has a one-component (1C) or two-component (2C) origin; these are typically identified as the Pfaffian (non-Abelian) or the $\Psi_{331}$ (Abelian) FQHSs, respectively. We report here our experimental study of the evolution of the correlated states of an ultrahigh-quality 2DES confined to a 72.5-nm-wide GaAs quantum well. At the lowest densities, the 2DES displays only odd-denominator FQHSs, and the ground state at $\nu = 1/2$ is a composite fermion Fermi sea. As the density is increased, a FQHS emerges at $\nu = 1/2$, and becomes very strong. In a finite density range where the 1/2 FQHS is strongest, we also observe its daughter FQHSs at $\nu = 8/17$ and 7/13, consistent with the theoretically expected daughter states of a Pfaffian 1/2 FQHS. At the highest densities, the 2DES becomes 2C, signaled by the emergence of a bilayer Wigner crystal state and the transitions of FQHSs flanking $\nu=1/2$. The 1/2 FQHS remains robust near this transition and, notably, its charge transport energy gap exhibits an \textit{upward} cusp with a maximum value of about 6 K on the 1C side of the transition; this is the largest gap reported for any even-denominator FQHS. Our observation of the transition of the 2DES ground states near $\nu=1/2$ to 2C states at high densities, and our measurements of the robustness of the 1/2 FQHS against charge distribution asymmetry, suggest that the 1/2 FQHS also makes a transition from 1C to 2C. Such a transition from a non-Abelian to Abelian state can open avenues for topological quantum information and quantum criticality.

cond-mat.mes-hall↗