Chengyu Wang
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Research Interests
Chengyu Wang has a total of 177 co-authors. Here are the top co-authors by number of shared publications:
- Jun Huang
Alibaba Group, Hangzhou, China
49 shared publications - Xiaofeng He
East China Normal University, Shanghai, China
20 shared publications - Cen Chen
East China Normal University, Shanghai, China
18 shared publications - Ming Gao
East China Normal University, Shanghai, China
9 shared publications - Weining Qian
East China Normal University
9 shared publications - Zhongjie Duan
East China Normal University
9 shared publications
Latest Publications
- Open Access
- 9 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 4446-4455
From Discord to Accord: A Consensus-Driven Multi-Agent Framework for Multi-Modal Question Answering
Multi-modal large language models (MLLMs) integrate heterogeneous inputs to solve tasks that require joint reasoning across modalities, such as multi-modal question answering (MQA). Beyond conventional prompt-based inference, multi-agent frameworks can elicit complementary evidence and aggregate diverse reasoning traces to better address challenging queries. Existing multi-agent methods are largely inspired by individual-level cognitive mechanisms (e.g., fast-slow thinking and metacognition), but often under-exploit social cognitive factors that govern how groups reconcile disagreements. Moreover, while debate-style strategies can improve logical consistency, their final summarization is frequently heuristic, making it prone to overlooking factual details. We propose CRATER, a consensus-driven multi-agent framework for MQA that aligns discordant perspectives through structured interactions among heterogeneous MLLM agents. CRATER proceeds in three stages. First, in Collaborative Multi-Agent Initialization, a discussion group generates candidate opinions from diverse perspectives and modalities to cover a broad space of viewpoints. Second, Multi-Dimensional Scaling (MDS) Projection embeds these opinions into a low-dimensional geometry, facilitating the identification of polarized stances. Third, polarized opinions are reconciled via an iterative Monotonic Concession Re-discussion procedure, which progressively narrows disagreements and establishes consensus. Experiments on public MQA benchmarks spanning general capability evaluation, text recognition, and multidisciplinary domains show that CRATER consistently outperforms strong baselines. Additional analyses validate the contributions of each core component.
- 9 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 4446-4455
Open AccessFrom Discord to Accord: A Consensus-Driven Multi-Agent Framework for Multi-Modal Question Answering
Multi-modal large language models (MLLMs) integrate heterogeneous inputs to solve tasks that require joint reasoning across modalities, such as multi-modal question answering (MQA). Beyond conventional prompt-based inference, multi-agent frameworks can elicit complementary evidence and aggregate diverse reasoning traces to better address challenging queries. Existing multi-agent methods are largely inspired by individual-level cognitive mechanisms (e.g., fast-slow thinking and metacognition), but often under-exploit social cognitive factors that govern how groups reconcile disagreements. Moreover, while debate-style strategies can improve logical consistency, their final summarization is frequently heuristic, making it prone to overlooking factual details. We propose CRATER, a consensus-driven multi-agent framework for MQA that aligns discordant perspectives through structured interactions among heterogeneous MLLM agents. CRATER proceeds in three stages. First, in Collaborative Multi-Agent Initialization, a discussion group generates candidate opinions from diverse perspectives and modalities to cover a broad space of viewpoints. Second, Multi-Dimensional Scaling (MDS) Projection embeds these opinions into a low-dimensional geometry, facilitating the identification of polarized stances. Third, polarized opinions are reconciled via an iterative Monotonic Concession Re-discussion procedure, which progressively narrows disagreements and establishes consensus. Experiments on public MQA benchmarks spanning general capability evaluation, text recognition, and multidisciplinary domains show that CRATER consistently outperforms strong baselines. Additional analyses validate the contributions of each core component.
- Open Access
- 9 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 14781-14790
M 3 Prune: Hierarchical Collaborative Pruning for Efficient Multi-Modal Multi-Agent Retrieval-Augmented Generation
- 9 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 14781-14790
Open AccessM 3 Prune: Hierarchical Collaborative Pruning for Efficient Multi-Modal Multi-Agent Retrieval-Augmented Generation
- 8 October 2026
- Published by Springer Nature
- p. 213-227
ReVerTree: Recursive Verification Tree for Self-checking Mathematical Reasoning
- 8 October 2026
- Published by Springer Nature
- p. 213-227
ReVerTree: Recursive Verification Tree for Self-checking Mathematical Reasoning
- Open Access
- 8 August 2026
- Published by Association for Computing Machinery (ACM)
- p. 4195-4206
Self-Evolutionary Reinforced Knowledge Distillation for Multi-Modal Tool-Use Agents
- 8 August 2026
- Published by Association for Computing Machinery (ACM)
- p. 4195-4206
Open AccessSelf-Evolutionary Reinforced Knowledge Distillation for Multi-Modal Tool-Use Agents
- Top Cited PapersOpen Access
- 19 February 2026
- Published by Springer Nature in Frontiers of Computer Science
- Vol. 20 (11) , 2011366
A short survey on small reasoning models: training, inference, applications, and research directions
Recently, the reasoning capabilities of Large Reasoning Models (LRMs), such as DeepSeek-R1, have witnessed significant advancements through computationally intensive “slow thinking” processes. These models have demonstrated impressive performance across a variety of complex reasoning tasks. However, despite their remarkable success, LRMs come with substantial computational demands that pose considerable challenges in terms of resource consumption, scalability, and accessibility. In contrast, Small Reasoning Models (SRMs), which are often distilled from larger models, offer a more efficient alternative while still achieving competitive performance. Beyond their efficiency, SRMs frequently exhibit distinct capabilities and cognitive trajectories compared with their larger counterparts, making them particularly interesting from both practical and theoretical perspectives. In this work, we provide a timely and comprehensive survey of recently published research focused on SRMs. We first review the current landscape of SRMs. Then, we analyze diverse training paradigms and inference techniques tailored to enhance the reasoning capabilities of SRMs. Furthermore, we offer an extensive review of domain-specific applications where SRMs have been effectively leveraged. Finally, we discuss promising future research directions that aim to bridge existing gaps. By consolidating recent advances, this survey serves as an essential reference for researchers and practitioners interested in leveraging or developing SRMs to unlock advanced reasoning functionalities with improved efficiency.
- 19 February 2026
- Published by Springer Nature in Frontiers of Computer Science
- Vol. 20 (11) , 2011366
Top Cited PapersOpen AccessA short survey on small reasoning models: training, inference, applications, and research directions
Recently, the reasoning capabilities of Large Reasoning Models (LRMs), such as DeepSeek-R1, have witnessed significant advancements through computationally intensive “slow thinking” processes. These models have demonstrated impressive performance across a variety of complex reasoning tasks. However, despite their remarkable success, LRMs come with substantial computational demands that pose considerable challenges in terms of resource consumption, scalability, and accessibility. In contrast, Small Reasoning Models (SRMs), which are often distilled from larger models, offer a more efficient alternative while still achieving competitive performance. Beyond their efficiency, SRMs frequently exhibit distinct capabilities and cognitive trajectories compared with their larger counterparts, making them particularly interesting from both practical and theoretical perspectives. In this work, we provide a timely and comprehensive survey of recently published research focused on SRMs. We first review the current landscape of SRMs. Then, we analyze diverse training paradigms and inference techniques tailored to enhance the reasoning capabilities of SRMs. Furthermore, we offer an extensive review of domain-specific applications where SRMs have been effectively leveraged. Finally, we discuss promising future research directions that aim to bridge existing gaps. By consolidating recent advances, this survey serves as an essential reference for researchers and practitioners interested in leveraging or developing SRMs to unlock advanced reasoning functionalities with improved efficiency.
- Open Access
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.
Open AccessAn 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.
- Open Access
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.
Open AccessM$^3$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.
- Open Access
- 1 September 2025
- Published by International Joint Conferences on Artificial Intelligence
- Vol. 11, 10198-10206
Hallucination-Aware Prompt Optimization for Text-to-Video Synthesis
The rapid advancements in AI-generated content (AIGC) have led to extensive research and application of deep text-to-video (T2V) synthesis models, such as OpenAI's Sora. These models typically rely on high-quality prompt-video pairs and detailed text prompts for model training in order to produce high-quality videos. To boost the effectiveness of Sora-like T2V models, we introduce VidPrompter, an innovative large multi-modal model supporting T2V applications with three key functionalities: (1) generating detailed prompts from raw videos, (2) enhancing prompts from videos grounded with short descriptions, and (3) refining simple user-provided prompts to elevate T2V video quality. We train VidPrompter using a hybrid multi-task paradigm and propose the hallucination-aware direct preference optimization (HDPO) technique to improve the multi-modal, multi-task prompt optimization process. Experiments on various tasks show our method surpasses strong baselines and other competitors.
- 1 September 2025
- Published by International Joint Conferences on Artificial Intelligence
- Vol. 11, 10198-10206
Open AccessHallucination-Aware Prompt Optimization for Text-to-Video Synthesis
The rapid advancements in AI-generated content (AIGC) have led to extensive research and application of deep text-to-video (T2V) synthesis models, such as OpenAI's Sora. These models typically rely on high-quality prompt-video pairs and detailed text prompts for model training in order to produce high-quality videos. To boost the effectiveness of Sora-like T2V models, we introduce VidPrompter, an innovative large multi-modal model supporting T2V applications with three key functionalities: (1) generating detailed prompts from raw videos, (2) enhancing prompts from videos grounded with short descriptions, and (3) refining simple user-provided prompts to elevate T2V video quality. We train VidPrompter using a hybrid multi-task paradigm and propose the hallucination-aware direct preference optimization (HDPO) technique to improve the multi-modal, multi-task prompt optimization process. Experiments on various tasks show our method surpasses strong baselines and other competitors.
- Open Access
- 1 September 2025
- Published by International Joint Conferences on Artificial Intelligence
- Vol. 11, 10180-10188
AdaptEdit: An Adaptive Correspondence Guidance Framework for Reference-Based Video Editing
Video editing is a pivotal process for customizing video content according to user needs. However, existing text-guided methods often lead to ambiguities regarding user intentions and restrict fine-grained control for editing specific aspects in videos. To overcome these limitations, this paper introduces a novel approach named \emph{AdaptEdit}, which focuses on reference-based video editing that disentangles the editing process. It achieves this by first editing a reference image and then adaptively propagating its appearance across other frames to complete the video editing. While previous propagation methods, such as optical flow and the temporal modules of recent video generative models, struggle with object deformations and large motions, we propose an adaptive correspondence strategy that accurately transfers the appearance from the reference frame to the target frames by leveraging inter-frame semantic correspondences in the original video. By implementing a proxy-editing task to optimize hyperparameters for image token-level correspondence, our method effectively balances the need to maintain the target frame's structure while preventing leakage of irrelevant appearance. To more accurately evaluate editing beyond the semantic-level consistency provided by CLIP-style models, we introduce a new dataset, PVA, which supports pixel-level evaluation. Our method outperforms the best-performing baseline with a clear PSNR improvement of 3.6 dB.
- 1 September 2025
- Published by International Joint Conferences on Artificial Intelligence
- Vol. 11, 10180-10188
Open AccessAdaptEdit: An Adaptive Correspondence Guidance Framework for Reference-Based Video Editing
Video editing is a pivotal process for customizing video content according to user needs. However, existing text-guided methods often lead to ambiguities regarding user intentions and restrict fine-grained control for editing specific aspects in videos. To overcome these limitations, this paper introduces a novel approach named \emph{AdaptEdit}, which focuses on reference-based video editing that disentangles the editing process. It achieves this by first editing a reference image and then adaptively propagating its appearance across other frames to complete the video editing. While previous propagation methods, such as optical flow and the temporal modules of recent video generative models, struggle with object deformations and large motions, we propose an adaptive correspondence strategy that accurately transfers the appearance from the reference frame to the target frames by leveraging inter-frame semantic correspondences in the original video. By implementing a proxy-editing task to optimize hyperparameters for image token-level correspondence, our method effectively balances the need to maintain the target frame's structure while preventing leakage of irrelevant appearance. To more accurately evaluate editing beyond the semantic-level consistency provided by CLIP-style models, we introduce a new dataset, PVA, which supports pixel-level evaluation. Our method outperforms the best-performing baseline with a clear PSNR improvement of 3.6 dB.
- Open Access
- 1 September 2025
- Published by International Joint Conferences on Artificial Intelligence
- Vol. 11, 10072-10080
FastBlend: Enhancing Video Stylization Consistency via Model-Free Patch Blending
With the emergence of diffusion models and the rapid development of image processing, generating artistic images in style transfer tasks has become effortless. However, these impressive image processing approaches face consistency issues in video processing due to the independent processing of each frame. In this paper, we propose a powerful, model-free approach called FastBlend to address the consistency problem in video stylization. FastBlend functions as a post-processor and can be seamlessly integrated with diffusion models to create a robust video stylization pipeline. Based on a patch-matching algorithm, we remap and blend the aligned content across multiple frames, thus compensating for inconsistent content with neighboring frames. Moreover, we propose a tree-like data structure and a specialized loss function, aiming to optimize computational efficiency and visual quality for different application scenarios. Extensive experiments have demonstrated the effectiveness of FastBlend. Compared with both independent video deflickering algorithms and diffusion-based video processing methods, FastBlend is capable of synthesizing more coherent and realistic videos.
- 1 September 2025
- Published by International Joint Conferences on Artificial Intelligence
- Vol. 11, 10072-10080
Open AccessFastBlend: Enhancing Video Stylization Consistency via Model-Free Patch Blending
With the emergence of diffusion models and the rapid development of image processing, generating artistic images in style transfer tasks has become effortless. However, these impressive image processing approaches face consistency issues in video processing due to the independent processing of each frame. In this paper, we propose a powerful, model-free approach called FastBlend to address the consistency problem in video stylization. FastBlend functions as a post-processor and can be seamlessly integrated with diffusion models to create a robust video stylization pipeline. Based on a patch-matching algorithm, we remap and blend the aligned content across multiple frames, thus compensating for inconsistent content with neighboring frames. Moreover, we propose a tree-like data structure and a specialized loss function, aiming to optimize computational efficiency and visual quality for different application scenarios. Extensive experiments have demonstrated the effectiveness of FastBlend. Compared with both independent video deflickering algorithms and diffusion-based video processing methods, FastBlend is capable of synthesizing more coherent and realistic videos.