publications
* denotes equal contribution
2026
- CoVe: Training Interactive Tool-Use Agents via Constraint-Guided VerificationJinpeng Chen*, Cheng Gong*, Hanbo Li* , Ziru Liu , Zichen Tian, Xinyu Fu , Shi Wu, Chenyang Zhang , Wu Zhang , Suiyun Zhang, Dandan Tu , and Rui LiuCoRR, 2026
Developing multi-turn interactive tool-use agents is challenging because real-world user needs are often complex and ambiguous, yet agents must execute deterministic actions to satisfy them. To address this gap, we introduce CoVe (Constraint-Verification), a post-training data synthesis framework designed for training interactive tool-use agents while ensuring both data complexity and correctness. CoVe begins by defining explicit task constraints, which serve a dual role: they guide the generation of complex trajectories and act as deterministic verifiers for assessing trajectory quality. This enables the creation of high-quality training trajectories for supervised fine-tuning (SFT) and the derivation of accurate reward signals for reinforcement learning (RL). Our evaluation on the challenging τ2-bench benchmark demonstrates the effectiveness of the framework. Notably, our compact CoVe-4B model achieves success rates of 43.0% and 59.4% in the Airline and Retail domains, respectively; its overall performance significantly outperforms strong baselines of similar scale and remains competitive with models up to 17× its size. These results indicate that CoVe provides an effective and efficient pathway for synthesizing training data for state-of-the-art interactive tool-use agents. To support future research, we open-source our code, trained model, and the full set of 12K high-quality trajectories used for training.
@article{DBLP:journals/corr/abs-2603-01940, author = {Chen, Jinpeng and Gong, Cheng and Li, Hanbo and Liu, Ziru and Tian, Zichen and Fu, Xinyu and Wu, Shi and Zhang, Chenyang and Zhang, Wu and Zhang, Suiyun and Tu, Dandan and Liu, Rui}, title = {CoVe: Training Interactive Tool-Use Agents via Constraint-Guided Verification}, journal = {CoRR}, volume = {abs/2603.01940}, year = {2026}, doi = {10.48550/ARXIV.2603.01940}, }
2025
- Beyond Confidence: Adaptive and Coherent Decoding for Diffusion Language ModelsKecheng Chen* , Ziru Liu* , Xijia Tao, Hui Liu, Xinyu Fu , Suiyun Zhang, Dandan Tu, Lingpeng Kong , Rui Liu , and Haoliang LiCoRR, 2025
Diffusion Language Models (DLMs) have recently achieved significant success due to their any-order generation capabilities. However, existing inference methods typically rely on local, immediate-step metrics—such as confidence or entropy—which inherently lack a more reliable perspective, leading to sub-optimal generation quality. To address this, we propose Coherent Contextual Decoding (CCD), a novel inference framework built upon two core innovations. First, CCD bypasses the potential bias of the single context to leverage historical contexts for approximating the marginal distribution of token prediction, leading to better sequence coherence and the early rejection of sub-optimal paths. More importantly, we demonstrate that this mechanism is theoretically equivalent to modeling the consistency of historical steps via the conditional mutual information between contexts and token predictions. Finally, CCD achieves significantly milder performance degradation under highly parallel decoding scenarios compared to baselines. Empirically, our method achieves a simultaneous enhancement in both inference speed and performance across diverse benchmarks on Dream and LLaDA.
@article{DBLP:journals/corr/abs-2512-02044, author = {Chen, Kecheng and Liu, Ziru and Tao, Xijia and Liu, Hui and Fu, Xinyu and Zhang, Suiyun and Tu, Dandan and Kong, Lingpeng and Liu, Rui and Li, Haoliang}, title = {Beyond Confidence: Adaptive and Coherent Decoding for Diffusion Language Models}, journal = {CoRR}, volume = {abs/2512.02044}, year = {2025}, doi = {10.48550/ARXIV.2512.02044}, } - MathCanvas: Intrinsic Visual Chain-of-Thought for Multimodal Mathematical ReasoningWeikang Shi*, Aldrich Yu*, Rongyao Fang*, Houxing Ren, Ke Wang, Aojun Zhou, Changyao Tian, Xinyu Fu , Yuxuan Hu, Zimu Lu, Linjiang Huang , Si Liu , Rui Liu† , and Hongsheng Li†CoRR, 2025
While Large Language Models (LLMs) have excelled in textual reasoning, they struggle with mathematical domains like geometry that intrinsically rely on visual aids. Existing approaches to Visual Chain-of-Thought (VCoT) are often limited by rigid external tools or fail to generate the high-fidelity, strategically-timed diagrams necessary for complex problem-solving. To bridge this gap, we introduce MathCanvas, a comprehensive framework designed to endow unified Large Multimodal Models (LMMs) with intrinsic VCoT capabilities for mathematics. Our approach consists of two phases. First, a Visual Manipulation stage pre-trains the model on a novel 15.2M-pair corpus, comprising 10M caption-to-diagram pairs (MathCanvas-Imagen) and 5.2M step-by-step editing trajectories (MathCanvas-Edit), to master diagram generation and editing. Second, a Strategic Visual-Aided Reasoning stage fine-tunes the model on MathCanvas-Instruct, a new 219K-example dataset of interleaved visual-textual reasoning paths, teaching it when and how to leverage visual aids. To facilitate rigorous evaluation, we introduce MathCanvas-Bench, a challenging benchmark with 3K problems that require models to produce interleaved visual-textual solutions. Our model, BAGEL-Canvas, trained under this framework, achieves an 86% relative improvement over strong LMM baselines on MathCanvas-Bench, demonstrating excellent generalization to other public math benchmarks. Our work provides a complete toolkit-framework, datasets, and benchmark-to unlock complex, human-like visual-aided reasoning in LMMs.
@article{DBLP:journals/corr/abs-2510-14958, author = {Shi, Weikang and Yu, Aldrich and Fang, Rongyao and Ren, Houxing and Wang, Ke and Zhou, Aojun and Tian, Changyao and Fu, Xinyu and Hu, Yuxuan and Lu, Zimu and Huang, Linjiang and Liu, Si and Liu, Rui and Li, Hongsheng}, title = {MathCanvas: Intrinsic Visual Chain-of-Thought for Multimodal Mathematical Reasoning}, journal = {CoRR}, volume = {abs/2510.14958}, year = {2025}, doi = {10.48550/ARXIV.2510.14958}, } - MMSearch-Plus: Benchmarking Provenance-Aware Search for Multimodal Browsing AgentsXijia Tao*, Yihua Teng*, Xinxing Su*, Xinyu Fu, Jihao Wu, Chaofan Tao , Ziru Liu, Haoli Bai , Rui Liu†, and Lingpeng Kong†CoRR, 2025
Existing multimodal browsing benchmarks often fail to require genuine multimodal reasoning, as many tasks can be solved with text-only heuristics without vision-in-the-loop verification. We introduce MMSearch-Plus, a 311-task benchmark that enforces multimodal understanding by requiring extraction and propagation of fine-grained visual cues through iterative image-text retrieval and cross-validation under retrieval noise. Our curation procedure seeds questions whose answers require extrapolating from spatial cues and temporal traces to out-of-image facts such as events, dates, and venues. Beyond the dataset, we provide a model-agnostic agent framework with standard browsing tools and a set-of-mark (SoM) module, which lets the agent place marks, crop subregions, and launch targeted image/text searches. SoM enables provenance-aware zoom-and-retrieve and improves robustness in multi-step reasoning. We evaluated closed- and open-source MLLMs in this framework. The strongest system achieves an end-to-end accuracy of 36.0%, and integrating SoM produces consistent gains in multiple settings, with improvements up to +3.9 points. From failure analysis, we observe recurring errors in locating relevant webpages and distinguishing between visually similar events. These results underscore the challenges of real-world multimodal search and establish MMSearch-Plus as a rigorous benchmark for advancing agentic MLLMs.
@article{DBLP:journals/corr/abs-2508-21475, author = {Tao, Xijia and Teng, Yihua and Su, Xinxing and Fu, Xinyu and Wu, Jihao and Tao, Chaofan and Liu, Ziru and Bai, Haoli and Liu, Rui and Kong, Lingpeng}, title = {MMSearch-Plus: Benchmarking Provenance-Aware Search for Multimodal Browsing Agents}, journal = {CoRR}, volume = {abs/2508.21475}, year = {2025}, doi = {10.48550/arxiv.2508.21475}, } - GHPO: Adaptive Guidance for Stable and Efficient LLM Reinforcement LearningZiru Liu*, Cheng Gong*, Xinyu Fu , Yaofang Liu , Ran Chen, Shoubo Hu , Suiyun Zhang , Rui Liu† , Qingfu Zhang†, and Dandan Tu†CoRR, 2025
Reinforcement Learning with Verifiable Rewards (RLVR) has recently emerged as a powerful paradigm for facilitating the self-improvement of large language models (LLMs), particularly in the domain of complex reasoning tasks. However, prevailing on-policy RL methods often contend with significant training instability and inefficiency. This is primarily due to a capacity-difficulty mismatch, where the complexity of training data frequently outpaces the model’s current capabilities, leading to critically sparse reward signals and stalled learning progress. This challenge is particularly acute for smaller, more resource-efficient LLMs. To overcome this, we introduce the Guided Hybrid Policy Optimization (GHPO), a novel difficulty-aware reinforcement learning framework. GHPO dynamically calibrates task difficulty by employing adaptive prompt refinement to provide targeted guidance. This unique approach adaptively balances direct imitation learning for problems currently beyond the model’s reach with exploration-based reinforcement learning for more manageable tasks, effectively creating a smooth and optimized learning curriculum. Extensive experiments demonstrate that GHPO achieves an average performance gain of approximately 5% across six challenging mathematics benchmarks, consistently outperforming strong on-policy reinforcement learning and curriculum learning baselines. Further analysis confirms that our framework significantly enhances both training stability and final reasoning performance, thus offering a scalable and efficient solution for developing powerful and robust reasoning models.
@article{DBLP:journals/corr/abs-2507-10628, author = {Liu, Ziru and Gong, Cheng and Fu, Xinyu and Liu, Yaofang and Chen, Ran and Hu, Shoubo and Zhang, Suiyun and Liu, Rui and Zhang, Qingfu and Tu, Dandan}, title = {{GHPO:} Adaptive Guidance for Stable and Efficient {LLM} Reinforcement Learning}, journal = {CoRR}, volume = {abs/2507.10628}, year = {2025}, doi = {10.48550/arXiv.2507.10628}, }
2024
- Client-Specific Hyperbolic Federated LearningIn FedKDD Workshop, 2024
Personalized Federated Learning (PFL) has gained attention for privacy-preserving training on heterogeneous data. However, existing methods fail to capture the unique inherent geometric properties across diverse datasets by assuming a unified Euclidean space for all data distributions. Drawing on hyperbolic geometry’s ability to fit complex data properties, we present FlatLand, a novel personalized Federated learning method that embeds different clients’ data in tailored Lorentz space. FlatLand is able to directly tackle the challenge of heterogeneity through the personalized curvatures of their respective Lorentz model of hyperbolic geometry, which is manifested by the time-like dimension. Leveraging the Lorentz model properties, we further design a parameter decoupling strategy that enables direct server aggregation of common client information, with reduced heterogeneity interference and without the need for client-wise similarity estimation. To the best of our knowledge, this is the first attempt to incorporate hyperbolic geometry into personalized federated learning. Empirical results on various federated graph learning tasks demonstrate that FlatLand achieves superior performance, particularly in low-dimensional settings.
@inproceedings{jiahong_flatland, author = {Liu, Jiahong and Fu, Xinyu and Yang, Menglin and Zhang, Weixi and Ying, Rex and King, Irwin}, title = {Client-Specific Hyperbolic Federated Learning}, booktitle = {{FedKDD} Workshop}, year = {2024}, } - Geometric View of Soft Decorrelation in Self-Supervised LearningYifei Zhang, Hao Zhu, Zixing Song , Yankai Chen, Xinyu Fu, Ziqiao Meng, Piotr Koniusz†, and Irwin King†In KDD, 2024
Contrastive learning, a form of Self-Supervised Learning (SSL), typically consists of an alignment term and a regularization term. The alignment term minimizes the distance between the embeddings of a positive pair, while the regularization term prevents trivial solutions and expresses prior beliefs about the embeddings. As a widely used regularization technique, soft decorrelation has been employed by several non-contrastive SSL methods to avoid trivial solutions. While the decorrelation term is designed to address the issue of dimensional collapse, we find that it fails to achieve this goal theoretically and experimentally. Based on such a finding, we extend the soft decorrelation regularization to minimize the distance between the covariance matrix and an identity matrix. We provide a new perspective on the geometric distance between positive definite matrices to investigate why the soft decorrelation cannot efficiently solve the dimensional collapse. Furthermore, we construct a family of loss functions utilizing the Bregman Matrix Divergence (BMD), with the soft decorrelation representing a specific instance within this family. We prove that a loss function (LogDet) in this family can solve the issue of dimensional collapse. Our novel loss functions based on BMD exhibit superior performance compared to the soft decorrelation and other baseline techniques, as demonstrated by experimental results on graph and image datasets.
@inproceedings{DBLP:conf/kdd/ZhangZS00MKK24, author = {Zhang, Yifei and Zhu, Hao and Song, Zixing and Chen, Yankai and Fu, Xinyu and Meng, Ziqiao and Koniusz, Piotr and King, Irwin}, title = {Geometric View of Soft Decorrelation in Self-Supervised Learning}, booktitle = {{KDD}}, pages = {4338--4349}, publisher = {{ACM}}, year = {2024}, doi = {10.1145/3637528.3671914}, } - A Survey of Trustworthy Federated Learning: Issues, Solutions, and ChallengesACM Transactions on Intelligent Systems and Technology, 2024
Trustworthy artificial intelligence (TAI) has proven invaluable in curbing potential negative repercussions tied to AI applications. Within the TAI spectrum, federated learning (FL) emerges as a promising solution to safeguard personal information in distributed settings across a multitude of practical contexts. However, the realm of FL is not without its challenges. Especially worrisome are adversarial attacks targeting its algorithmic robustness and systemic confidentiality. Moreover, the presence of biases and opacity in prediction outcomes further complicates FL’s broader adoption. Consequently, there is a growing expectation for FL to instill trust. To address this, we chart out a comprehensive road-map for Trustworthy Federated Learning (TFL) and provide an overview of existing efforts across four pivotal dimensions: Privacy and Security, Robustness, Fairness, and Explainability. For each dimension, we identify potential pitfalls that might undermine TFL and present a curated selection of defensive strategies, enriched by a discourse on technical solutions tailored for TFL. Furthermore, we present potential challenges and future directions to be explored for in-depth TFL research with broader impacts.
@article{DBLP:journals/tist/ZhangZLFCXK24, author = {Zhang, Yifei and Zeng, Dun and Luo, Jinglong and Fu, Xinyu and Chen, Guanzhong and Xu, Zenglin and King, Irwin}, title = {A Survey of Trustworthy Federated Learning: Issues, Solutions, and Challenges}, journal = {ACM Transactions on Intelligent Systems and Technology}, volume = {15}, number = {6}, pages = {112:1--112:47}, year = {2024}, doi = {10.1145/3678181}, } - A Systematic Survey on Federated Semi-supervised LearningIn IJCAI, 2024
Federated learning (FL) revolutionizes distributed machine learning by enabling devices to collaboratively learn a model while maintaining data privacy. However, FL usually faces a critical challenge with limited labeled data, making semi-supervised learning (SSL) crucial for utilizing abundant unlabeled data. The integration of SSL within the federated framework gives rise to federated semi-supervised learning (FSSL), a novel approach that exploits unlabeled data across devices without compromising privacy. This paper systematically explores FSSL, shedding light on its four basic problem settings that commonly appear in real-world scenarios. By examining the unique challenges, generic solutions, and representative methods tailored for each setting of FSSL, we aim to provide a cohesive overview of the current state of the art and pave the way for future research directions in this promising field.
@inproceedings{DBLP:conf/ijcai/SongYZ0XK24, author = {Song, Zixing and Yang, Xiangli and Zhang, Yifei and Fu, Xinyu and Xu, Zenglin and King, Irwin}, title = {A Systematic Survey on Federated Semi-supervised Learning}, booktitle = {{IJCAI}}, pages = {8244--8252}, publisher = {ijcai.org}, year = {2024}, doi = {10.24963/ijcai.2024/911}, } - MECCH: Metapath Context Convolution-based Heterogeneous Graph Neural NetworksXinyu Fu†, and Irwin KingNeural Networks, 2024
Heterogeneous graph neural networks (HGNNs) were proposed for representation learning on structural data with multiple types of nodes and edges. To deal with the performance degradation issue when HGNNs become deep, researchers combine metapaths into HGNNs to associate nodes closely related in semantics but far apart in the graph. However, existing metapath-based models suffer from either information loss or high computation costs. To address these problems, we present a novel Metapath Context Convolution-based Heterogeneous Graph Neural Network (MECCH). MECCH leverages metapath contexts, a new kind of graph structure that facilitates lossless node information aggregation while avoiding any redundancy. Specifically, MECCH applies three novel components after feature preprocessing to extract comprehensive information from the input graph efficiently: (1) metapath context construction, (2) metapath context encoder, and (3) convolutional metapath fusion. Experiments on five real-world heterogeneous graph datasets for node classification and link prediction show that MECCH achieves superior prediction accuracy compared with state-of-the-art baselines with improved computational efficiency.
@article{DBLP:journals/nn/FuK24, author = {Fu, Xinyu and King, Irwin}, title = {{MECCH:} Metapath Context Convolution-based Heterogeneous Graph Neural Networks}, journal = {Neural Networks}, volume = {170}, pages = {266--275}, year = {2024}, doi = {10.1016/j.neunet.2023.11.030}, }
2023
- FedHGN: A Federated Framework for Heterogeneous Graph Neural NetworksXinyu Fu, and Irwin KingIn IJCAI, 2023
Heterogeneous graph neural networks (HGNNs) can learn from typed and relational graph data more effectively than conventional GNNs. With larger parameter spaces, HGNNs may require more training data, which is often scarce in real-world applications due to privacy regulations (e.g., GDPR). Federated graph learning (FGL) enables multiple clients to train a GNN collaboratively without sharing their local data. However, existing FGL methods mainly focus on homogeneous GNNs or knowledge graph embeddings; few have considered heterogeneous graphs and HGNNs. In federated heterogeneous graph learning, clients may have private graph schemas. Conventional FL/FGL methods attempting to define a global HGNN model would violate schema privacy. To address these challenges, we propose FedHGN, a novel and general FGL framework for HGNNs. FedHGN adopts schema-weight decoupling to enable schema-agnostic knowledge sharing and employs coefficients alignment to stabilize the training process and improve HGNN performance. With better privacy preservation, FedHGN consistently outperforms local training and conventional FL methods on three widely adopted heterogeneous graph datasets with varying client numbers.
@inproceedings{DBLP:conf/ijcai/0004K23, author = {Fu, Xinyu and King, Irwin}, title = {FedHGN: {A} Federated Framework for Heterogeneous Graph Neural Networks}, booktitle = {{IJCAI}}, pages = {3705--3713}, publisher = {ijcai.org}, year = {2023}, doi = {10.24963/ijcai.2023/412}, }
2020
- MAGNN: Metapath Aggregated Graph Neural Network for Heterogeneous Graph EmbeddingIn WWW, 2020
A large number of real-world graphs or networks are inherently heterogeneous, involving a diversity of node types and relation types. Heterogeneous graph embedding is to embed rich structural and semantic information of a heterogeneous graph into low-dimensional node representations. Existing models usually define multiple metapaths in a heterogeneous graph to capture the composite relations and guide neighbor selection. However, these models either omit node content features, discard intermediate nodes along the metapath, or only consider one metapath. To address these three limitations, we propose a new model named Metapath Aggregated Graph Neural Network (MAGNN) to boost the final performance. Specifically, MAGNN employs three major components, i.e., the node content transformation to encapsulate input node attributes, the intra-metapath aggregation to incorporate intermediate semantic nodes, and the inter-metapath aggregation to combine messages from multiple metapaths. Extensive experiments on three real-world heterogeneous graph datasets for node classification, node clustering, and link prediction show that MAGNN achieves more accurate prediction results than state-of-the-art baselines.
@inproceedings{DBLP:conf/www/0004ZMK20, author = {Fu, Xinyu and Zhang, Jiani and Meng, Ziqiao and King, Irwin}, title = {{MAGNN:} Metapath Aggregated Graph Neural Network for Heterogeneous Graph Embedding}, booktitle = {{WWW}}, pages = {2331--2341}, publisher = {{ACM} / {IW3C2}}, year = {2020}, doi = {10.1145/3366423.3380297}, }
