arXiv subjects

Chuxiong Sun

Publications and source records attributed to Chuxiong Sun.

14 recordsLinked to original sources

arXiv2026

Discovering Efficient and Explainable Communication Topologies for LLM-based Multi-Agent Systems via Causal Inference

The performance of large language model (LLM)-based multi-agent systems (MAS) largely depends on effective communication topologies. Existing topology generation methods, however, typically learn communication topologies through black-box optimization driven solely by task-level rewards. While effective, such optimization provides little insight into why particular communication edges are selected, making it difficult to identify the critical communication subgraphs responsible for successful collaboration. To address this limitation, we propose E2-Explainer, a model-agnostic framework for providing interpretable explanations of communication topologies produced by arbitrary topology generators. Specifically, we formulate topology explanation as a causal attribution problem that identifies compact communication subgraphs supported by edge-level evidence of task preservation. We obtain this evidence with a Granger-style objective that measures how masking each communication channel changes the task outcome and the stability of the final response. The resulting budgeted subgraphs are then distilled into an amortized explainer, enabling efficient post-hoc explanation without repeated edge-level evaluations at deployment. Extensive experiments on multiple reasoning and coding benchmarks demonstrate that E2-Explainer identifies critical communication subgraphs that preserve successful collaboration. These subgraphs can also be executed directly to prune redundant communication edges, substantially reducing communication costs while maintaining competitive task performance.

cs.MA↗

arXiv2026

TIDE-Bench: Task-Aware and Diagnostic Evaluation of Tool-Integrated Reasoning

Tool-integrated reasoning has emerged as a promising paradigm for enhancing large language models with external computation, retrieval, and execution capabilities. However, the field still lacks a high-quality and unified evaluation benchmark, and existing TIR evaluations remain limited in dataset quality, task diversity, diagnostic comprehensiveness, and evaluation efficiency. In this work, we introduce TIDE-Bench, a holistic and efficient benchmark for evaluating TIR methods, featuring three key advantages. First, it provides diverse task settings, combining widely used mathematical reasoning and knowledge-intensive QA tasks with two newly designed tasks, namely the tool-grounded experimental design task and the dynamic interactive task, to probe models' abilities in complex tool invocation and multi-tool coordination. Second, TIDE-Bench adopts a comprehensive yet task-aware evaluation protocol, jointly measuring final answer quality, process reliability, tool-use efficiency, and inference cost across heterogeneous task settings. Third, TIDE-Bench constructs high-quality and discriminative evaluation sets by filtering low-discrimination instances from existing datasets, substantially reducing evaluation cost while focusing on more challenging samples. Extensive experiments on multiple foundation models and TIR methods reveal persistent bottlenecks in tool grounding, offering insights for future TIR research.

cs.AI↗

arXiv2026

Multi-modal Test-time Adaptation via Adaptive Probabilistic Gaussian Calibration

Multi-modal test-time adaptation (TTA) enhances the resilience of benchmark multi-modal models against distribution shifts by leveraging the unlabeled target data during inference. Despite the documented success, the advancement of multi-modal TTA methodologies has been impeded by a persistent limitation, i.e., the lack of explicit modeling of category-conditional distributions, which is crucial for yielding accurate predictions and reliable decision boundaries. Canonical Gaussian discriminant analysis (GDA) provides a vanilla modeling of category-conditional distributions and achieves moderate advancement in uni-modal contexts. However, in multi-modal TTA scenario, the inherent modality distribution asymmetry undermines the effectiveness of modeling the category-conditional distribution via the canonical GDA. To this end, we introduce a tailored probabilistic Gaussian model for multi-modal TTA to explicitly model the category-conditional distributions, and further propose an adaptive contrastive asymmetry rectification technique to counteract the adverse effects arising from modality asymmetry, thereby deriving calibrated predictions and reliable decision boundaries. Extensive experiments across diverse benchmarks demonstrate that our method achieves state-of-the-art performance under a wide range of distribution shifts. The code is available at https://github.com/XuJinglinn/AdaPGC.

cs.CV↗

arXiv2025

HTG-GCL: Leveraging Hierarchical Topological Granularity from Cellular Complexes for Graph Contrastive Learning

Graph contrastive learning (GCL) aims to learn discriminative semantic invariance by contrasting different views of the same graph that share critical topological patterns. However, existing GCL approaches with structural augmentations often struggle to identify task-relevant topological structures, let alone adapt to the varying coarse-to-fine topological granularities required across different downstream tasks. To remedy this issue, we introduce Hierarchical Topological Granularity Graph Contrastive Learning (HTG-GCL), a novel framework that leverages transformations of the same graph to generate multi-scale ring-based cellular complexes, embodying the concept of topological granularity, thereby generating diverse topological views. Recognizing that a certain granularity may contain misleading semantics, we propose a multi-granularity decoupled contrast and apply a granularity-specific weighting mechanism based on uncertainty estimation. Comprehensive experiments on various benchmarks demonstrate the effectiveness of HTG-GCL, highlighting its superior performance in capturing meaningful graph representations through hierarchical topological information.

cs.LG↗

arXiv2025

Group Causal Policy Optimization for Post-Training Large Language Models

Recent advances in large language models (LLMs) have broadened their applicability across diverse tasks, yet specialized domains still require targeted post training. Among existing methods, Group Relative Policy Optimization (GRPO) stands out for its efficiency, leveraging groupwise relative rewards while avoiding costly value function learning. However, GRPO treats candidate responses as independent, overlooking semantic interactions such as complementarity and contradiction. To address this challenge, we first introduce a Structural Causal Model (SCM) that reveals hidden dependencies among candidate responses induced by conditioning on a final integrated output forming a collider structure. Then, our causal analysis leads to two insights: (1) projecting responses onto a causally informed subspace improves prediction quality, and (2) this projection yields a better baseline than query only conditioning. Building on these insights, we propose Group Causal Policy Optimization (GCPO), which integrates causal structure into optimization through two key components: a causally informed reward adjustment and a novel KL regularization term that aligns the policy with a causally projected reference distribution. Comprehensive experimental evaluations demonstrate that GCPO consistently surpasses existing methods, including GRPO across multiple reasoning benchmarks.

cs.LG↗

arXiv2025

Training Language Model to Critique for Better Refinement

Large language models (LLMs) have demonstrated remarkable evaluation and critique capabilities, providing insightful feedback and identifying flaws in various tasks. However, limited research has explored which types of critiques are most effective for improving model responses or how to generate such critiques. To address this gap, we introduce \textbf{R}efinement-oriented \textbf{C}ritique \textbf{O}ptimization (RCO), a novel framework designed to train critic models using refinement signals. RCO uses a feedback loop where critiques, generated by the critic model, guide the actor model in refining its responses. The critique utility (CU) quantifies the effectiveness of these refinements, serving as the reward signal for training the critic model. By focusing on critiques that lead to better refinements, RCO eliminates the need for direct critique preference assessment, ensuring that critiques driving meaningful improvements are rewarded. We evaluate RCO across five tasks, i.e., dialog generation, summarization, question answering, mathematical reasoning, and code generation, and show that it significantly outperforms traditional methods and open-source models in terms of critique quality and refinement outcomes. Our contributions include the introduction of RCO, a novel supervision scheme based on refined response preferences, and comprehensive experimental results that highlight the method's effectiveness in enhancing LLM critique-refinement loops.

cs.CL↗

arXiv2025

Token-level Accept or Reject: A Micro Alignment Approach for Large Language Models

With the rapid development of Large Language Models (LLMs), aligning these models with human preferences and values is critical to ensuring ethical and safe applications. However, existing alignment techniques such as RLHF or DPO often require direct fine-tuning on LLMs with billions of parameters, resulting in substantial computational costs and inefficiencies. To address this, we propose Micro token-level Accept-Reject Aligning (MARA) approach designed to operate independently of the language models. MARA simplifies the alignment process by decomposing sentence-level preference learning into token-level binary classification, where a compact three-layer fully-connected network determines whether candidate tokens are "Accepted" or "Rejected" as part of the response. Extensive experiments across seven different LLMs and three open-source datasets show that MARA achieves significant improvements in alignment performance while reducing computational costs. The source code and implementation details are publicly available at https://github.com/IAAR-Shanghai/MARA, and the trained models are released at https://huggingface.co/IAAR-Shanghai/MARA_AGENTS.

cs.CL↗

arXiv2025

Revisiting Communication Efficiency in Multi-Agent Reinforcement Learning from the Dimensional Analysis Perspective

In this work, we introduce a novel perspective, i.e., dimensional analysis, to address the challenge of communication efficiency in Multi-Agent Reinforcement Learning (MARL). Our findings reveal that simply optimizing the content and timing of communication at sending end is insufficient to fully resolve communication efficiency issues. Even after applying optimized and gated messages, dimensional redundancy and confounders still persist in the integrated message embeddings at receiving end, which negatively impact communication quality and decision-making. To address these challenges, we propose Dimensional Rational Multi-Agent Communication (DRMAC), designed to mitigate both dimensional redundancy and confounders in MARL. DRMAC incorporates a redundancy-reduction regularization term to encourage the decoupling of information across dimensions within the learned representations of integrated messages. Additionally, we introduce a dimensional mask that dynamically adjusts gradient weights during training to eliminate the influence of decision-irrelevant dimensions. We evaluate DRMAC across a diverse set of multi-agent tasks, demonstrating its superior performance over existing state-of-the-art methods in complex scenarios. Furthermore, the plug-and-play nature of DRMAC's key modules highlights its generalizable performance, serving as a valuable complement rather than a replacement for existing multi-agent communication strategies.

cs.MA↗

arXiv2024

M2I2: Learning Efficient Multi-Agent Communication via Masked State Modeling and Intention Inference

Communication is essential in coordinating the behaviors of multiple agents. However, existing methods primarily emphasize content, timing, and partners for information sharing, often neglecting the critical aspect of integrating shared information. This gap can significantly impact agents' ability to understand and respond to complex, uncertain interactions, thus affecting overall communication efficiency. To address this issue, we introduce M2I2, a novel framework designed to enhance the agents' capabilities to assimilate and utilize received information effectively. M2I2 equips agents with advanced capabilities for masked state modeling and joint-action prediction, enriching their perception of environmental uncertainties and facilitating the anticipation of teammates' intentions. This approach ensures that agents are furnished with both comprehensive and relevant information, bolstering more informed and synergistic behaviors. Moreover, we propose a Dimensional Rational Network, innovatively trained via a meta-learning paradigm, to identify the importance of dimensional pieces of information, evaluating their contributions to decision-making and auxiliary tasks. Then, we implement an importance-based heuristic for selective information masking and sharing. This strategy optimizes the efficiency of masked state modeling and the rationale behind information sharing. We evaluate M2I2 across diverse multi-agent tasks, the results demonstrate its superior performance, efficiency, and generalization capabilities, over existing state-of-the-art methods in various complex scenarios.

cs.MA↗

arXiv2024

Rethinking Generalizability and Discriminability of Self-Supervised Learning from Evolutionary Game Theory Perspective

Representations learned by self-supervised approaches are generally considered to possess sufficient generalizability and discriminability. However, we disclose a nontrivial mutual-exclusion relationship between these critical representation properties through an exploratory demonstration on self-supervised learning. State-of-the-art self-supervised methods tend to enhance either generalizability or discriminability but not both simultaneously. Thus, learning representations jointly possessing strong generalizability and discriminability presents a specific challenge for self-supervised learning. To this end, we revisit the learning paradigm of self-supervised learning from the perspective of evolutionary game theory (EGT) and outline the theoretical roadmap to achieve a desired trade-off between these representation properties. EGT performs well in analyzing the trade-off point in a two-player game by utilizing dynamic system modeling. However, the EGT analysis requires sufficient annotated data, which contradicts the principle of self-supervised learning, i.e., the EGT analysis cannot be conducted without the annotations of the specific target domain for self-supervised learning. Thus, to enhance the methodological generalization, we propose a novel self-supervised learning method that leverages advancements in reinforcement learning to jointly benefit from the general guidance of EGT and sequentially optimize the model to chase the consistent improvement of generalizability and discriminability for specific target domains during pre-training. Theoretically, we establish that the proposed method tightens the generalization error upper bound of self-supervised learning. Empirically, our method achieves state-of-the-art performance on various benchmarks.

cs.AI↗

arXiv2024

Proxy-RLHF: Decoupling Generation and Alignment in Large Language Model with Proxy

Reinforcement Learning from Human Feedback (RLHF) is the prevailing approach to ensure Large Language Models (LLMs) align with human values. However, existing RLHF methods require a high computational cost, one main reason being that RLHF assigns both the generation and alignment tasks to the LLM simultaneously. In this paper, we introduce Proxy-RLHF, which decouples the generation and alignment processes of LLMs, achieving alignment with human values at a much lower computational cost. We start with a novel Markov Decision Process (MDP) designed for the alignment process and employ Reinforcement Learning (RL) to train a streamlined proxy model that oversees the token generation of the LLM, without altering the LLM itself. Experiments show that our method achieves a comparable level of alignment with only 1\% of the training parameters of other methods.

cs.CL↗

arXiv2024

T2MAC: Targeted and Trusted Multi-Agent Communication through Selective Engagement and Evidence-Driven Integration

Communication stands as a potent mechanism to harmonize the behaviors of multiple agents. However, existing works primarily concentrate on broadcast communication, which not only lacks practicality, but also leads to information redundancy. This surplus, one-fits-all information could adversely impact the communication efficiency. Furthermore, existing works often resort to basic mechanisms to integrate observed and received information, impairing the learning process. To tackle these difficulties, we propose Targeted and Trusted Multi-Agent Communication (T2MAC), a straightforward yet effective method that enables agents to learn selective engagement and evidence-driven integration. With T2MAC, agents have the capability to craft individualized messages, pinpoint ideal communication windows, and engage with reliable partners, thereby refining communication efficiency. Following the reception of messages, the agents integrate information observed and received from different sources at an evidence level. This process enables agents to collectively use evidence garnered from multiple perspectives, fostering trusted and cooperative behaviors. We evaluate our method on a diverse set of cooperative multi-agent tasks, with varying difficulties, involving different scales and ranging from Hallway, MPE to SMAC. The experiments indicate that the proposed model not only surpasses the state-of-the-art methods in terms of cooperative performance and communication efficiency, but also exhibits impressive generalization.

cs.MA↗

arXiv2022

Adaptive Graph Diffusion Networks

Graph Neural Networks (GNNs) have received much attention in the graph deep learning domain. However, recent research empirically and theoretically shows that deep GNNs suffer from over-fitting and over-smoothing problems. The usual solutions either cannot solve extensive runtime of deep GNNs or restrict graph convolution in the same feature space. We propose the Adaptive Graph Diffusion Networks (AGDNs) which perform multi-layer generalized graph diffusion in different feature spaces with moderate complexity and runtime. Standard graph diffusion methods combine large and dense powers of the transition matrix with predefined weighting coefficients. Instead, AGDNs combine smaller multi-hop node representations with learnable and generalized weighting coefficients. We propose two scalable mechanisms of weighting coefficients to capture multi-hop information: Hop-wise Attention (HA) and Hop-wise Convolution (HC). We evaluate AGDNs on diverse, challenging Open Graph Benchmark (OGB) datasets with semi-supervised node classification and link prediction tasks. Until the date of submission (Aug 26, 2022), AGDNs achieve top-1 performance on the ogbn-arxiv, ogbn-proteins and ogbl-ddi datasets and top-3 performance on the ogbl-citation2 dataset. On the similar Tesla V100 GPU cards, AGDNs outperform Reversible GNNs (RevGNNs) with 13% complexity and 1% training runtime of RevGNNs on the ogbn-proteins dataset. AGDNs also achieve comparable performance to SEAL with 36% training and 0.2% inference runtime of SEAL on the ogbl-citation2 dataset.

cs.LG↗

arXiv2021

Scalable and Adaptive Graph Neural Networks with Self-Label-Enhanced training

It is hard to directly implement Graph Neural Networks (GNNs) on large scaled graphs. Besides of existed neighbor sampling techniques, scalable methods decoupling graph convolutions and other learnable transformations into preprocessing and post classifier allow normal minibatch training. By replacing redundant concatenation operation with attention mechanism in SIGN, we propose Scalable and Adaptive Graph Neural Networks (SAGN). SAGN can adaptively gather neighborhood information among different hops. To further improve scalable models on semi-supervised learning tasks, we propose Self-Label-Enhance (SLE) framework combining self-training approach and label propagation in depth. We add base model with a scalable node label module. Then we iteratively train models and enhance train set in several stages. To generate input of node label module, we directly apply label propagation based on one-hot encoded label vectors without inner random masking. We find out that empirically the label leakage has been effectively alleviated after graph convolutions. The hard pseudo labels in enhanced train set participate in label propagation with true labels. Experiments on both inductive and transductive datasets demonstrate that, compared with other sampling-based and sampling-free methods, SAGN achieves better or comparable results and SLE can further improve performance.

cs.LG↗