Arrow Research search

Author name cluster

Hanqing Yang

Possible papers associated with this exact author name in Arrow. This page groups case-insensitive exact name matches and is not a full identity disambiguation profile.

2 papers
1 author row

Possible papers

2

AAAI Conference 2026 Conference Paper

Contribution-aware Token Compression for Efficient Video Understanding via Reinforcement Learning

  • Yinchao Ma
  • Qiang Zhou
  • Zhibin Wang
  • Xianing Chen
  • Hanqing Yang
  • Jun Song
  • Bo Zheng

Video large language models have demonstrated remarkable capabilities in video understanding tasks. However, the redundancy of video tokens introduces significant computational overhead during inference, limiting their practical deployment. Many compression algorithms are proposed to prioritize retaining features with the highest attention scores to minimize perturbations in attention computations. However, the correlation between attention scores and their actual contribution to correct answers remains ambiguous. To address the above limitation, we propose a novel contribution-aware token compression algorithm for video understanding (CaCoVID) that explicitly optimizes the token selection policy based on the contribution of tokens to correct predictions. First, we introduce a reinforcement learning-based framework that optimizes a policy network to select video token combinations with the greatest contribution to correct predictions. This paradigm shifts the focus from passive token preservation to active discovery of optimal compressed token combinations. Secondly, we propose a combinatorial policy optimization algorithm with online combination space sampling, which dramatically reduces the exploration space for video token combinations and accelerates the convergence speed of policy optimization. Extensive experiments on diverse video understanding benchmarks demonstrate the effectiveness of CaCoVID. Codes will be released.

PDF Details DOI

TMLR Journal 2026 Journal Article

The Five Ws of Multi-Agent Communication: Who Talks to Whom, When, What, and Why - A Survey from MARL to Emergent Language and LLMs

  • Jingdi Chen
  • Hanqing Yang
  • Zongjun Liu
  • Carlee Joe-Wong

Multi-agent sequential decision-making underpins many real-world systems, from autonomous vehicles and robotics to collaborative AI assistants. In dynamic and partially observable environments, effective communication is essential for reducing uncertainty and enabling coordination. Although research on multi-agent communication (MA-Comm) spans diverse paradigms, we organize this survey explicitly around the Five Ws of communication: who communicates with whom, what is communicated, when communication occurs, and why communication is beneficial. This lens provides a coherent structure for synthesizing diverse approaches and exposing shared design principles across paradigms. Within Multi-Agent Reinforcement Learning (MARL), early work relied on hand-designed or implicit communication protocols, followed by trainable, end-to-end mechanisms optimized for reward and control. While effective, these approaches often yield task-specific and weakly interpretable communication, motivating research on Emergent Language (EL), where agents develop more structured or symbolic protocols through interaction. EL methods, however, still face challenges in grounding, generalization, and scalability, which have driven recent interest in large language models (LLMs) as a means to leverage natural language priors for reasoning, planning, and coordination in open-ended multi-agent settings. This progression motivates our survey: we analyze how communication paradigms evolve in response to the limitations of earlier approaches and how MARL, EL, and LLM-based systems address complementary aspects of multi-agent communication. This paper provides a unified survey of MA-Comm across MARL, EL, and LLM-based multi-agent systems. Organized around the Five Ws, we examine how different paradigms motivate, structure, and operationalize communication, reveal cross-paradigm trade-offs, and identify open challenges in communication, coordination, and learning. By offering systematic comparisons and design-oriented insights, this survey helps the community extract effective communication design patterns and supports the development of hybrid systems that combine learning, language, and control to meet diverse task, scalability, and interpretability requirements.

PDF Details
v2026.09.13