On May 1st, the admission results of the 42nd International Conference on Machine Learning (ICML 2025) were announced. The research paper "Hypergraph Coordination Networks with Dynamic Grouping for Multi-Agent Reinforcement Learning" of Professor Li Dazi's research group in our college has been successfully accepted. The first author of this paper is Liu Chiqiang, a doctoral student from the School of Information Science and Technology, and the corresponding author is Professor Li Dazi. Beijing University of Chemical Technology is the sole completing institution.
This study addresses the core challenges in the field of multi-agent reinforcement learning and proposes a systematic coordination framework, HYGMA. Traditional multi-agent systems have inherent limitations in modeling dynamic relationships and efficiently organizing information exchange, especially when agents need to adjust the collaboration mode according to environmental changes. HYGMA realizes the adaptive learning and evolution mechanism of agent relationships through dynamic spectral clustering and hypergraph networks. This framework effectively breaks through the limitation that the traditional graph structure can only express binary relations. It innovatively applies the hypergraph theory to the representation of high-order relations in multi-agent systems and combines the adaptive attention mechanism to achieve accurate selective information processing, providing a theoretically complete coordination method for complex agent systems. Based on mathematics, the research has established a strict theoretical connection between the dynamic formation of groups and learning performance, providing a solid theoretical support for the optimal design of multi-agent systems.
The International Conference on Machine Learning (ICML) is one of the most authoritative and influential international academic conferences in the field of machine learning. It was founded in 1980 and is hosted by the International Society for Machine Learning (IMLS). According to the latest global academic journal and conference impact ranking released by Google Scholar (covering all subject areas), ICML ranks 17th, with an h5 index of 268. ICML, along with NeurIPS and ICLR, is known as one of the "three most challenging, high-level and influential conferences" in the field of artificial intelligence. It is recognized as an A-class international conference in the field of artificial intelligence by the China Computer Federation (CCF). ICML is renowned for its strict requirements for theoretical contributions and methodological innovations, enjoying an extremely high reputation in the academic community. ICML25 received a total of 12,107 valid submissions and ultimately accepted 3,260, with an acceptance rate of 26.9%. The acceptance of this paper marks a breakthrough in the top-level conferences of our college in the field of artificial intelligence.