PhD Dissertation Defense by GENG Minghong | Scaling up Cooperative Multi-agent Reinforcement Learning

Scaling up Cooperative Multi-agent Reinforcement Learning

ABOUT THE TALK

Over the past decade, multi-agent reinforcement learning (MARL) has emerged as a powerful paradigm for enabling collaborative behaviors among autonomous agents within MAS to solve complex tasks. This dissertation discusses a critical scalability gap and addresses the question: How can we design multi-agent learning systems that simultaneously scale to large agent teams and extended temporal horizons while maintaining the generalizability for practical deployment? As a response, the dissertation makes four interconnected contributions that collectively advance the field from recognizing scalability limitations to implementing practical multi-agent frameworks: a comprehensive survey of scalable multi-agent reinforcement learning, a long-horizon multi-objective multi-agent reinforcement learning benchmark (MOSMAC), a hierarchical multi-agent learning framework with self-organizing neural networks (HiSOMA), and a hierarchical multi-agent learning framework integrating large language models (L2M2). Through these contributions, this dissertation establishes that hierarchical, heterogeneous, modular architectures provide a unified solution to both structural and temporal scalability in multi-agent systems.

SPEAKER BIOGRAPHY

GENG Minghong is a Ph.D. Candidate in Computer Science, under the supervision of Professor TAN Ah Hwee. His primary research explores methods for scaling multi-agent reinforcement learning (MARL) systems, specifically addressing coordination and learning challenges within large-scale, complex environments. His work on hierarchical multi-agent systems has been published in leading academic venues, including IJCAI (2025), AAMAS (2023, 2024, 2025), and ESwA. He is a recipient of the SMU Presidential Doctoral Fellowship. During his candidature, he took a half-year academic visit to BNRist, Tsinghua University. He previously earned his Master of IT in Business (MITB) from SCIS, SMU, and a Bachelor's degree from Southwestern University of Finance and Economics, majoring in Finance.