Learning Approximate Nash Equilibria in Cooperative Multi-Agent Reinforcement Learning via Mean-Field Subsampling

Researchers propose ALTERNATING-MARL, a new framework for cooperative multi-agent reinforcement learning that enables a global agent to learn with massive populations under communication constraints. The method achieves approximate Nash equilibrium convergence while only observing a subset of local agent states, with applications in multi-robot control and federated optimization.

• →ALTERNATING-MARL framework enables cooperative learning between one global agent and many local agents under strict observability constraints.
• →The method achieves O(1/√k)-approximate Nash Equilibrium convergence where k is the number of observed local agents.
• →The approach separates sample complexity between joint state space and action space, improving computational efficiency.
• →Framework has practical applications in multi-robot control systems and federated optimization scenarios.
• →The research addresses real-world challenges in large-scale networked systems with centralized decision makers.