Scalable Actor-Critic Reinforcement Learning for Mean Field Control Games
A deep reinforcement learning method that approximates equilibria of Mean Field Control Games by recasting the infinite-agent stochastic control problem as a Markov Decision Process in which a representative agent learns against an evolving population distribution. It replaces direct solution of the coupled Hamilton-Jacobi-Bellman and Fokker-Planck equations with an actor-critic scheme made scalable through batching, target networks, PPO, generalized advantage estimation, and entropy regularization.
2501.00052
Mean Field Control Games (MFCGs) provide a powerful theoretical framework for analyzing systems of infinitely many interacting agents, blending elements from Mean Field Games (MFGs) and Mean Field Co…