Multi-Agent and Game-Theoretic RL
A study of QMIX and zero-sum games, where multiple agents interact in cooperative or competitive settings.
Chapter 22: Multi-Agent RL Basics Problem Definition and Research Motivation Research Directions: Cooperative, competitive, mixed settings Frameworks: MARL challenges, agent modeling Future Study: Scalable MARL, real-world coordination References Chapter 23: Decentralized and Centralized Training Independent Q-learning, QMIX, WQMIX COMA, QTRAN, CollaQ, ATOC Centralized Training with Decentralized Execution References Chapter 24: Game-Theoretic RL (Nash equilibria, Stackelberg games, mean-field games) Chapter 25: Zero-Sum Games Problem Definition and Research Motivation Research History: Minimax, AlphaGo, poker solvers Algorithms: CFR, fictitious play, neural MCTS Future Prospects: General-sum extensions, real-time games References Chapter 26: Emergent Behaviors in MARL (Coordination, communication, social dilemmas)