Topics in Multiagent Learning

3-0-9 units · Graduate

Examines the foundations of multi-agent systems from a combined game-theoretic, optimization, and learning-theoretic perspective. Studies settings involving multiple learning agents with distinct objectives, where standard training methods may induce nonstationary or unstable behaviors. Introduces fundamental models including matrix games, stochastic games, and imperfect-information games. Studies modern settings deviating from standard models wherein strategies and utilities are represented using deep neural networks. Presents equilibrium concepts, equilibrium computation methods, and equilibrium learning methods. Studies the complexity of solution concepts. Discusses applications in machine learning, including multi-agent reinforcement learning, adversarial learning, AI-AI interactions, and strategic decision-making systems underlying game-playing agents in domains such as Go, poker, Diplomacy, and Stratego.