CSCI 4156

This course presents the theory and applications of reinforcement learning by asking: How do we teach computers to play games? Two basic themes will be developed: 1) Markov decision processes, dynamic programming and Monte Carlo Tree search, culminating in the development of the Temporal difference method. 2) Episodic policy search through evolutionary computation. Case studies will consider results from games of complete information (e.g. Back-gammon, Chess, Go) and incomplete information (e.g. FPS, StarCraft, Dota2, Poker). The role of self-play will also be considered.