Reinforcement Learning for Game Theory

Abstract

Reinforcement learning, especially deep reinforcement learning, has recently achieved impressive results in playing complex board games like chess and Go, as well as video games such as StarCraft II. However, there has been limited research into how these techniques work with strategic games from game theory. This project aims to create a reinforcement learning system that learns to play a repeated game, such as the iterated prisoner's dilemma, against itself and to compare its performance with traditional strategies.