Deep Reinforcement Learning for GO AI

Aranda Ramirez, Esteban; Graham, Thomas

Student Work

Deep Reinforcement Learning for GO AI

Public

Go is considered to be the most challenging classical game for artificial intelligence. Following the structure of Google’s AlphaGo which defeated the world’s best Go player in 2016, this project aimed to develop an intelligent player that could achieve human-level of play for TicTacToe, Othello, and most importantly Go. Using Python and the machine learning framework PyTorch, the team developed two tree search algorithms, Minimax and Monte Carlo Tree Search, and two neural networks with the goal of combining them into a single intelligent agent. Testing was only performed for TicTacToe and without combining algorithms. Despite the neural networks not achieving the same performance as the tree search algorithms, their improvement over time showcase the potential of machine learning, even with a limited amount of data and computational power for training.

This report represents the work of one or more WPI undergraduate students submitted to the faculty as evidence of completion of a degree requirement. WPI routinely publishes these reports on its website without editorial or peer review.

Creator

Publisher