Learning of Position Evaluation
A. Leouski · 2001
Conventional Othello programs are based on a thorough analysis of the game, crafting sophisticated evaluation functions and supervised learning techniques with use of large expert-labeled game databases. This paper presents an alternative by training a neural network to evaluate Othello positions via temporal difference (TD) learning. The approach is based on network architecture that reflects the spatial and temporal organization of the domain. The network begins as a random network and by simply playing against itself achieves an intermediate level of play.