Learning to Play Draughts using Temporal Difference Learning with Neural Networks and Databases
J-P. Patist, Wiering · Utrecht University Repository (Utrecht University) · 2003
This paper describes several aspects of using temporal difference learning (TD) and neural networks to learn game evaluation functions, and the benefits of using databases.Experiments in tic-tac-toe and international draughts have been done to measure the effectiveness of using databases.The experiment of Tic-Tac-Toe showed that training from database games resulted in better play than learning from playing against a random player.In the experiment of learning international draughts, the program reached after just a few hours of training a better level than a strong computer playing program and occasionally drew a very strong program.Thus, using temporal difference learning and neural networks on database games is a time-efficient way to reach a considerable level of play.