Abstract

The area of game player strategies was first examined in computer science by Charles Babbage in the 1840's and since then, in an endeavour to improve game search techniques, it has been looked at extensively by researchers in computer science. Because the continued use of brute force techniques produces diminishing returns for performance, recent interest has focused on learning the strategies of expert players. This means more intelligent programs must be designed if greater performance gains are to be achieved. This paper will look at improving the ability to learn the strategies in games, with emphasis on the game of Chess. Using the ideas of Jansen et al. (2000) as a departure point, this thesis examines ways of improving machine learning of strategies in the game of chess. This thesis investigates three main areas. First, evaluation functions are revisited and then secondly the actual searching technique is inferred. The third part involves inferring both the search strategy and the evaluation technique simultaneously.