Seminar on Computational Learning and Adaptation


Experience-Based Chess Play

Robert Levinson
Department of Computer Science
University of California, Santa Cruz

We begin by reviewing the state of the art in computer chess and game playing in general. Despite the obvious success of chess computers, which compete on a level approximately equal to the best humans, one can always construct chess positions in which the computer is almost clueless. Moreover, typical programs do not grasp the underlying structure of chess, and human chess players' use of patterns and learning is obviously more efficient. We present an alternative architecture, Morph, that incorporates technologies such as neural networks, genetic algorithms, graph matching, and representation change and that, with little human assistance, goes from random chess play to a rating over 1500, equivalent to beating 45 percent of tournament players. Although our combination and understanding of these technologies (and chess) is far from perfect, we believe the future is bright for their use in complex domains. The talk is suitable for a general computer science audience.

Date: Wednesday, April 6, 2005

Time: 4:15-5:30PM

Place: Gates 104


Return to the seminar schedule