Learning in Bebe¹²

Article type: Research Article

Authors: Scherzer, T. | Scherzer, L. | Tjaden Jr., D.

Affiliations: Hoffman Estates, Illinois

Note: [1] ‘Learning in Bebe’ by T. Scherzer, L. Scherzer and D. Tjaden Jr. (1990) has been distinguished by the 1990/1991 Mephisto Best-Publication Award. It appears here in a version abridged and edited from its original disclosure as Chapter 12 in Computers, Chess, and Cognition (eds. T. Anthony Marsland and Jonathan Schaeffer), Springer-Verlag, New York. ISBN 0-387-97415-6 and ISBN 3-540-97415-6. Permission to publish in this form was graciously given by Springer-Verlag, New York.

Note: [2] The abridgement is the authors’; slight editing changes have been made in order to bring the publication into line with this Journal’s editorial practice and in order to promote its readability as a self-contained article.

Abstract: Determinism in chess programs implies the risk of repeating one’s losses when the opponent copies a previous winning sequence of moves. Preserving the transposition table between games counters this effect by disconcerting opponents. In order to have efficient action against repeating successful lines, a form of long-term memory is introduced, making the program’s experience heritable in the longer run. This paper provides a partially affirmative answer to the question whether this technique is effective: in crude terms it is in many games against the same opponent and it cannot be determined what its effects are against dissimilar opponents.

DOI: 10.3233/ICG-1991-14402

Journal: ICGA Journal, vol. 14, no. 4, pp. 183-191, 1991