A Chess Program that Uses its Transposition Table to Learn from Experience

Abstract

The transposition table is a common tool for conveying information between different branches of a computer-chess search tree. This paper discusses a technique, based on a method described by A. Samuel (1959), for accumulating selected information from many games and then utilizing it subsequently via the transposition table. This transfer of information constitutes a kind of learning from experience. Although position-specific, it is shown to be capable of a modest degree of generalization. A program was implemented to test this algorithm, and preliminary examples of its behavior are given, illustrating some of its capabilities and limitations.