Searching Game Trees Under Memory Constraints

No. 1: Agents, AI in Art and Entertainment, Knowledge Representation, and Learning

Volume

Proceedings of the AAAI Conference on Artificial Intelligence, 13

Game-Tree Search

The best-first game-tree search algorithm SSS* has greater pruning power than the depth-first algorithm Alpha-Beta. Yet it is seldom used in practice because it is slow in execution and requires substantial memory. Variants of SSS* have been proposed in recent years that overcome some, but not all, of its limitations. The recursive controlled-memory best-first search scheme MemSSS* described here is a new derivative of SSS* that compares favourably with Alpha-Beta in respect of all three major performance measures, namely, pruning power, running time and memory needs. MemSSS* improves upon an earlier controlled-memory algorithm IterSSS* which has most of the desired properties but is slow in execution.