Monte-Carlo Tree Search in chess endgames

Andraž Kohne (2012) Monte-Carlo Tree Search in chess endgames. EngD thesis.

Preview

PDF Download (636Kb)

Abstract

The Monte-Carlo Tree Search (MCTS) algorithm has in recent years captured the attention of many researchers due to its notable success in the game of Go. In spite of this success, so far it has not been used much in the game of chess. In this thesis, we attempt to apply MCTS to chess endgames. The reason for this is the existence of chess tablebases, i.e. databases that provide an exact value of each chess board position in terms of distance to mate. With this information at disposal we are able to measure more objectively the quality of play, and thus assess how well the Monte-Carlo Tree Search algorithm performs. We propose some guidelines about how the algorithm should be configured to achieve good performance in chess endgames. The question remains whether our findings are applicable to other domains as well.

Actions (login required)