A Comparison of Monte-Carlo Methods for Phantom Go

Throughout recent years, Monte-Carlo methods have considerably improved computer Go programs. In particular, Monte-Carlo Tree Search algorithms such as UCT have enabled significant advances in this domain. Phantom Go is a variant of Go which is complicated by the condition of imperfect information. This article compares four Monte-Carlo methods for Phantom Go in a self-play experiment: (1) Monte-Carlo evaluation with standard sampling, (2) MonteCarlo evaluation with all-as-first sampling, (3) UCT with late random opponent-move guessing heuristic, and (4) UCT with early probabilistic opponent-move guessing heuristic. Our experimental findings indicate that Monte-Carlo methods can be applied to Phantom Go effectively. Surprisingly, Monte-Carlo Tree Search performs comparable to Monte-Carlo evaluation but not much better.

[1]  Hiroyuki Iida,et al.  Solving Kriegspiel-like Problems: Exploiting a Transposition Table , 2000, J. Int. Comput. Games Assoc..

[2]  Erik van der Werf,et al.  AI techniques for the game of Go , 2001 .

[3]  Bruno Bouzy,et al.  Monte-Carlo Go Developments , 2003, ACG.

[4]  Paolo Ciancarini,et al.  Searching over Metapositions in Kriegspiel , 2004, Computers and Games.

[5]  V. S. Subrahmanian,et al.  Game-Tree Search with Combinatorially Large Belief States , 2005, IJCAI.

[6]  Bruno Bouzy,et al.  Monte-Carlo Go Reinforcement Learning Experiments , 2006, 2006 IEEE Symposium on Computational Intelligence and Games.

[7]  Tristan Cazenave,et al.  A Phantom-Go Program , 2006, ACG.

[8]  Bruno Bouzy,et al.  Monte-Carlo strategies for computer Go , 2006 .

[9]  Rémi Coulom,et al.  Efficient Selectivity and Backup Operators in Monte-Carlo Tree Search , 2006, Computers and Games.

[10]  Csaba Szepesvári,et al.  Bandit Based Monte-Carlo Planning , 2006, ECML.

[11]  Paolo Ciancarini,et al.  A Program to Play Kriegspiel , 2007, J. Int. Comput. Games Assoc..

[12]  Jahn-Takeshi Saito,et al.  Grouping Nodes for Monte-Carlo Tree Search , 2007 .