Monte-Carlo Tactic Search

Mr. Peigang Zhang

Computer Science Dept., UNC Charlotte

Friday, November 2, 2007

3:00 p.m. - 4:00 p.m.

Woodward Hall, Room 106

Complete Description:

Go is the last board game that is not able to provide a strong computer player, and this is a great challenge to AI researchers. Solving Go tactic problems is an important step towards building a strong 19x19 Go playing program. In this talk, I will introduce Monte-Carlo Tactic Search. Standard Monte-Carlo UCT tree search algorithm is modified and extended to provide an efficient Go capturing problem solver. Experiment results show that this method outperforms traditional game tree search methods in solving capturing problems in Go. Bio: Mr. Peigang Zhang is a Ph. D student (advisor: Dr. Keh-Hsun Chen) in the department of computer science at UNC-Charlotte. He is a member of the Heuristic Search Lab. His research area includes Computer Go, Monte-Carlo Tree Search, Game Tree Search and Machine Learning.