Adding expert knowledge and exploration in Monte-Carlo tree search

Abstract

We present a new exploration term, more efficient than classical UCT-like exploration terms. It combines efficiently expert rules, patterns extracted from datasets, All-Moves-As-First values, and classical online values. As this improved bandit formula does not solve several important situations (semeais, nakade) in computer Go, we present three other important improvements which are central in the recent progress of our program MoGo. We show an expert-based improvement of Monte-Carlo simulations for nakade situations; we also emphasize some limitations of this modification. We show a technique which preserves diversity in the Monte-Carlo simulation, which greatly improves the results in 19x19. Whereas the UCB-based exploration term is not efficient in MoGo, we show a new exploration term which is highly efficient in MoGo. MoGo recently won a game with handicap 7 against a 9Dan Pro player, Zhou JunXun, winner of the LG Cup 2007, and a game with handicap 6 against a 1Dan pro player, Li-Chen Chien. © 2010 Springer-Verlag Berlin Heidelberg.