PROGRESSIVE STRATEGIES FOR MONTE-CARLO TREE SEARCH

被引:199
作者
Chaslot, Guillaume M. J-B. [1 ]
Winands, Mark H. M. [1 ]
Van den Herik, H. Jaap [1 ]
Uiterwijk, Jos W. H. M. [1 ]
Bouzy, Bruno [2 ]
机构
[1] Univ Maastricht, Fac Humanities & Sci, MICC IKAT Games & AI Grp, POB 616, NL-6200 MD Maastricht, Netherlands
[2] Univ Paris 5 Descartes, Ctr Rech Informat Paris 5, F-75270 Paris 06, France
关键词
Monte-Carlo Tree Search; heuristic search; Computer Go;
D O I
10.1142/S1793005708001094
中图分类号
O1 [数学];
学科分类号
0701 ; 070101 ;
摘要
Monte-Carlo Tree Search (MCTS) is a new best-first search guided by the results of Monte-Carlo simulations. In this article, we introduce two progressive strategies for MCTS, called progressive bias and progressive unpruning. They enable the use of relatively time-expensive heuristic knowledge without speed reduction. Progressive bias directs the search according to heuristic knowledge. Progressive unpruning first reduces the branching factor, and then increases it gradually again. Experiments assess that the two progressive strategies significantly improve the level of our Go program Mango. Moreover, we see that the combination of both strategies performs even better on larger board sizes.
引用
收藏
页码:343 / 357
页数:15
相关论文
共 21 条
[1]   Associating domain-dependent knowledge and Monte Carlo approaches within a Go program [J].
Bouzy, B .
INFORMATION SCIENCES, 2005, 175 (04) :247-257
[2]   Computer go: An AI oriented survey [J].
Bouzy, B ;
Cazenave, T .
ARTIFICIAL INTELLIGENCE, 2001, 132 (01) :39-103
[3]  
BOUZY B, 2005, IEEE 2005 S COMP INT, P176
[4]  
BOUZY B, 2006, IEEE 2006 S COMP INT, P187
[5]  
Bouzy B, 2003, ADV COMPUTER GAMES, P159
[6]  
Brugmann Bernd, 1993, TECHNICAL REPORT
[7]   Deep blue [J].
Campbell, M ;
Hoane, AJ ;
Hsu, FH .
ARTIFICIAL INTELLIGENCE, 2002, 134 (1-2) :57-83
[8]  
Cazenave T, 2007, ICGA J, V30, P35
[9]  
Chaslot G., 2006, PROC 18 BENELUX C AR, P91
[10]  
Coquelin P.-A., 2007, BANDIT ALGORIT UNPUB