Monte Carlo *-Minimax Search

Marc Lanctot; Abdallah Saffidine; Joel Veness; Christopher Archibald; Mark H M Winands

Monte Carlo *-Minimax Search

Marc Lanctot, Abdallah Saffidine, Joel Veness, Christopher Archibald, Mark H M Winands

Networks and Strategic Optimization

Research output: Chapter in Book/Report/Conference proceeding › Conference article in proceeding › Academic › peer-review

Abstract

This paper introduces Monte Carlo *-Minimax Search (MCMS), a Monte Carlo search algorithm for turned-based, stochastic, two-player, zero-sum games of perfect information. The algorithm is designed for the class of of densely stochastic games; that is, games where one would rarely expect to sample the same successor state multiple times at any particular chance node. Our approach combines sparse sampling techniques from MDP planning with classic pruning techniques developed for adversarial expectimax planning. We compare and contrast our algorithm to the traditional *-Minimax approaches, as well as MCTS enhanced with the Double Progressive Widening, on four games: Pig, EinStein W\"urfelt Nicht!, Can't Stop, and Ra. Our results show that MCMS can be competitive with enhanced MCTS variants in some domains, while consistently outperforming the equivalent classic approaches given the same amount of thinking time.

Original language	English
Title of host publication	Proceedings of the 23rd International Joint Conference on Artificial Intelligence
Pages	580-586
Number of pages	7
Publication status	Published - 2013
Event	Twenty-third International Conference on Artificial Intelligence - Beijing, China Duration: 3 Aug 2013 → 9 Aug 2013 http://ijcai-13.org/

Conference

Conference	Twenty-third International Conference on Artificial Intelligence
Abbreviated title	IJCAI-13
Country/Territory	China
City	Beijing
Period	3/08/13 → 9/08/13
Internet address	http://ijcai-13.org/

Access to Document

http://www.mendeley.com/research/monte-carlo-minimax-search

Cite this

@inproceedings{2e1a036727ea4539b96c8f894e0fc8f7,

title = "Monte Carlo *-Minimax Search",

abstract = "This paper introduces Monte Carlo *-Minimax Search (MCMS), a Monte Carlo search algorithm for turned-based, stochastic, two-player, zero-sum games of perfect information. The algorithm is designed for the class of of densely stochastic games; that is, games where one would rarely expect to sample the same successor state multiple times at any particular chance node. Our approach combines sparse sampling techniques from MDP planning with classic pruning techniques developed for adversarial expectimax planning. We compare and contrast our algorithm to the traditional *-Minimax approaches, as well as MCTS enhanced with the Double Progressive Widening, on four games: Pig, EinStein W\{"}urfelt Nicht!, Can't Stop, and Ra. Our results show that MCMS can be competitive with enhanced MCTS variants in some domains, while consistently outperforming the equivalent classic approaches given the same amount of thinking time.",

author = "Marc Lanctot and Abdallah Saffidine and Joel Veness and Christopher Archibald and Winands, {Mark H M}",

year = "2013",

language = "English",

pages = "580--586",

booktitle = "Proceedings of the 23rd International Joint Conference on Artificial Intelligence",

note = "Twenty-third International Conference on Artificial Intelligence , IJCAI-13 ; Conference date: 03-08-2013 Through 09-08-2013",

url = "http://ijcai-13.org/",

}

TY - GEN

T1 - Monte Carlo *-Minimax Search

AU - Lanctot, Marc

AU - Saffidine, Abdallah

AU - Veness, Joel

AU - Archibald, Christopher

AU - Winands, Mark H M

PY - 2013

Y1 - 2013

N2 - This paper introduces Monte Carlo *-Minimax Search (MCMS), a Monte Carlo search algorithm for turned-based, stochastic, two-player, zero-sum games of perfect information. The algorithm is designed for the class of of densely stochastic games; that is, games where one would rarely expect to sample the same successor state multiple times at any particular chance node. Our approach combines sparse sampling techniques from MDP planning with classic pruning techniques developed for adversarial expectimax planning. We compare and contrast our algorithm to the traditional *-Minimax approaches, as well as MCTS enhanced with the Double Progressive Widening, on four games: Pig, EinStein W\"urfelt Nicht!, Can't Stop, and Ra. Our results show that MCMS can be competitive with enhanced MCTS variants in some domains, while consistently outperforming the equivalent classic approaches given the same amount of thinking time.

AB - This paper introduces Monte Carlo *-Minimax Search (MCMS), a Monte Carlo search algorithm for turned-based, stochastic, two-player, zero-sum games of perfect information. The algorithm is designed for the class of of densely stochastic games; that is, games where one would rarely expect to sample the same successor state multiple times at any particular chance node. Our approach combines sparse sampling techniques from MDP planning with classic pruning techniques developed for adversarial expectimax planning. We compare and contrast our algorithm to the traditional *-Minimax approaches, as well as MCTS enhanced with the Double Progressive Widening, on four games: Pig, EinStein W\"urfelt Nicht!, Can't Stop, and Ra. Our results show that MCMS can be competitive with enhanced MCTS variants in some domains, while consistently outperforming the equivalent classic approaches given the same amount of thinking time.

M3 - Conference article in proceeding

SP - 580

EP - 586

BT - Proceedings of the 23rd International Joint Conference on Artificial Intelligence

T2 - Twenty-third International Conference on Artificial Intelligence

Y2 - 3 August 2013 through 9 August 2013

ER -