Comparing Randomization Strategies for Search-Control Parameters in Monte-Carlo Tree Search

Chiara F. Sironi; Mark H. M. Winands

doi:10.1109/CIG.2019.8848056

Comparing Randomization Strategies for Search-Control Parameters in Monte-Carlo Tree Search

^*Corresponding author for this work

Networks and Strategic Optimization

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

45 Downloads (Pure)

Abstract

Monte-Carlo Tree Search (MCTS) has been applied successfully in many domains. Previous research has shown that adding randomization to certain components of MCTS might increase the diversification of the search and improve the performance. In a domain that tackles many games with different characteristics, like General Game Playing (GGP), trying to diversify the search might be a good strategy. This paper investigates the effect of randomizing search-control parameters for MCTS in GGP. Four different randomization strategies are compared and results show that randomizing parameter values before each simulation has a positive effect on the search in some of the tested games. Moreover, parameter randomization is compared with on-line parameter tuning.

Original language	English
Title of host publication	2019 IEEE Conference on Games (CoG)
Publisher	IEEE
Pages	1-8
Number of pages	8
ISBN (Print)	9781728118840
DOIs	https://doi.org/10.1109/CIG.2019.8848056
Publication status	Published - Aug 2019
Event	IEEE Conference on Games (CoG) 2019 - People's Palace, Queen Mary University of London, London, United Kingdom Duration: 20 Aug 2019 → 23 Aug 2019

Publication series

Series	IEEE Conference on Computational Intelligence and Games
ISSN	2325-4270

Conference

Conference	IEEE Conference on Games (CoG) 2019
Abbreviated title	COG
Country/Territory	United Kingdom
City	London
Period	20/08/19 → 23/08/19

Keywords

Monte-Carlo tree search
search-control parameter
randomization
General Game Playing

Access to Document

10.1109/CIG.2019.8848056

Full text Final published version, 291 KBLicence: Taverne

https://ieeexplore.ieee.org/document/8848056/

Cite this

@inproceedings{5db8ae9bfa7e4a87b4d9deba32fb320b,

title = "Comparing Randomization Strategies for Search-Control Parameters in Monte-Carlo Tree Search",

abstract = "Monte-Carlo Tree Search (MCTS) has been applied successfully in many domains. Previous research has shown that adding randomization to certain components of MCTS might increase the diversification of the search and improve the performance. In a domain that tackles many games with different characteristics, like General Game Playing (GGP), trying to diversify the search might be a good strategy. This paper investigates the effect of randomizing search-control parameters for MCTS in GGP. Four different randomization strategies are compared and results show that randomizing parameter values before each simulation has a positive effect on the search in some of the tested games. Moreover, parameter randomization is compared with on-line parameter tuning.",

keywords = "Monte-Carlo tree search, search-control parameter, randomization, General Game Playing",

author = "Sironi, {Chiara F.} and Winands, {Mark H. M.}",

note = "Funding Information: This work is funded by the Netherlands Organisation for Scientific Research (NWO) in the framework of the project GoGeneral, grant number 612.001.121. Publisher Copyright: {\textcopyright} 2019 IEEE.; IEEE Conference on Games (CoG) 2019 , COG ; Conference date: 20-08-2019 Through 23-08-2019",

year = "2019",

month = aug,

doi = "10.1109/CIG.2019.8848056",

language = "English",

isbn = "9781728118840",

series = "IEEE Conference on Computational Intelligence and Games",

publisher = "IEEE",

pages = "1--8",

booktitle = "2019 IEEE Conference on Games (CoG)",

address = "United States",

}

Comparing Randomization Strategies for Search-Control Parameters in Monte-Carlo Tree Search. / Sironi, Chiara F.; Winands, Mark H. M.
2019 IEEE Conference on Games (CoG). IEEE, 2019. p. 1-8 (IEEE Conference on Computational Intelligence and Games).

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

TY - GEN

T1 - Comparing Randomization Strategies for Search-Control Parameters in Monte-Carlo Tree Search

AU - Sironi, Chiara F.

AU - Winands, Mark H. M.

N1 - Funding Information: This work is funded by the Netherlands Organisation for Scientific Research (NWO) in the framework of the project GoGeneral, grant number 612.001.121. Publisher Copyright: © 2019 IEEE.

PY - 2019/8

Y1 - 2019/8

N2 - Monte-Carlo Tree Search (MCTS) has been applied successfully in many domains. Previous research has shown that adding randomization to certain components of MCTS might increase the diversification of the search and improve the performance. In a domain that tackles many games with different characteristics, like General Game Playing (GGP), trying to diversify the search might be a good strategy. This paper investigates the effect of randomizing search-control parameters for MCTS in GGP. Four different randomization strategies are compared and results show that randomizing parameter values before each simulation has a positive effect on the search in some of the tested games. Moreover, parameter randomization is compared with on-line parameter tuning.

AB - Monte-Carlo Tree Search (MCTS) has been applied successfully in many domains. Previous research has shown that adding randomization to certain components of MCTS might increase the diversification of the search and improve the performance. In a domain that tackles many games with different characteristics, like General Game Playing (GGP), trying to diversify the search might be a good strategy. This paper investigates the effect of randomizing search-control parameters for MCTS in GGP. Four different randomization strategies are compared and results show that randomizing parameter values before each simulation has a positive effect on the search in some of the tested games. Moreover, parameter randomization is compared with on-line parameter tuning.

KW - Monte-Carlo tree search

KW - search-control parameter

KW - randomization

KW - General Game Playing

U2 - 10.1109/CIG.2019.8848056

DO - 10.1109/CIG.2019.8848056

M3 - Conference article in proceeding

SN - 9781728118840

T3 - IEEE Conference on Computational Intelligence and Games

SP - 1

EP - 8

BT - 2019 IEEE Conference on Games (CoG)

PB - IEEE

T2 - IEEE Conference on Games (CoG) 2019

Y2 - 20 August 2019 through 23 August 2019

ER -