Monte-Carlo simulation balancing applied to 9 × 9 Go

Shih Chieh Huang, Rémi Coulom, Shun Shii Lin

Research output: Contribution to journalArticle

1 Citation (Scopus)

Abstract

Simulation balancing is a new technique to tune parameters of a playout policy for a Monte-Carlo game-playing program. So far, this algorithm had only been tested in an artificial setting: it was limited to 5 × 5 and 6 × 6 Go, and required a stronger external program that served as a supervisor. In this article, the effectiveness of simulation balancing is demonstrated in a realistic setting. A state-of-the-art program, ERICA, learned an improved playout policy on the 9 × 9 board, without requiring any external expert to provide position evaluations. Evaluations were collected by letting the program analyze positions by itself. This evaluation was run with playout parameters estimated by the minorization-maximization (MM) algorithm. Thanks to simulation balancing, ERICA'S playing strength was improved from a winning rate of 69% (with playout parameters trained by MM) to 78% (with playout parameters trained by SB) against FUEGO 0.4.

Original languageEnglish
Pages (from-to)191-201
Number of pages11
JournalICGA Journal
Volume33
Issue number4
DOIs
Publication statusPublished - 2010 Dec

Fingerprint

Supervisory personnel
Monte Carlo simulation

ASJC Scopus subject areas

  • Computer Science (miscellaneous)
  • Computational Mechanics
  • Human-Computer Interaction
  • Computer Graphics and Computer-Aided Design

Cite this

Monte-Carlo simulation balancing applied to 9 × 9 Go. / Huang, Shih Chieh; Coulom, Rémi; Lin, Shun Shii.

In: ICGA Journal, Vol. 33, No. 4, 12.2010, p. 191-201.

Research output: Contribution to journalArticle

Huang, Shih Chieh ; Coulom, Rémi ; Lin, Shun Shii. / Monte-Carlo simulation balancing applied to 9 × 9 Go. In: ICGA Journal. 2010 ; Vol. 33, No. 4. pp. 191-201.
@article{0107692592e84e0892bc7103b73209e1,
title = "Monte-Carlo simulation balancing applied to 9 × 9 Go",
abstract = "Simulation balancing is a new technique to tune parameters of a playout policy for a Monte-Carlo game-playing program. So far, this algorithm had only been tested in an artificial setting: it was limited to 5 × 5 and 6 × 6 Go, and required a stronger external program that served as a supervisor. In this article, the effectiveness of simulation balancing is demonstrated in a realistic setting. A state-of-the-art program, ERICA, learned an improved playout policy on the 9 × 9 board, without requiring any external expert to provide position evaluations. Evaluations were collected by letting the program analyze positions by itself. This evaluation was run with playout parameters estimated by the minorization-maximization (MM) algorithm. Thanks to simulation balancing, ERICA'S playing strength was improved from a winning rate of 69{\%} (with playout parameters trained by MM) to 78{\%} (with playout parameters trained by SB) against FUEGO 0.4.",
author = "Huang, {Shih Chieh} and R{\'e}mi Coulom and Lin, {Shun Shii}",
year = "2010",
month = "12",
doi = "10.3233/ICG-2010-33402",
language = "English",
volume = "33",
pages = "191--201",
journal = "ICGA Journal",
issn = "1389-6911",
publisher = "Tilburg Centre for Cogination and Communication",
number = "4",

}

TY - JOUR

T1 - Monte-Carlo simulation balancing applied to 9 × 9 Go

AU - Huang, Shih Chieh

AU - Coulom, Rémi

AU - Lin, Shun Shii

PY - 2010/12

Y1 - 2010/12

N2 - Simulation balancing is a new technique to tune parameters of a playout policy for a Monte-Carlo game-playing program. So far, this algorithm had only been tested in an artificial setting: it was limited to 5 × 5 and 6 × 6 Go, and required a stronger external program that served as a supervisor. In this article, the effectiveness of simulation balancing is demonstrated in a realistic setting. A state-of-the-art program, ERICA, learned an improved playout policy on the 9 × 9 board, without requiring any external expert to provide position evaluations. Evaluations were collected by letting the program analyze positions by itself. This evaluation was run with playout parameters estimated by the minorization-maximization (MM) algorithm. Thanks to simulation balancing, ERICA'S playing strength was improved from a winning rate of 69% (with playout parameters trained by MM) to 78% (with playout parameters trained by SB) against FUEGO 0.4.

AB - Simulation balancing is a new technique to tune parameters of a playout policy for a Monte-Carlo game-playing program. So far, this algorithm had only been tested in an artificial setting: it was limited to 5 × 5 and 6 × 6 Go, and required a stronger external program that served as a supervisor. In this article, the effectiveness of simulation balancing is demonstrated in a realistic setting. A state-of-the-art program, ERICA, learned an improved playout policy on the 9 × 9 board, without requiring any external expert to provide position evaluations. Evaluations were collected by letting the program analyze positions by itself. This evaluation was run with playout parameters estimated by the minorization-maximization (MM) algorithm. Thanks to simulation balancing, ERICA'S playing strength was improved from a winning rate of 69% (with playout parameters trained by MM) to 78% (with playout parameters trained by SB) against FUEGO 0.4.

UR - http://www.scopus.com/inward/record.url?scp=79960538544&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=79960538544&partnerID=8YFLogxK

U2 - 10.3233/ICG-2010-33402

DO - 10.3233/ICG-2010-33402

M3 - Article

AN - SCOPUS:79960538544

VL - 33

SP - 191

EP - 201

JO - ICGA Journal

JF - ICGA Journal

SN - 1389-6911

IS - 4

ER -