MONTE-CARLO GO WITH KNOWLEDGE-GUIDED SIMULATIONS¹

Article type: Research Article

Authors: Chen, Keh-Hsun | Zhang, Peigang²

Affiliations: Charlotte, USA

Note: [1] This article is a revised version of a contribution with the same title to the Computer Games Workshop 2007 held in Amsterdam, The Netherlands, June 15-17, 2007. The revised version has been subjected to the Journal’s normal refereeing procedure.

Note: [2] Department of Computer Science, University of North Carolina at Charlotte, Charlotte, NC 28223, USA. Email: {chen,pzhang1}@uncc.edu

Abstract: This article explores light Go-domain knowledge items. Light items are computable in a speed suitable to be used in random simulations in Monte-Carlo (MC) Go. For our purpose, we evaluated and filtered the desired knowledge items for simulation games in Monte-Carlo Tree Search (MCTS) through extensive testing against GNU GO. As a result, we arrived at an appropriate collection of five light basic Go knowledge items to be included in the simulation routine of GO INTELLECT. This collection of knowledge items improved the program performance by 30 percentage points in winning rates over using bare-bone solid-eye-avoidance knowledge alone in random simulations when tested against GNU GO.

DOI: 10.3233/ICG-2008-31202

Journal: ICGA Journal, vol. 31, no. 2, pp. 67-76, 2008