Faculty Advisor

Sarkozy, Gabor N

Center

Budapest, Hungary

Abstract

Since its introduction in 2006, Monte-Carlo Tree Search has been a major breakthrough in computer Go. Performance of an MCTS engine is highly dependent on the quality of its simulations, though despite this, simulations remain one of the most poorly understand aspects of MCTS. In this paper, we explore in-depth the simulations policy of Pachi, an open-source computer Go agent. This research attempts to better understand how simulation policies affect the overall performance of MCTS, building on prior work in the field by doing so. Through this research we develop a deeper understanding of the underlying components in Pachi's simulation policy, which are common to many modern MCTS Go engines, and evaluate the metrics used to measure them.

Publisher

Worcester Polytechnic Institute

Date Accepted

April 2014

Major

Computer Science

Major

Physics

Project Type

Major Qualifying Project

Accessibility

Unrestricted

Advisor Department

Computer Science

Project Center

Budapest, Hungary

Share

COinS