This paper implements and analyzes four algorithms for improving computer play of the board game Go. These algorithms use machine pattern learning to find better Monte-Carlo simulation policies for use with Monte-Carlo Tree Search. Two of these algorithms maximize individual move strength, and two minimize overall simulation error. These algorithms are tested using UCT on 9x9 Go with 3x3 patterns.

• This report represents the work of one or more WPI undergraduate students submitted to the faculty as evidence of completion of a degree requirement. WPI routinely publishes these reports on its website without editorial or peer review.

Creator

• Anderson, David Adler

Publisher

• Worcester Polytechnic Institute

Identifier

• E-project-042909-035846

Advisor

• Selkow, Stanley M.
• Sarkozy, Gabor N

Year

• 2009

Center

• Budapest, Hungary Project Center - MQP

Sponsor

• MTA SZTAKI

Date created

• 2009-04-29

Location

• Budapest

Resource type

• Major Qualifying Project

Major

• Computer Science

Rights statement

• In Copyright