Game stage classification using machine learning and fuzzy sets for the game of Othello

Author

Lawrence R. Webb

Date of Award

8-1988

Degree Type

Thesis

Degree Name

Master of Science

Major

Computer Science

Major Professor

Maria Zemankova

Committee Members

David C. Mutchler

Abstract

Can the performance of a program to play Othello be improved using machine learning techniques? In this research two paths of research were pursued to attempt to answer this question.

In the first method, a program playing an uninformed strategy competed against a program playing a fully informed weighted square strategy. Locations occupied by the winner of each game were used to modify the weightings of the uninformed strategy. After a series of games the uninformed strategy was able to compete favorably against the expert strategy. This showed improvement in the performance of the learning program when competing against an informed strategy. Low and high thresholds of acceptance for location weightings were used showing that the high threshold produced a better playing program.

Once it was shown that machine learning can be used to improve the performance of an Othello playing program, the machine learning technique was implemented for a second testing method. By identifying characteristics of opening, middle and end game situations, game board configurations at each turn in the game were classified into these sets using fuzzy set theory. Different strategies at each stage of the game were utilized to maximize the performance of the program at each of these stages.The parameters of these fuzzy sets were automatically modified by the program at the end of each game. By adjusting the point in the game when the opening game strategy stopped playing and the middle game strategy began, and by adjusting the point in the game when the middle game strategy stopped playing and the end game strategy began, an improved strategy was defined.

This study shows that the performance of a program playing Othello can be improved using even simple machine learning techniques. Combining fuzzy set theory with machine learning to solve game theory problems is also shown to be an effective technique to improve the quality of play when exact quantifications of characteristics of a game are uncertain.

Recommended Citation

Webb, Lawrence R., "Game stage classification using machine learning and fuzzy sets for the game of Othello. " Master's Thesis, University of Tennessee, 1988.
https://trace.tennessee.edu/utk_gradthes/13370