Academic Commons


The Expected-Outcome Model of Two-Player Games

Abramson, Bruce

This paper introduces a new, crisp definition of two-player evaluation functions. These functions calculate a node's expected-outcome value. or the probability that a randomly chosen leaf beneath it will represent a win. The utility of these values to game programs will be assessed by a series of experiments that compare the performance of expected-outcome functions with that of some popular, previously studied evaluators. To help demonstrate the domain-independence of these new functions, the experiments will be run on variants of several games, including tic-tac-toe, Othello, and chess. In addition, the paper outlines a. new probabilistic model of game-trees which involves rethinking many long-accepted assumptions in light of the newly defined expected-outcome functions.



More About This Work

Academic Units
Computer Science
Department of Computer Science, Columbia University
Columbia University Computer Science Technical Reports, CUCS-315-87
Published Here
December 7, 2011