Evolving intelligent game-playing agents : research article
N. Franken, Andries Petrus Engelbrecht · South African Computer Journal · 2004
Traditional game playing programs have relied on advanced search algorithms and hand-tuned evaluation functions to play 'intelligently'. A historical overview of these techniques is provided, followed by a revealing look at recent developments in coevolutionary strategies to facilitate game learning. The use of particle swarms in conjunction with neural networks to learn how to play tic-tac-toe is experimentally compared to current game learning research. The use of a new particle swarm neighbourhood structure and innovative board state representation show promising results that warrant further investigation to its application in more complex games like checkers.