Evolution of Neural Networks to Play the Game of Dots-and-Boxes

摘要

Dots-and-Boxes is a child's game which remains analytically unsolved. Weimplement and evolve artificial neural networks to play this game, evaluatingthem against simple heuristic players. Our networks do not evaluate or predictthe final outcome of the game, but rather recommend moves at each stage.Superior generalisation of play by co-evolved populations is found, and acomparison made with networks trained by back-propagation using simpleheuristics as an oracle.