A learning framework for zero-knowledge game playing agents

摘要: The subjects of perfect information games, machine learning and computational intelligence combine in an experiment that investigates a method to build the skill game-playing agent from zero game knowledge. playing is determined by two aspects, first quantity quality knowledge it uses second aspect its search capacity. This thesis introduces novel representation language combines symbols numeric elements capture Insofar concerned, extension existing knowledge-based developed. Empirical tests show improvement over alpha-beta, especially conditions where may be weak. Current techniques as applied agents reviewed. From these framework established. data-mining algorithm, ID3, technique, Particle Swarm Optimisation (PSO), form key components this framework. classification trees produced ID3 subjected new post-pruning processes specifically defined for mentioned language. Different combinations pruning are tested dominant combination chosen use As PSO, tournaments introduced relative fitness function. A variety alternative tournament methods described some experiments conducted evaluate these. final design decisions incorporated into configuration, on Checkers variations Checkers. These has occurred, but also highlights need further development experimentation. Some ideas regard concludes thesis.