A connectionist approach to heuristically pruning large search trees

Artificial neural systems (ANS) technology is a form of information processing based on models derived from the observed behavior in neurobiological systems. While not necessarily accurate brain models, neural networks appear to be ideally suited for parallel state-space search applications, particularly when the state-space can be represented as regularly occurring patterns.;In this thesis I present an approach to using neural networks as a means of performing the state-space search in two variations of the two-player, zero-sum game of tic-tac-toe. The methods for encoding the game representation are described, as well as the dynamics that enable the ANS to successfully play the game. In the final implementation, the neural networks "suggest" likely winning strategies for a traditional tree-search algorithm, and the improvements obtained as a result of using the ANS to heuristically prune the game tree for the search algorithm are described.