Artificial neural network using the training set of DTS for Pacman game

Abstract

Recently Pac-Man game has received some attention of AI researchers. Some artificial intelligence researches have been performed with the Pac-Man game. The Monte Carlo, UCT (Upper Confidence Bound for Tree) and DTS (Dynamically Expanding UCT Tree Search) method have some good performances when used in Pac-Man game. However, these methods are based on intensive computation and so they usually could not be used in multiplayer online games. This paper discusses the advantages of Artificial Neural Network (ANN) method compared with DTS. It uses the results of DTS - called training set. The advantage of ANN is that it uses much less CPU resources than DTS. So it can be used in multiplayer online games whose AI program is running on the server, or in some other situations that require low computational intensity. This paper aims to discuss one main issue about ANN method based on the training set provided by DTS: how to implement ANN using the training set of DTS.