Using evolutionary programing to create neural networks that are capable of playing tic-tac-toe

The use of evolutionary programming for adapting the design and weights of a multi-layer feedforward perceptron in the context of machine learning is described. Specifically, it is desired to evolve the structure and weights of a single hidden layer perceptron such that it can achieve a high level of play in the game tic-tac-toe without the use of heuristics or credit assignment algorithms. Conclusions from the experiments are offered regarding the relative importance of specific mutation operations, the necessity for credit assignment procedures, and the efficiency and effectiveness of evolutionary search