Evolving and training of Neural Network to Play DAMA Board Game Using NEAT Algorithm

Abstract

Neuroevolutionary algorithms, such as NeuroEvolution of Augmenting Topologies (NEAT) in Machine Learning (ML) methods, are utilized for training and playing computer games due to increased research in Artificial Intelligence (AI) field. NEAT is a genetic algorithm for the generation of evolving artificial neural networks. In this paper, a new study is presented. A Dama board game is designed, and the NEAT algorithm is implemented to develop and train the populations of neural networks for playing the game efficiently. Different inputs and outputs for the network are used, and various network sizes are tried for the game to reach or pass the human level. This paper aims to make a neural network that plays a Dama game like humans or is close by training different neural networks for many generations. The experimental results show that neural networks have been trained for several thousands of generations, and they have played more than one million games. It is concluded that using more input to handle information is better for the learning process. It is also found that a set of values for NEAT parameters is suitable for extensive neural networks like those used in this paper. Povzetek: je predstavljena nova študija. Zasnovana je tradicionalna igra, imenovana "Dama", in implementiran je algoritem NEAT za razvoj in usposabljanje populacij nevronskih omrežij, da igrajo igro, prvič, na način, ki presega človeške zmožnosti.