The evolution of neural networks for decision making in non-cooperative repetitive games

Classic game theory analyzes the interactions between individuals (Players) under assumptions of perfect rationality and homogeneity. Nevertheless, new theories have arisen; such as evolutionary game theory. The evolutionary game theory is not based upon assumptions of perfect rationality, but under processes of Darwinian natural selection. This work portrays the evolutionary process of neural networks (perceptron, a radial basis network) using genetic algorithms for the learning of decision making strategies in non-cooperative repetitive games, in which the parameters to set up the topology of the networks are obtained experimentally. Results obtained through the evolutionary process of neural networks are comparable to the ones obtained on literature using genetic algorithms and particle swarms for games such as: Prisoner's Dilemma, Chicken Games and Stag Hunt.