Evolutionary markov games based on neural network

Abstract

Based on the dynamic characteristics of evolutionary game and Markov process, this paper presents a dynamic decision model for evolutionary Markov games. In this model, players' strategy-choosing is mapped to a Markov decision process with payoffs, and transition probability is made by Boltzmann distribution. This paper uses neural network to simulate strategy-choosing in evolutionary Markov games, Experimental results show that the neural network can successfully simulate players' dynamic learning and actions in evolutionary Markov games. © 2009 Springer Berlin Heidelberg.