In this paper, we focus on models dealing with interactions between players playing a coordination game in networks. For a detailed examination of equilibrium selection, we employ agent-based simulation analysis, and consider the influence of network structures on interaction of players in networks. In a newly developed simulation system, agents repeatedly play the games, and they can adaptively select strategies through this learning process. Decision making and learning mechanism of agents are implemented by using a neural network and the genetic algorithm. As a result of the simulations where the values of the parameters of the networks and the coordination games are varyingly arranged, we find that the risk-dominant strategy in a random network is superiority over the payoff-dominant one, and the payoff-dominant one in a local network is superiority over the risk-dominant one, compared to the other networks. Moreover, it is also found that depending on the situation, the behavior of agents in a small-world network exhibits properties similar to that of agents in a random network or in a local one.
View full abstract