Evolution of Cooperation in Spatial Prisoner’s Dilemma Game Based on Incremental Learning

The evolution of cooperation among intelligent agents is a fundamental issue in multi-agent systems. It is well accepted that the individual strategy-updating rules play a significant role in the cooperation dynamics on graphs. The imitation mechanisms account for a large proportion of these rules, in which an individual will choose a neighbor with higher payoff and follows its strategy. In this paper, we propose a strategy-updating rule based on incremental learning process for continuous prisoner’s dilemma game. Under our strategy-updating rule, each individual refreshes its decision according to original strategy (self-opinion) and new strategy learnt from one of neighbors (social-opinion). The simulation results show the incremental learning rule can enhance cooperation dramatically when individual has higher resistance to imitate others or lower payoff sensitivity. We also find that the incremental learning rule has greater influence when individual obtains fewer information of neighbors’ payoff. The reason behind the phenomena is also given. Our results may shed some light on how cooperative strategies are actually adopted and spread in spatial network.