Effects of ensemble action selection with different usage of player's memory resource on the evolution of cooperative strategies for iterated prisoner's dilemma game

Abstract

In our previous study, we proposed an ensemble action selection model where each player has multiple strategies with different memory length for the iterated prisoner's dilemma (IPD) game. An action was suggested by each strategy based on its memory about opponent's single, two or three actions. Majority vote was used for action selection. Under these settings, the evolution of cooperation was examined for various ensembles (i.e., various combinations of strategies). In this paper, we extend our ensemble model to a more general case where strategies have different memory usage. Each strategy of a player has a memory of opponent's and/or player's previous actions. For example, a memory of a strategy can be opponent's single and player's two actions. Another strategy's memory can be player's three actions. Various combinations of strategies for ensemble action selection are examined in this paper. It is shown through computational experiments that the use of ensemble action selection enhances the evolution of cooperation. It is also shown that no cooperation is evolved among strategies with no memory about opponent's actions. An interesting observation is that cooperation is evolved by players with the combination of the following three strategies: two strategies with no memory about opponent's actions, and a single strategy with a memory of both player's and opponent's actions.