Effective Decision Making by Self-evaluation in the Multi-agent Environment

Abstract

Generally, in multi-agent systems, there are close relations between behavior of each individual agent and the group of agents as a whole, so a certain information about the relative state of each agent in the group may be hided in each agent behavior. If this information can be extracted, each agent has the possibility to improve its state by seeing only its own behavior without seeing other agents’ behaviors. In this paper, we focus on “power-law” which is interesting character seen in the behavior of each node of various kinds of networks as one of such information. Up to now, we have already found that power-law can be seen in the efficiently behaving agents in Minority Game which is the competitive multi-agent simulation environment. So, in this paper we have verified whether it is possible for each agent in the game to improve its state by seeing only its own behavior, and confirmed that the performance gain was actually possible. Keyword: Minority game, Indirect coordination, Power law.