Acquisition of the action strategy of a Multi-Agent System
(Prof. H. YUASA and Prof. T. ARAI)

In actual society, we are pressed for selection of various actions. Then, game theory was devised as a general theory of decision-making and action selection in a rational agent. In order for an agent to act well, it is important to “solve” game theory. That is, it is important to change a self-behavioral pattern according to a surrounding situation. However, in multi-agent system, each agent is environment for other agents. And they are going to optimize their action pattern according to the other’s action pattern. In such a situation, the optimal behavioral pattern does not necessarily exist. U-Mart (http://www.u-mart.econ.kyoto-u.ac.jp/) is mentioned as one of the action environment of such a multi-agent. U-Mart is an artificial futures trading market model. Originally it is proposed for research and education of a market. It is interesting also as a multi-agent's action environment.

By this research, we consider U-Mart is as action environment of a multi-agent system, and aim at acquiring certain knowledge out of an action strategy in multi-agent system. The validity of the action strategy in data analyses, such as modeling / wave let analysis of a market price, has so far been verified.

Keywords: Multi-Agent System, U-mart

Fig. 1 U-Mart System

Fig. 2 U-Mart simulator