抄録
We propose a metric for learning processes of agents as a useful tool in experimental analysis of multiagent systems. This method employs quantity of information for decision making which quantifies how decision making reduces uncertainty of relations between input and output. We evaluate this method in Artificial Stock Market which consists of many autonomous learning agents. The results show the proposed metric is valuable in the following two points. One is that the metric gives a criterion for determining the size of initial learning. The other is that the metric can be used to extract dominant sets of attributes in decision making. We conclude this metric will prove to be a significant tool to investigate the behavior of multiagent systems.
