MACC'97 Session: Business Application of Multiagent Systems
Organizational-learning oriented Classifier System (OCS) is a new multiagent-based learning model with an evolutionary computational mechanism. It is composed of a production system, a reinforcement learning mechanism, and rule generation/exchange mechanisms together with introducing concepts of organizational learning. In OCS, agents acquire their own appropriate functions through interaction among their neighborhoods, and form an organizational structure without an explicit control mechanism. To investigate the effectiveness of our model, we have conducted intensive experiments on real scale printed circuit boards (PCBs) re-design problems in the computer aided design (CAD). The experimental results have suggested that (1) Total wiring length with OCS is shorter than that with human experts; (2) OCS finds feasible solutions even if the size of PCBs becomes large or the type of PCBs is different; and (3) Knowledge acquired with the small size of PCBs in OCS is also applicable to the large sized one, and this utilization of the knowledge contributes to reducing the iteration counts.
organizational learning, learning classifier system, cooperative agents, multiagent reinforcement learning, print circuit board design
PS file(+gzip)