Learning organizational roles for negotiated search in a multiagent system

Abstract

This paper presents studies in learning a form of organizational knowledge called organizational roles in a multi-agent agent system. It attempts to demonstrate the viability and utility of self-organization in an agent-based system involving complex interactions within the agent set. We present a multi-agent parametric design system called L-TEAM where a set of heterogeneous agents learn their organizational roles in negotiated search for mutually acceptable designs. We tested the system on a steam condenser design domain and empirically demonstrated its usefulness. L-TEAM produced better results than its non-learning predecessor, TEAM, which required elaborate knowledge engineering to hand-code organizational roles for its agent set. In addition, we discuss experiments with L-TEAM that highlight the importance of certain learning issues in multi-agent systems.