Learning Experiments in a Heterogeneous Multi-agent System

     M V Nagendra Prasad, Victor Lesser and Susan Lander
             Department of Computer Science 
              University of Massachusetts 
                   Amherst, MA 01003


                       Abstract

Self-organization for efficient distributed search control has
received much attention previously but the work presented in this
paper represents one of the few attempts at demonstrating its
viability and utility in an agent-based system involving complex
interactions within the agent set.  We discuss experiments with a
heterogeneous multi-agent parametric design system called L-TEAM
where machine learning techniques endow the agents with
capabilities to learn their organizational roles in negotiated
search and to learn meta-level knowledge about the composite
search spaces. We tested the system on a steam condenser design
domain and empirically demonstrated its usefulness.