Organizational Learning and Network Adaptation in Multi-Agent Systems

In both real and artificial societies, successful organizations are highly dependent upon a structure that fosters effective and efficient behavior at both the individual and the organizational levels. In multi-agent systems, groups of agents must coordinate effectively in order to solve problems, allocate tasks across a distributed organization, collectively distribute knowledge and information, and achieve collective goals. The organizational structure of a multi-agent system dictates the interactions among the agents, and can play a significant role in the overall performance of a society of agents.

Given the importance of organizational network structures for multi-agent systems, distributed network adaptation is a promising approach for organizational learning. After reviewing related work in multi-agent learning, the structure and dynamics of networks, and organizational learning in multi-agent systems, I present the concept of agent-organized networks as an approach for organizational learning by distributed network adaptation. Supported by theoretical evidence of the complexity of organizational network design, a general learning-based agent-organized network framework is proposed and applied to two general environments: multi-agent team formation and a production and exchange economy. In addition, the general framework is used to develop distributed network adaptation strategies for specific applications in supply network formation and wireless sensor networks. Experimental results for both the general and specific multi-agent environments support the hypothesis that distributed management and adaptation of organizational network structure leads to improved collective performance in multi-agent systems. Analyses of the structural characteristics of the networks as they evolve are used to aid in understanding the behavior of agent-organized networks and further support the utility of distributed network adaptation for organizational learning in multi-agent systems.

Committee Members: Dr. Marie desJardins ,(Chairperson) Dr. Tim Finin, Dr. David Jensen, Dr. Tim Oates and Dr. Yun Peng. Academic Advisor: Dr. Marie desJardins .