Environment classification in multiagent systems inspired by the adaptive immune system

Abstract

The adaptive immune system in vertebrates is a complex, distributed, adaptive system capable of effecting collective mul-ticellular responses. Our study introduces many of the desirable properties of this biological system to decentralized multiagent systems. We adopt the crossregulation model of the adaptive immune system involving interactions between effector and regulatory cells. Effector cells can mount beneficial immune responses to microbial antigens as well as pathologic autoimmune responses to self-antigens. Deleterious autoimmunity is prevented by regulatory cells that suppress the effectors to tolerate the self-antigens. We redeploy the crossregulation model within a multiagent system by letting each agent run an ODE-based instance of the model. Results of extensive simulation-based experiments demonstrate that a distributed multiagent system can mount different responses to distinct objects in their environment. These responses are solely a result of the dynamics between virtual cells in each agent and interactions between neighboring agents. The collective dynamics gives rise to a meaningful "self"- "nonself" classification of the environment by individual agent, even if these categories were not prescribed a priori in the agents.