Bottom-Up Norm Creation in Open Distributed Computing Grids by Means of eXtended Classifier Systems

Abstract:
Open, distributed multi-agent systems with heterogeneous agent societies need to be robust against malicious agents in the system. These malicious agents can show complex behaviours to manipulate or exploit the system. Trust mechanisms are a popular method to improve the performance and robustness of such systems. Yet, these mechanisms leave room for improvement. In this paper, we present a norm approach to enhance the robustness and, in consequence, the performance of the open multi-agent system. We introduce a concept, where norms are created bottom-up by the agents with the help of an eXtended Classifier System learning mechanism. We evaluate our approach within the simulation of a trust-based desktop computing grid and different agent stereotypes.