Cautious cooperative learning with distributed case-based reasoning

In this paper, we propose a cautious cooperative learning approach using distributed case-based reasoning. Our approach consists of two learning mechanisms: individual and cooperative learning. Normally, an agent conducts individual learning to learn from its past behavior. When the agent encounters a problem that it has failed to solve (satisfactorily), it triggers cooperative learning, asking for help from its neighboring agents. To avoid corrupting its own casebase and incurring costs on itself and other agents, our agent employs an axiomatic, cautious strategy that includes the notion of a chronological casebase, a profile-based neighbor selection, and a case review and adaptation before adopting an incoming case. Here we report on the approach and some results in a real-time negotiation domain.