Distributed Interactive Learning in Multi-Agent Systems

Both explanation-based and inductive learning techniques have proven successful in a variety of distributed domains. However, learning in multi-agent systems does not necessarily involve the participation of other agents directly in the inductive process itself. Instead, many systems frequently employ multiple instances of induction separately, or singleagent learning. In this paper we present a new framework, named the Multi-Agent Inductive Learning System (MAILS), that tightly integrates processes of induction between agents. The MAILS framework combines inverse entailment with an epistemic approach to reasoning about knowledge in a multiagent setting, facilitating a systematic approach to the sharing of knowledge and invention of predicates when required. The benefits of the new approach are demonstrated for inducing declarative program fragments in a multi-agent distributed programming system.