A distributed problem-solving approach to rule induction : learning in distributed artificial intelligence systems

Abstract: "One of the interesting characteristics of multi-agent problem solving in distributed artificial intelligence (DAI) systems is that the agents are able to learn from each other, thereby facilitating the problem-solving process and enhancing the quality of the solution generated. This paper aims at studying the multi-agent learning mechanism involved in a specific group learning situation: the induction of concepts from training examples. Based on the mechanism, a distributed problem-solving approach to inductive learning, referred to as DLS, is developed and analyzed. This approach not only provides a method for solving the inductive learning problem in a distributed fashion, it also helps shed light on the essential elements contributing to multi-agent learning in DAI systems.Am empirical study is used to evaluate the efficacy of DLS for rule induction as well as its performance patterns in relation to various group parameters. The ensuing analysis helps form a model for characterizing multi-agent learning."