A Distributed Problem-Solving Approach to Inductive Learning

Abstract

In this paper we propose a distributed approach to the inductive learning problem and present an implementation of the Distributed Learning System (DLS). Our method involves breaking up the data set into different sub-samples, using an inductive learning program (in our case PLS1) for each sample, and finally synthesizing the results given by each program into a final concept by using a genetic algorithm. We show that such an approach gives significantly better results than using the whole data set on an inductive learning program. We then show how DLS can be generalized to incorporate any learning algorithm and present some of the implications of this approach to Distributed AI (DAI) systems in general and learning methodologies in particular. Complexity analysis further shows that the time complexity of DLS can be made linear with respect to the size of the problem (data set) irrespective of the time complexity of the learning algorithm it uses.