Combining Competition and Cooperation in Supervised Inductive Learning

Abstract

Tools for automatic generation, verification, and maintenance of knowledge bases become more and more important with the amount of widely available information growing. For supervised concept learning in attribute–based spaces, many approaches have been proposed including the symbolic AQ and ID based algorithms. These algorithms exhibit competitive characteristics since either partial covers or the attributes compete for consideration at any given moment. In this paper, we describe a new full memory approach which implements a very unique search mechanism combining the competition with cooperation. This approach uses the VL1 language in a framework utilizing operators of inductive learning methodology and an inference engine modeled upon genetic algorithms. We also present some experiments indicating the applicability of this approach to both quantitative and qualitative learning.