Learning and representing concepts with graded structure

Abstract

The author presents a novel method for representing and learning concepts with graded structure. The method uses a hybrid concept representation that combines symbolic and numeric representations. In learning a concept, the method builds a general concept description for representing common cases of the concept. Such a description is in the form of decision rules, interpreted by a weighted distance measure, and numerical thresholds. The method has been implemented in the system FCLS (flexible concept learning system) and tested on a variety of problems.<>