Learning relationships among classes specified by examples

Abstract

Learning by examples is a commonly used method of knowledge acquisition in expert systems. The learning examples are past cases whose classification is known. When the number of classes is high, learning relationships amongst classes aids in sequential classification and therefore results in better explanations. A methodology to learn relationships amongst a given set of classes aids in sequential classification and therefore results in better explanations. A methodology to learn relationships amongst a given set of classes using a distance measure is described in this paper. This distance which is shown to be a metric is evaluated on the descriptions of the classes. The description of a class is learnt from its examples. The relationship learnt using this distance metric is shown to converge in the limit. The methodology described learns meaningful relationships for two applications commonly used by machine learning research groups.