Knowledge Acquisition from Examples Using Maximal Representation Learning

Abstract

In this paper we describe a new Knowledge Acquisition technique based on the learning from examples paradigm. This technique uses the principle of maximal representation of attributes which clusters the examples into sub-descriptions using the frequencies, and orders the attributes based on how well it represents the class. The sub-descriptions aid in simple definitions of importance and redundancy of attributes for a class which are then used to discard unimportant attributes and thereby simplify class description. The resultant description is the simplest characteristic description which describes the class completely with respect to the learning examples and the attributes specified. This helps not only in describing the class well but also in potentially discriminating the current class from all other classes envisaged. An inference algorithm based on the frequencies of the sub-descriptions and importance of attributes is used to classify new examples. This system has been tested on two real-life applications, the results of which are highly encouraging.