A Framework for Improving Efficiency and Accuracy

Abstract

This chapter reviews a framework for improving efficiency and accuracy. One of the main goals of machine learning is to improve performance, and the two most studied performance measures are efficiency and predictive accuracy. The knowledge acquired by a learning system must be efficient to use and accurate. Thus, research under such a theme aims essentially at producing a framework where both efficiency and accuracy may achieve a simultaneous incremental nature and where advances in one of the these is not occurring at the cost of the other. The framework presented in the chapter is an extension of work in explanation-based and inductive learning. Research on inductive learning has primarily focused on improving predictive accuracy. The task of learning from examples is a form of inductive learning that builds concept descriptions based upon instances labeled with their class information. Concept hierarchies are an important component of many inductive learning methods. The hierarchy is used to classify instances by beginning with the concept in question. Whenever a concept is matched, all the bindings produced by the match are stored with the concept. The matcher will do less work if it matches the concept multiple times.