Machine Learning for Software Reuse

Abstract

Recent work on learning apprentice systems suggests new approaches for using interactive programming environments to promote software reuse. Methodologies for software specification and validation yield natural domains of application for explanation-based learning techniques. This paper develops a relation between data abstractions in software and explanationbased generalization problems and shows how explanation-based learning can be used to generalize program abstractions to promote their reuse. This method is applied in the design of a system called LASR (Learning Apprentice for Software Reuse) which will acquire programming knowledge by capturing and generalizing interconnections between abstract data type theories. The technical role of theories in defining learned concepts in this application suggests their more general use in representing problems in explanation-based learning.