Date of Award

1993

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Computer Science

First Advisor

Sukhamay Kundu

Abstract

We consider the problem of learning complex recursive rules which involve new concepts other than those given in the input relations and which must be discovered by the learning algorithm in the course of finding rules. The existing learning methods (FOIL, FORGE, and etc.) create rules based on the given concepts or relations but they cannot create new concepts. However, in many cases one must use new intermediate concepts in order to form the recursive rules. We give a new technique for constructing such intermediate concepts and learning rules based on those concepts. We illustrate the new technique with several examples, none of which can be handled by the existing methods. We have implemented the new technique in Common Lisp and tested many different examples.

Pages

69

Share

COinS