Concept Learning by Experiment

Abstract

Mich of the emphasis in current research on con-cept learning and rule Induction is based on two assuaptione. F i rs t l y, a l l that needs to be known to learn a concept can be obtained di rect ly from exaaplea given to I t, without reference to previously learnt knowledge. Secondly, a suff icient nuaber of exaaplee to learn the concept is available, and a l l exaaplee are presented slaultaneously. Relatively l i t t l e attention has been given to developing systems which are able to improve their performance over time by using knowledge that has been learnt before. Two exceptions are (Winston70a] and [Cohen78a] • The usual task for a learning program is: Given a set of positive Instances and a set of negative Instances, produce a concept description which dis-tinguishes between these two sets. The program pas-sively accepts its input and otherwise does not interact with the environment. Furthermore, It Is expected that the concept w i l l be learnt in one ses-sion. A program has been developed which learns con-cepts by searching a knowledge base which is augaented each time a new concept is learnt. A concept descrip-t ion aay be treated as a program which may be execu-ted. The output w i l l be the description.of an object which is an instance of the concept. A t r i a l concept may be tested by executing the description as an experiment to see if the desired result is produced. 2. Represent ing Concepts Suppose a trainer gives the program a description of an object which characterises the concept to be learnt. For example, an instance of "on-top-of " is,