Learning from knowledge systems

This thesis describes a case-based reasoning (CBR) system, ISCN Student, which acquires its knowledge from a previously developed rule-based knowledge system, ISCN Expert. That is, ISCN Student is a second generation knowledge system that learns from a first generation one. ISCN Student has been shown to perform with the same competence as ISCN Expert once trained. The architecture for this solution is based upon the creation of a general purpose object-oriented CBR framework, written in Smalltalk, that has been specialized to develop ISCN Student, but which is applicable to other CBR problem domains. ISCN is a notation used by geneticists to describe chromosome defects; the functional purpose of both ISCN Expert and Student is to interpret expressions in ISCN. To this end, several supporting paths of novel research have been pursued in addition to the above. First, a grammar and associated parser for ISCN were created. Second was the development of a visual manipulation system for displaying chromosome defects and introducing new abnormalities as cases to ISCN Student.