Using Background Knowledge in Concept Formation

The utility of background knowledge about components for the task of incremental concept formation in structured domains is described, and LABYRINTH, which is a system that carries out structured concept formation, and its use of backgroundknowledge to improve its predictive accuracy is described.

Abstract

This chapter describes the utility of background knowledge about components for the task of incremental concept formation. It describes LABYRINTH, which is a system that carries out structured concept formation, and its use of background knowledge to improve its predictive accuracy. LABYRINTH is a model of incremental concept formation in structured domains, in which objects have components. LABYRINTH represents concepts by storing a set of attributes, their values, and associated probabilities. LABYRINTH can also store structured concepts, in which attributes have values that are primitive concepts. The system represents a CUP in terms of two other learned concepts, one that generalizes several cup bodies, and another generalizing the observed cup handles. Thus, LABYRINTH defines structured concepts not as monolithic structures but in terms of other concepts in memory. LABYRINTH can be primed with classes that characterize arbitrary subsets of attributes, provided instances are decompose d in the same way. Thus, LABYRINTH's use of components in classification enables it to take advantage of a form of background knowledge that is common in many domains.