ABSTRACT
This chapter, adapted from Chris Owens' thesis,1 explores the mechan ics o f reminding-how does an intelligent system confronted with a new problem retrieve relevant solutions that it has stored in memory? The central question behind Owens' work is how a system can characterize a new problem in the right set o f abstract terms, thus allowing it to match against an isomorphic problem in a different domain. This is very hard to do using traditional, bottom-up feature detectors, because most problems can be characterized in terms o f many alternative sets of abstract features, most o f which will not lead to useful remindings. The core idea behind Owens' solution to this problem is what he calls active memory: using the feature sets used to describe cases already in memory to provide top-down guidance to the feature extraction process.
