GENERALIZATION AND EXCLUSIVE ALLOCATION OF CREDIT IN UNSUPERVISED CATEGORY LEARNING

A new way of measuring generalization in unsupervised learning is pres ented. The measure is based on an exclusive allocation, or credit assi gnment, criterion. In a classifier that satisfies the criterion, input patterns are parsed so that the credit for each input feature is assi gned exclusively to one of multiple, possibly overlapping, output cate gories. Such a classifier achieves context-sensitive, global represent ations of pattern data. Two additional constraints, sequence masking a nd uncertainty multiplexing, are described; these can be used to refin e the measure of generalization. The generalization performance of EXI N networks, winner-take-all competitive learning networks, linear deco rrelator networks, and Nigrin's SONNET-2 network are compared.