Diffuse pattern learning with Fuzzy ARTMAP and PASS

Abstract

Fuzzy ARTMAP is compared to a classifier system (CS) called PASS (predictive adaptive sequential system). Previously reported results in a benchmark classification task suggest that Fuzzy ARTMAP systems perform better and are more parsimonious than systems based on the CS architecture. The tasks considered here differ from ordinary classificatory tasks in the amount of output uncertainty associated with input categories. To be successful, learning systems must identify not only correct input categories, but also the most likely outputs for those categories. Performance under various types of diffuse patterns is investigated using a simulated scenario.