Pattern classification in dynamic environments: tagged feature-class representation and the classifiers

The author discusses: a tagged feature and class representation of the pattern recognition problem in a dynamic environment; univariate cooperative classifiers that are based on statistical feature evaluation and impose no constraint on the variations of the sets of classes and features; and inductive learning procedures that are used to create a class-feature space adaptive to the variations of the dynamic environment. The univariate classifier and the cooperative classifier apply a classify-by-rejection approach to a candidate class set. The classification is based on the individual evaluation of the features presented in the sample patterns and the classes. The tagged feature-class space permits convenient building of a hierarchical structure of the classifications A content-addressable data retrieved characteristic is possessed by both types of classifier. Experimental results on the classifiers are presented.<
>