Classification by balanced, linearly separable representation

Classifiers for binary and for real-valued data, consisting of a single internal layer of spherical threshold cells, are completely defined by two fundamental requirements: linear separability of the internal representations, which defines the cells´ activation threshold, and input-space covering, which defines the minimal number of cells required. Class assignments are learnt by applying Rosenblatt´s learning rule to the internal representations which are balanced, having equally probable bit values. The separation capacity may be increased by increasing the number-of cells, at a possible cost in generalization. Our analysis extends to the classification of binarized symbolic (or enumerated) data and explains an empirical observation made in the literature on the separability of such data