Stability of Object Classes and Selection of the Latent Features

Abstract

The problem of selection new latent features space in pattern recognition with teacher is considered. To form the space is used by the partition the set of the initial features to the previously unknown number of disjoint groups. The extremum of the criterion of grouping is calculated by the value of stability of each object in class along the set of features included in the group. Displaying such a set to the numerical axis represents the value of the latent feature in the description of admissible objects.