METHODS AND APPARATUS FOR PERFORMING DIMENSIONALITY REDUCTION IN A SUPERVISED APPLICATION DOMAIN

A technique of the system for generating a reduced representation of input data, wherein the input data has a first set of feature variables and a class variable associated therewith, comprises the following steps. A second set of feature variables is determined from the first set of feature variables, wherein the second set of feature variables corresponds to mutually orthogonal vectors. Then, one or more of the feature variables associated with the second set of feature variables are selected based on a level of discrimination with respect to the class variable. The input data is then represented using the one or more selected feature variables.