Spectral approach to classification based on generalized unconditional tests

摘要

Abstract: Spectral approach to distribution-free classificationis presented. The discriminant function is based ongeneralized unconditional tests. The main steps of thealgorithms are: (1) finding a set of deadlockgeneralized tests, (2) computing a local discriminantfunction for each such a test, and (3) performing theactual classification of the observed pattern into aclass. The spectral algorithms involve computation ofthe Walsh and Reed-Muller (conjunctive) spectra usingfast algorithms.!10