Constrained PCA techniques for the identification of common factors in data

We present an analysis of a constrained principal components analysis netwo rk that identifies the common factors in data sets in a manner similar to p rincipal factor analysis. This network responds to the covariance of the in put data (not both variance and covariance as in PCA) and so is resistant t o noise and varying levels of power on the inputs. The network naturally le nds itself to the sparse coding of data, however, by enforcing this sparsen ess further we are able to decipher dual components in data. (C) 1998 Elsev ier Science B.V. All rights reserved.