Sanger's Like Systems for Generalized Principal and Minor Component Analysis

Abstract

In this paper generalizations of Sanger's learning rule for nondefinite matrices are explored. It is shown that the left and right principal components of any matrix can be computed so that these components upper triangulize the original matrix. We also modified the original Sanger's system to obtain new dynamical systems with a larger domain of attraction. Stability analysis for several Sanger's type systems for the standard and generalized principal, and minor component analyzers applied to nonsymmetric matrices is developed