A Simple and Flexible Nonlinearty Approach to Independent Component Analysis

A simple and flexible nonlinearity approach to independent component analysis is presented, which is able to blindly separate mixed super-Gaussian, Gaussian and sub-Gaussian sources. The parameter of the nonlinearity is estimated by representing it as a function of the kurtosis of sources. Further, the stability conditions for the proposed algorithm are analyzed to give a robust algorithm for independent component analysis. We show that this algorithm can interestingly be used to find hidden physiological processes inherent in gene expression experiments