Proposal of a Search for Rotation Based Independent Component Analysis (SRICA) Algorithm

Abstract

AbstractExisting Independent Component Analysis (ICA) algorithms are using varying independence measures derived through varying independence definitions and approximations. It will be interesting to study the effect of these variation on ICA solution and applications. This study require an ICA algorithm allowing use of varying independence measures as optimization criteria, assuring global solution and being truly blind. The article derives and verifies experimentally for the need, the Search for Rotation based ICA (SRICA) algorithm. It uses the fact that the independent components can be found by rotation of the whiten components. It uses Genetic Algorithm (GA), as a global search technique, to find the optimal angle of rotation. Also, the required study through SRICA finds minimization of sum of marginal entropies with kernel method for density estimation as the best independence measure, in terms of source matching, compare to the used four other independence measures. KeywordsIndependent Component Analysis (ICA)Blind Source Separation (BSS)Genetic Algorithm (GA)Search for Rotation based ICA (SRICA)Overlearning