Parametric independent component analysis for stable distributions
Mohammad Reza Ameri, Mona Shokripour, Adel Mohammadpour, Vahid Nassiri · Artificial Intelligence Research · 2013
Independent Component Analysis (ICA) is a method for blind source separation of a multivariate dataset that transforms observations to new statistically independent linear forms. Infinite variance of non-Gaussian α-stable distributions makes algorithms like ICA non-appropriate for these distributions. In this note, we propose an algorithm which computes mixing matrix of ICA, in a parametric subclass of α -stable distributions.