Generalizing geometric ICA to nonlinear settings

Abstract

In this paper, a geometry-based algorithm for nonlinear blind source separation is presented. The mixture space is decomposed in a set of concentric rings, in which ordinary linear ICA is performed in order to get a set of images of ring points under the original mixing mapping. Putting those together the mixing mapping can be reconstructed. Various applications to two- and three-dimensional artificial and natural data sets are presented. © Springer-Verlag Berlin Heidelberg 2003.