New Riemannian metrics for improvement of convergence speed in ICA based algorithms

In this paper three different Riemannian metrics are defined in the matrix space, according to various translations defined in the parameter space. Such metrics allowed the authors to derive correspondingly novel learning rules for two ICA based algorithms in blind source separation (BSS). Experimental results in several case studies have shown that a significant improvement of convergence speed can be achieved by employing such a new approach in comparison to the one based on right and left translations, which has appeared in the literature so far.