New Kurtosis Optimization Algorithms for Independent Component Analysis

This paper considers the independent component analysis (ICA) case in blind source separation (BSS), in which observations result from the linear and instantaneous mixture of sources. Inspired from the recently proposed reference-based contrast criteria, a similar contrast function is proposed, based on which novel optimization algorithms are proposed. They are very similar to the former classical fast fixed-point (FastICA) algorithms based on the kurtosis, but differ in the fact that they are more efficient than the corresponding latter ones respectively in terms of the computational speed, which is particularly striking when the number of samples is large. The validity and performance of the new algorithms are investigated through simulations, in which comparison and analysis are also performed.