Optimizing ICA Using Prior Information

In this work we introduce a novel algorithm for Independent Component Analysis (ICA) that takes available prior information on the sources into account. This prior information is included in the form of a “weak” constraint and is exploited simultaneously with independence in order to separate the sources. Optimization is performed by means of Simulated Annealing. We show how it outperforms classical ICA algorithms in the case of low SNR. Moreover, additional prior information on the sources enforces the ordering of the components according to their significance.