Faster and better sparse blind source separation through mini-batch optimization

Sparse Blind Source Separation (sBSS) plays a key role in scientific domains as different as biomedical imaging, remote sensing or astrophysics. Such fields however require the development of increasingly faster and scalable BSS methods without sacrificing the separation performances. To that end, we introduce in this work a new distributed sparse BSS algorithm based on a mini-batch extension of the Generalized Morphological Component Analysis algorithm (GMCA). Precisely, it combines a robust projected alternated least-squares method with mini-batch optimization. The originality further lies in the use of a manifold-based aggregation of the asynchronously estimated mixing matrices. Numerical experiments are carried out on realistic spectroscopic spectra, and highlight the ability of the proposed distributed GMCA (dGMCA) to provide very good separation results even when very small mini-batches are used. Quite unexpectedly, the algorithm can further outperform the (non-distributed) state-of-the-art methods for highly sparse sources.