Robust Sparse Blind Source Separation

Abstract: Blind Source Separation (BSS) has become a key tool for the analysis of multichannel data encountered in various domains, including astrophysics, terrestrial remote-sensing and bioengineering. Most of the BSS methods assume that the observations correspond to a linear combination of underlying sources, to be retrieved. However, in many real-world applications, large deviations from this simplistic linear model occur, e.g. unexpected physical events, non-linearities of the observed physical process, or instrumental artifacts. These deviations are detrimental for standard BSS methods, thus creating a need for new robust BSS strategies. Building upon sparse modeling of the components, we have proposed robust BSS methods able to perform in a wide variety of settings, including the full-rank regime. Various types of deviations are considered and illustrated, including in astrophysics with the Planck data.

Bio: Cecile Chenot received an Engineering degree from Supelec, France and a MSc in Electrical Engineering from ETHZ, Switzerland in 2014. Since 2014, she has been a PhD student at CEA (Alternative Energies and Atomic Energy Commission), France, under the supervision of J.Bobin. Her research interests include multichannel data analysis, sparse modeling, and optimization.