Radio-astronomy is entering the golden age, with new very sensitive instruments such as LOFAR, LWA and SKA). This allows us to reinvestigate the radio emission from ground. At these frequencies, both the close and distant universe can be studied. Radio instruments require however to reconstruct first images from a subset of Fourier components. Very nice progress have been made in the recent years in this domain, thanks to recent breakthrough in applied mathematics such compressed sensing theories or Deep learning. These tools presents however serious limitations because they consider only mono-canal data set. The goal of this project is to build a multichannel-data reconstruction, using both the deep learning concept and advanced component separation techniques. Within this project, we seek one PhD student who will develop novel data-driven approaches for the robust reconstruction of radio maps at different frequencies, which will offer uncertainty quantification, as well as the extension of these methods to the sphere. Emphasis will be given to the analysis of large-scale datasets such those that will be acquired with Square Kilometre Array Telescope (SKA). The doctoral student will be located at the premises of FORTH with a strong collaboration with the CosmoStat Laboratory at CEA Saclay.