Abstract :
The microwave emission from our own Galaxy is observed as a foreground contamination to detect the Cosmic Microwave Background (CMB) B-mode polarization. In the past decades, techniques (commonly known as component separation algorithms) have been proposed in the literature have been shown to successfully separate and reconstruct the unpolarized CMB emission from the foreground one. This is also due to the availability of unprecedented templates from multi-frequency observations (e.g. WMAP, Planck). However, the situation in polarization is still challenged by the fact that the B-mode polarization is orders of magnitude weaker and that the Galactic
polarized emission has just started to be probed. In this talk, we will show how novel techniques involving supervised and unsupervised learning encoding clustering techniques can be adopted in order to improve the performances of parametric component separation. Moreover, we have identified the number of clouds along the line of sight as a good tracer to infer the properties of Galactic dust. This is particularly relevant for future CMB experiments (e.g. SO, LiteBIRD, CMB-S4 ), where high sensitivities are expected to be achieved.
