Foreground challenge to CMB polarization: present methodologies and new concepts

In this thesis, I focus on the issue of contamination to the polarization of the Cosmic Microwave Background (CMB) anisotropies from diffuse Galactic foregrounds, which is known to be one of the greatest challenges to the detection of the curl (B) modes of the signal, which might be sourced by cosmological gravitational waves. I take parallel approaches along these lines. I apply the most recent techniques capable of parametrizing, fitting, and removing the main known Galactic foregrounds in a multi-frequency CMB dataset to one of the forthcoming powerful CMB polarization experiments, the Large Scale Polarization Explorer (LSPE). I presented the result of the complete simulation done for the parametric component separation pipeline of this experiment. On the other hand, I explored the latest Machine Learning and Artificial Intelligence algorithms and their application in CMB data analysis, specifically component separation and foreground cleaning. I start the investigation of the relevance of Neural Networks (NNs) in the understanding of the physical properties of foregrounds, as it is necessary before the foreground removal layer, by implementing a novel algorithm, which I test on simulated data from future B-mode probes. The results of the implemented NN’s prediction in discerning the correct foreground model address the high accuracy and suitability of this model as a preceding stage for the component separation procedure. Finally, I also investigate how different NNs, as a generative model, could be used for reconstructing CMB anisotropies where the removal is impossible, and data have to be abandoned in the analysis. Lots remain to be done along each of these three investigations, which have been published in scientific journals, and constitute the basis of my future research.