In this work we present a Neural Network (NN) algorithm for the identification of the appropriate parametrization of diffuse polarized Galactic emissions in the context of Cosmic Microwave Background (CMB) $B$-mode multi-frequency observations. In particular, we have focused our analysis on low frequency foregrounds relevant for polarization observation: namely Galactic Synchrotron and Anomalous Microwave Emission (AME). We have implemented and tested our approach on a set of simulated maps corresponding to the frequency coverage and sensitivity represented by future satellite and low frequency ground based probes. The NN efficiency in recognizing the right parametrization of foreground emission in different sky regions reaches an accuracy of about $90\%$. We have compared this performance with the $\chi^{2}$ information following parametric foreground estimation using multi-frequency fitting, and quantify the gain provided by a NN approach. Our results show the relevance of model recognition in CMB $B$-mode observations, and highlight the exploitation of dedicated procedures to this purpose.

The Large Scale Polarization Explorer (LSPE) is a cosmology program for the measurement of large scale curl-like features (B-modes) in the polarization of the Cosmic Microwave Background. Its goal is to constrain the background of inflationary gravity waves traveling through the universe at the time of matter-radiation decoupling. The two instruments of LSPE are meant to synergically operate by covering a large portion of the northern microwave sky. LSPE/STRIP is a coherent array of receivers planned to be operated from the Teide Observatory in Tenerife, for the control and characterization of the low-frequency polarized signals of galactic origin; LSPE/SWIPE is a balloon-borne bolometric polarimeter based on 330 large throughput multi-moded detectors, designed to measure the CMB polarization at 150 GHz and to monitor the polarized emission by galactic dust above 200 GHz. The combined performance and the expected level of systematics mitigation will allow LSPE to constrain primordial B-modes down to a tensor/scalar ratio of $10^{-2}$. We here report the status of the STRIP pre-commissioning phase and the progress in the characterization of the key subsystems of the SWIPE payload (namely the cryogenic polarization modulation unit and the multi-moded TES pixels) prior to receiver integration.

In this work, we propose a new method to inpaint the CMB signal in regions masked out following a point source extraction process. We adopt a modified Generative Adversarial Network (GAN) and compare different combinations of internal (hyper-)parameters and training strategies. We study the performance using a suitable $\mathcal{C}_r$ variable in order to estimate the performance regarding the CMB power spectrum recovery. We consider a test set where one point source is masked out in each sky patch with a 1.83 $\times$ 1.83 squared degree extension, which, in our gridding, corresponds to 64 $\times$ 64 pixels. The GAN is optimized for estimating performance on Planck 2018 total intensity simulations. The training makes the GAN effective in reconstructing a masking corresponding to about 1500 pixels with $1\%$ error down to angular scales corresponding to about 5 arcminutes.