Filling in Cosmic Microwave Background map missing regions via Generative Adversarial Networks

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