Inpainting CMB maps using partial convolutional neural networks

Abstract We present a novel application of partial convolutional neural networks (PCNN) that can inpaint masked images the cosmic microwave background. The network reconstruct both maps and power spectra to few percent for circular irregularly shaped masks covering up 10% image area. By performing Kolmogorov-Smirnov test we show reconstructed are indistinguishable from input at 99.9% level. Moreover, PCNNs with regular irregular same accuracy. This should be particularly beneficial CMB come astrophysical sources such as Galactic foregrounds. proof concept shown in this paper shows an important tool data analysis pipelines cosmology.