Recovering the CMB Signal with Machine Learning

Abstract The cosmic microwave background (CMB), carrying the inhomogeneous information of very early universe, is great significance for understanding origin and evolution our universe. However, observational CMB maps contain serious foreground contaminations from several sources, such as Galactic synchrotron thermal dust emissions. Here, we build a deep convolutional neural network (CNN) to recover tiny signal various huge contaminations. Focusing on temperature fluctuations, find that CNN model can successfully with high accuracy, deviation recovered power spectrum C ? smaller than variance at > 10. We then apply this method current Planck observations, quite consistent disclosed by Collaboration, which indicates provide promising approach component separation observations. Furthermore, test simulated polarization based CMB-S4 experiment. result shows both EE BB spectra be accuracy. Therefore, will helpful detection primordial gravitational waves in future experiments. designed analyze two-dimensional images, thus not only able process full-sky maps, but also partial-sky maps. it used other similar experiments, radio surveys like Square Kilometer Array.