Single frequency CMB B-mode inference with realistic foregrounds from a single training image

With a single training image and using wavelet phase harmonic augmentation, we present polarized Cosmic Microwave Background (CMB) foreground marginalization in high-dimensional likelihood-free (Bayesian) framework. We demonstrate robust removal only frequency of simulated data for BICEP-like sky patch. Using Moment Networks estimate the pixel-level posterior probability underlying {E,B} signal validate statistical model with quantile-type test estimated marginal moments. The use hierarchy U-Net convolutional neural networks. This work validates such an approach most difficult limiting case: pixel-level, noise-free, highly non-Gaussian dust foregrounds at frequency. For real CMB experiment, small number representative patches would provide required full cosmological inference. These results enable likelihood-free, simulation-based parameter inference primordial B-mode detection observed polarization data.