De-noising non-Gaussian fields in cosmology with normalizing flows by Adam Rouhiainen et al. on Monday 28 NovemberFields in cosmology, such as the matter distribution, are observed byexperiments up to experimental noise. The first step in cosmological dataanalysis is usually to de-noise the observed field using an analytic orsimulation driven prior. On large enough scales, such fields are Gaussian, andthe de-noising step is known as Wiener filtering. However, on smaller scalesprobed by upcoming experiments, a Gaussian prior is substantially sub-optimalbecause the true field distribution is very non-Gaussian. Using normalizingflows, it is possible to learn the non-Gaussian prior from simulations (or frommore high-resolution observations), and use this knowledge to de-noise the datamore effectively. We show that we can train a flow to represent the matterdistribution of the universe, and evaluate how much signal-to-noise can begained as a function of the experimental noise under idealized conditions. Wealso introduce a patching method to reconstruct fields on arbitrarily largeimages by dividing them up into small maps (where we reconstruct non-Gaussianfeatures), and patching the small posterior maps together on large scales(where the field is Gaussian).arXiv: http://arxiv.org/abs/http://arxiv.org/abs/2211.15161v1