Obtaining cosmological covariance matrices with machine learning

Paper highlights:

• Introduces a machine learning approach to denoise cosmological covariance matrices estimated from as few as 50 power spectra.
• Trains convolutional neural networks on inexpensive mock simulations to produce covariances nearly indistinguishable from those built using thousands of spectra.
• Shows that denoised covariances recover cosmological parameters with accuracy comparable to analyses based on much larger simulation sets.

You can read the paper here.