CG: Wiener filter for CMB maps using Neural Networks – Costanza

In this work, we have studied a Convolutional Neural Network (CNN) called WienerNet to apply the Wiener Filter to noisy CMB (Cosmic Microwave Background) maps. We present how these neural networks work and how good the results are compared to the traditional method of Wiener Filter with the conjugate gradient. Also, we show that the predictions of the neural network are faster and more efficient than the conjugate gradient method, which is a bottleneck in CMB analyses. For these purposes, we have applied this neural network to CMB maps with different numbers of pixels, from 28×28 to 512×512, to study how the computation time scales with the size of the maps.