Orateur
Dr
Niall Jeffrey
(École normale supérieure)
Description
I will present single frequency morphological foreground removal and denoising using deep learning (in a Bayesian likelihood-free framework) using phase harmonic augmentation. With only a single training image of polarized dust foregrounds, we are able to generate new realisations using wavelet phase harmonic synthesis. In a likelihood-free inference framework, these new realisations can be used for Bayesian foreground map-cleaning (using high dimensional Moment Networks) and cosmological parameter inference.
Auteur principal
Dr
Niall Jeffrey
(École normale supérieure)
Co-auteurs
Bruno Regaldo-Saint Blancard
(Ecole Normale Supérieure)
Francois Boulanger
(Ecole Normale Supérieure)
Erwan Allys
(LPENS, Paris)
