Orateur
Didier BARRET
(IRAP)
Description
In this paper, I will introduce simulation-based inference for X-ray spectral fitting, emphasizing on its application to the challenging newAthena mock X-IFU high-resolution X-ray spectra. Training a neural density estimator on dimension reduced spectra, computed either through a compact and light encoder or an embedding network enables to quickly derive posterior approximates. I will then describe how the approximate posterior estimates can be corrected by the known likelihood, via importance-resampling, to generate an asymptotically exact estimate of the posteriors, identical to the ones produced by BXA, yet within a run time at least an order of magnitude shorter.
Auteur
Didier BARRET
(IRAP)
Co-auteur
Simon Dupourqué
(Institut de Recherche en Astrophysique et Planétologie)
