Speaker
Description
Physical constraints on accretion in extreme gravitational environments can be provided through the spectral fitting of Black hole X-ray binaries. Traditional methods of spectral fitting are limited by intractable computation times. We apply machine learning methods to expedite the spectral fitting of thousands of spectra collected by NICER. Specifically, we use a probabilistic approach, consisting of a variational auto-encoder with a normalizing flow, trained to adopt a physical latent space, building upon previous deterministic networks. This neural network predicts spectral-model parameters as well as their full probability distributions. Our method significantly improves in spectral reconstructions over the deterministic model while remaining three orders of magnitude faster than traditional methods, achieving the best fit parameters.