Dalek: A Deep Learning Emulator for TARDIS

Published in The Astrophysical Journal Letters, 2021

Recommended citation: Kerzendorf, Wolfgang; et al. (2021). https://ui.adsabs.harvard.edu/abs/2021ApJ...910L..23K/abstract

This paper presents DALEK, a neural network based emulator for the radiative transfer code TARDIS. DALEK is trained on hundreds of thousands of full TARDIS simulations and can predict the TARDIS output synthetic spectrum to extremely high accuracy. DALEK is 6 orders of magnitude faster than TARDIS, so millions of evaluations can be made in only a few hours. This unlocks the ability to perform Bayesian inference and fit models to observed spectra, quantifying degeneracies and uncertainties for the high dimensional TARDIS model parameters.