2020-11-08, 16:25–16:25, Posters
One of the most widely used methods for deriving physical properties of galaxies involves modeling their SEDs. While several modern Bayesian codes are routinely used to fit observed photometries with underlying models of star formation, evolution, and dust attenuation, results based on hydrodynamical simulations have quantitatively shown that uncertainties in the derived properties--for example, galaxy mass, star formation history, gas phase dust mass--at a factor of a few level are inescapable. Furthermore, Bayesian fitting is an intrinsically compute-intensive task, often requiring access to expensive hardware for long periods of time. To overcome these shortcomings, we develop 'mirkwood': an explainable, user-friendly tool comprising of an ensemble of supervised machine-learning models capable of mapping galaxy fluxes to their properties. For the first time, we delineate the contribution of each photometric band to galaxy parameters of interest via the use of Shapley values. We also derive both reducible and irreducible predictions errors -- the former refer to uncertainties arising both from missing observations in informative bands and intrinsic noise in observations, while the latter refer to those arising from finite training data and incorrect modeling assumptions. We demonstrate mirkwood's superior performance by training it on a combined data set of z=0 galaxies from SIMBA, EAGLE and ILLUSTRIS-TNG, and comparing the derived results with those obtained from the Prospector. We envisage mirkwood to be an evolving, open-source framework that will phase out traditional SED fitting.