mirkwood: Fast and Accurate SED Modeling Using Machine Learning

Traditional spectral energy distribution (SED) fitting codes used to derive galaxy physical properties are often uncertain at the factor of a few level owing uncertainties in star formation histories and dust attenuation curves. Beyond this, Bayesian (which is typically SED software) an intrinsically compute-intensive task, requiring access expensive hardware for long periods time. To overcome these shortcomings, we have developed {\sc mirkwood}: user-friendly tool comprising ensemble supervised machine learning-based models capable non-linearly mapping fluxes their properties. By stacking multiple models, marginalize against any individual model's poor performance given region parameter space. We demonstrate \textsc{mirkwood}'s significantly improved over traditional techniques by training it on combined data set mock photometry z=0 galaxies from \textsc{Simba}, \textsc{EAGLE} \textsc{IllustrisTNG} cosmological simulations, comparing derived results with those obtained techniques. \textsc{mirkwood} also able account arising both intrinsic noise observations, finite incorrect modeling assumptions. increase added value observational community, use Shapley explanations (SHAP) fairly evaluate relative importance different bands understand why particular predictions were reached. envisage be evolving, open-source framework that will provide highly accurate observations as compared fitting.