Millions of Galaxies, Sparse Information: Reliable SED Inference for HETDEX with Neural Density Estimators

Interpreting galaxy images and spectroscopy is central to studies of galaxy formation and evolution. Widely-used forward-modeling approaches infer galaxy parameters via MCMC but take hours per galaxy due to model generation cost; this is prohibitive for modern astronomical surveys with millions of galaxies. Here, we present a simulation-based inference framework for fast, amortized inference of galaxy physical parameters in the HETDEX survey. A unique challenge is that astrophysical inference is not performed upon raw measurements but instead processed observables which have strongly heteroscedastic uncertainties, epistemic uncertainties, and patchy imaging coverage. Furthermore, galaxy inference requires complex physical forward models whose correctness cannot be empirically verified, making simulation-based inference a natural framework for posterior predictive checks at scale. We address these challenges with a neural posterior estimator trained on ∼10^7 simulated galaxies generated with a 17-parameter model, paired with lin–log asinh magnitudes and a tailored uncertainty model; we contrast the predictive performance and speed with classic MCMC inference.