Mapping interstellar dust with Gaussian processes

Interstellar dust corrupts nearly every stellar observation and accounting for it is crucial to measuring physical properties of stars. We model the distribution as a spatially varying latent field with Gaussian process (GP) develop likelihood inference method that scales millions astronomical observations. Modeling interstellar complicated by two factors. The first integrated data come from vantage point on Earth, each an integral unobserved function along our line sight, resulting in complex more difficult problem than classical GP inference. second complication scale; catalogs have To address these challenges, we ziggy, scalable approach observations based stochastic variational study ziggy synthetic Ananke dataset, high-fidelity mechanistic Milky Way reliably infers spatial map well-calibrated posterior uncertainties.