Abstract

The Gaia satellite has provided first-ever positions and motions for billions of stars over the last decade, thereby revolutionizing our understanding of the local universe. However, observing constraints mean that Gaia’s measurements get worse for faint stars and truncate past a certain brightness. These faint stars are particularly interesting for many open astrophysical questions, such as the nature of dark matter and the formation history of the Milky Way and Local Group. Using my BP3M tool (Bayesian Positions, Parallaxes, and Proper Motions), I combine Hubble images with Gaia data through joint Bayesian inference, simultaneously solving for each image’s alignment to the Gaia reference frame and the improved motions of every star in the image. The faintest stars see the largest gains: their motions are up to 50x more precise than with Gaia alone, and we can measure first-ever motions for many stars too faint for Gaia to detect at all. In this talk, I will explain BP3M’s statistical techniques and the challenges of multi-telescope fusion. Each telescope has its own biases, distortions, and time-evolving optical system, and misspecifying any one of them leads to errors that compound across the combined measurement. I will present work testing modern machine learning techniques—such as multi-layer perceptrons and tabular foundation models—to characterize and remove these systematic errors directly from the data. I will also highlight science applications with dwarf galaxies and stellar streams to understand Local Group formation and the nature of dark matter. Finally, I will discuss extending BP3M-like methods to James Webb, ground-based surveys, and the next generation of telescopes (e.g. Euclid, Roman, Rubin/LSST).