Learning Gaia: A Generative Model Trained on the Gaia Data Release 3 Source Catalog

概要

I train a conditional flow matching model on the positions, photometry, and astrometry of $\sim$ 579 million sources from the Gaia Data Release 3 Source Catalog. Using a probabilistic masking scheme during training, I develop a model capable of generating mock Gaia catalog sources with and without partial measurements of real Gaia sources. The model can produce coarse-grained mock Gaia Source Catalogs that reproduce the Milky Way disk and Magellanic Clouds, and can impute photometric and kinemat…