Speaker
Description
The extinction of starlight, even though a nuisance in stellar astrophysics, allows mapping out the distribution of dust in the Milky Way. Thanks to the number of sightlines available with reliable parallax information, Gaia data have revolutionized the field, allowing high-resolution 3D dust maps out to kiloparsec distances. Mathematically, such reconstructions are ill-posed, e. g. due to the observational uncertainties, and require regularization. Gaussian processes and Bayesian variational inference can provide this regularization at quasi-linear compute. Over the last couple of years, the resulting maps, specifically a 3D dust map out to 1.25 kpc, have provided the basis for completely new kinds of studies of the interstellar medium. I will review the underlying techniques in my talk, also commenting on their current limitations and future potential. Gaia DR4 will extend the accuracy and reach of such 3D maps and I will show preliminary results from our latest reconstructions.