Speaker
Description
The fundamental nature of dark matter so far eludes direct detection experiments, but it has left its imprint in the population of low-mass sub-halos in the Milky Way (MW). Measurements of the MW phase space from observatories like Gaia will probe the lightest dark halos ever detectable (< 10^7 M_Sol) through their gravitational perturbations with thin tidal streams of stars. It is thus paramount that as much information as possible is extracted, while carefully accounting for the complex modelling uncertainties and instrumental systematics through detailed simulations. I will present a new approach to robust inference of fundamental physics from stellar stream data using graph neural networks to compress the full 6D phase space and normalising flows to infer posterior distributions. I will demonstrate that this deep learning method can improve constraints on halo properties by up to a factor of 100 compared to existing density power spectrum measurements. I will discuss plans to prepare this pipeline for upcoming Gaia data, in particular the modelling challenge.
