Mapping the Local Dark Matter Distribution through Machine Learning

Abstract: The particle nature of dark matter remains one of the most profound unsolved problems in modern astrophysics and cosmology. While Lambda-CDM provides excellent agreement with observations on cosmological scales, significant tensions emerge at galactic and sub-galactic scales, where predictions from different dark matter models yield distinct observable signatures. Dwarf galaxies and stellar streams in the Milky Way environment retain subtle signatures of dark matter physics in their phase-space distribution, making them exceptional laboratories for constraining dark matter properties. However, inferring dark matter halo properties from stellar kinematics is challenging due to incomplete phase-space information, measurement uncertainties, and degeneracies in dynamical models. In this talk, I present how modern machine learning methods, specifically neural simulation-based inference, can provide a complementary approach to traditional techniques, enabling robust constraints on dark matter halo parameters while naturally incorporating these observational complexities. I demonstrate applications to both dwarf galaxies and stellar streams, highlighting how this framework yields precise characterization of dark matter density profiles, substructure, and implications for the particle nature of dark matter.