AstroAI Lunch Talks - June 22, 2026 - Ben Dodge & Robert Zimmermann
22 Jun 2026 - Joshua Wing
The video can be found here: https://www.youtube.com/watch?v=40Qa7Ia-6vE
Speaker 1: Ben Dodge
Title 1: GraphGP: Scalable Gaussian Processes for Mapping the Milky Way
Abstract 1: Gaussian processes are a powerful tool for modeling continuous fields in astronomy, but their naive computational cost and memory requirements often limit their practical use. Vecchia’s approximation is a sparse precision matrix approximation for stationary, decaying kernels that conditions each point only on its nearest neighbors. I will present GraphGP, a new implementation of Vecchia’s approximation that scales to nearly a billion parameters on a single GPU with linear time and memory requirements. Our key contributions are (1) a bit-reversed k-d tree ordering that allows efficient neighbors searches while also maximizing batch parallelism and (2) a differentiable CUDA implementation which is more than an order of magnitude faster and more memory efficient compared to a pure JAX baseline. I will also discuss my ongoing work on two applications of GraphGP: all-sky modeling of diffuse emission with SPHEREx and sodium tomography with the upcoming Via survey, illustrating the utility of Gaussian processes in astronomy and the importance of scaling them to large domains.
Speaker 2: Robert Zimmermann
Title 2: MW-Atlas: Towards the Three-Dimensional Stellar Density of the Local Galactic Disk & Halo within 1.25 kpc
Abstract 2: Applying information field theory to Gaia DR3 data, we aim to reconstruct the Milky Way’s stellar density of the local Galactic disk and halo within 1.25 kpc of the Sun as a non-parametric, three-dimensional field using a Gaussian process prior. Our model explicitly accounts for measurement quality and uncertainties, enabling reliable reconstruction down to small scales. This allows us to move beyond the limitations of parametric profiles and capture the Milky Way’s complex spatial structures and correlations. I will present the methods employed in this ongoing project and the first results, demonstrating the feasibility of accurate, non-parametric mapping of the local Galactic disc and halo up to small scales.