Teaching Dark Matter Simulations to Speak the Galactic Language

by Dr Shivam Pandey

Monday 10 Feb 2025, 15:00 → 16:00 Europe/Madrid

C7b-0-058 - Seminar (IFAE Main Building C7b)

Developing fast and efficient methods for simulating our observable Universe is a key bottleneck in maximizing information extraction from cosmological datasets. I will discuss a novel approach to obtaining galaxies and their properties from fast dark matter simulations by treating this as a language translation problem. Using a transformer-based multi-modal architecture, I will first demonstrate how we can obtain high-resolution halo catalogs from very low-resolution (and approximate) dark matter simulations. Then, I will show how the same architecture can be augmented to directly generate hydrodynamical simulation-based galaxy catalogs and their observable photometry, conditioned on dark matter-only simulations. Finally, I will discuss the potential of this approach in the context of ongoing large-scale surveys such as DESI and Euclid.