Diffusion-based galaxy emulator

We have also developed techniques to use diffusion-based learning models to accurately predict the distribution of galaxies on small scales, based on the full cosmological hydrodynamical simulations carried out by the training set generation group. This technique is producing remarkably accurate galaxy distributions, even on small scales. Related approaches include painting galaxies onto dark matter simulations using transformer-based models (Pandey et al., 2025) and inpainting galaxy counts across multiple cosmologies (Bourdin et al., 2024).