Can machine learning finally match simulations for cosmic lensing maps?

Weak lensing maps reveal how matter is distributed across the universe, but generating realistic ones requires massive simulations. This work trains a generative AI model (flow matching) that learns to create maps matching both the statistics and full distribution of N-body simulations, while conditioning on two key cosmological parameters. The result is a fast, accurate surrogate that could speed up constraining dark matter and dark energy properties.