Emulation of cosmological mass maps with conditional generative adversarial networks

Mass maps created using weak gravitational lensing techniques play a crucial role in understanding the evolution of structures in the universe and our ability to constrain cosmological models. The mass maps are based on computationally expensive N-body simulations, which can create a computational bottleneck for data analysis. Simulation-based emulators of observables are starting to play an increasingly important role in cosmology, as the analytical predictions are expected to reach their precision limits for upcoming experiments. Modern deep generative models, such as Generative Adversarial Networks (GANs), have demonstrated their potential to significantly reduce the computational cost of generating such simulations and generate the observable mass maps directly. Until now, most GAN approaches produce simulations for a fixed value of the cosmological parameters, which limits their practical applicability. We instead propose a new conditional model that is able to generate simulations for arbitrary cosmological parameters spanned by the space of simulations. Our results show that unseen cosmologies can be generated with high statistical accuracy and visual quality. This contribution is a step towards emulating weak lensing observables at the map level, as opposed to the summary statistic level.