Efficient and differentiable simulations for cosmological forward modelling

Abstract: The cosmic large-scale structure will take a key role in testing fundamental physics in the next decade. To extract the maximum cosmological information from the current and next generation surveys, modelling has to stay on par with the increasing quality of the data. The next generation of cosmological forward modelling has to face both high-dimensional cosmological parameter space as well as robustness and efficiency as we push deeper into the non-linear regime. This requires progress in both numerical and data-driven methods for LSS cosmology. I will discuss our recent efforts to devise highly accurate and very fast field-level models in a new framework that can be integrated with both gradient-based inference or simulation-based inference.