Description |
In this talk, I will present the forward modeling approaches to cosmological inference for the next generation of surveys. In the first part of the talk, I will motivate these approaches with the focus on a specific method- reconstruction of the cosmological fields. I will present two examples where this enhances the information that can be extracted from the large-scale structure surveys, one for galaxy clustering surveys and the other for 21-cm intensity mapping. In the second part of the talk, I will focus on the challenges for forward modeling approaches and the necessity of differentiable simulations to overcome them. I will present FlowPM- a differentiable particle mesh code in TensorFlow and demonstrate how such simulations can be combined with machine learning tools to speed up the reconstruction of cosmological fields by upto an order of magnitude.
|