Chirag Modi, Flatiron
When
Abstract: Upcoming cosmological surveys such as DESI, LSST, Euclid will probe our Universe at the largest volumes with a variety of cosmological observables, including galaxy clustering and weak lensing. The statistical power of these datasets provides an exciting opportunity to constrain cosmological parameters with unprecedented precision and answer longstanding fundamental questions about the birth, and evolution of our Universe, and the nature of dark energy and dark matter. However, current approaches to cosmological analysis will be insufficient in extracting all the information from this high-quality data. In this talk, I will discuss the limitations of these traditional methods, and motivate how they can be overcome using computational forward models. I will then present two forward modeling approaches in detail. The first approach is “simulation-based inference” which allows us to use new, powerful summary statistics without having access to theoretical models and analytic likelihood for the data. The second approach is “field-level inference”, which promises to eventually be the optimal way of doing cosmological analysis. By simultaneously inferring the cosmological parameters and the initial conditions of the Universe, field-level inference also opens doors to completely new science cases, not possible before. I will discuss the advances made in cosmology, statistics and machine learning over the last 5 years that have made these approaches possible. Finally, I will present challenges in applying forward modeling approaches at the scale necessary for the next generation of cosmological surveys and outline strategies to overcome them.