Mocking the Universe & Machine Learning Approaches to Large-Scale Structure

Efficient generation of accurate mock observations of the sky, tailored specifically to near-future large-scale structure and cosmic microwave background surveys, is a key technical challenge in cosmology. I will first discuss the creation of these synthetic observations, or mocks, through the use of rapid so-called ‘approximate’ simulation techniques, with a focus on the Websky extragalactic CMB simulations. Next, I will overview a few recent applications of machine learning in cosmology, including the use of deep learning techniques to speed up cosmological simulations, and the possible failure of supervised methods when trained on mocks with the goal of extracting information from data.

Speaker: George Stein, UC Berkeley