Accelerating the Next Generation of Strong Gravitational Lensing Science using Machine Learning

In the near future, a new generation of sky surveys conducted by the Euclid Space Telescope and the Rubin Observatory's Legacy Survey of Space and Time (LSST) is expected to discover over 200,000 new strong gravitational lenses. This surge in known samples demands an accurate and rapid analysis of massive amounts of data within a clear statistical framework to facilitate scientific breakthroughs enabled by strong lensing. In this work, we delve into the critical role of simulation-based inference (SBI) in strong gravitational lensing analysis for these surveys. Our proposed approach involves machine learning models to obtain posteriors of macro-parameters of individual strong lenses, which can be integrated into a hierarchical framework for inferring population-level statistics of strong lenses. We demonstrate the potential for modeling the selection function of strong lenses using neural networks and building a generative model of real observational noise that can be readily integrated into SBI frameworks to perform accurate analysis under realistic noise conditions.