Title: Simulation-Based Inference for Higher-order Statistics in Weak Lensing Cosmology

Speaker: Kunhao Zhong

Abstract: The next-generation galaxy survey, LSST, Euclid, and Roman, will provide us with an enormous amount of data to learn the growth history of the universe. These surveys will probe over billions of galaxies and make unprecedented precise measurements of weak gravitational lensing and clustering. These advancements promise significant insights into the inflationary epoch, the nature of dark energy, and neutrino masses. However, traditional methods such as 2-point statistics only capture Gaussian information. In this talk, I will introduce simulation-based inference (or likelihood-free inference) as an efficient way to extract cosmological information from non-gaussian summary statistics. Specifically, I will show the recent DES-Y3 effort in using moments and wavelet transform and their validation processes. I will present the recent work in making field-level compression using convolutional neural networks. I will also introduce an analytical generative model as an extension to the widely used log-normal model. Lastly, I will discuss the latest developments, challenges, and the future direction of simulation-based inference in cosmology.

Bio: Kunhao Zhong is a second-year PhD student at the University of Pennsylvania. He is advised by Prof. Bhuvnesh Jain to work on weak lensing higher order statistics, simulation-based inference, and machine learning. Kunhao has a broad interest in astrophysics and particle physics. Previously he was a Master's student at Stony Brook University working on dark energy phenomenology with Prof. Vivian Miranda.

Time: 10:30-11:30, 27/Dec, Friday
Venue: Room 506 (Large seminar room), Department of Astronomy