Boruah: Novel ML-based statistical methods for cosmological data analysis with LSST / Porras-Valverde: Why do SAMs and Sims Disagree on the Stellar Mass - Halo Mass Scatter?

Oct 14, 2022 - 10:40 am to 11:30 am

SLAC, Kavli 3rd Floor Conf. Room

Supranta Sarma Boruah (University of Arizona) / Antonio Porras-Valverde (Vanderbilt) In Person and zoom

Zoom info: zoom

Multiprobe cosmological analyses require thousands of simulated likelihood analyses (each of which cost thousands of CPU hours) to study the impact of systematic effects, optimizing survey strategies, and assessing tensions between different probes. This will form a major computational bottleneck for Stage-IV surveys such as Vera Rubin Observatory's LSST. In this talk I will talk about a recent work (, where we proposed an iterative emulator that leads to fast inference in high dimensions by accelerating the model vector prediction. I will also talk about new results where we use this emulator to perform a comprehensive study of the impact of different systematic effects on 3x2 point analysis with LSST. Such study can be used to prioritize the understanding of different systematic effects to maximize the scientific return from LSST.
Time permitting, I will also talk about some recent progress on Bayesian field-level pipelines we are developing for LSST weak lensing data analysis.