NBIA Gravity Seminar: Stephen Green

Speaker: Stephen Green

Title: Simulation-Based Inference for Gravitational Waves

Abstract: During the first half of the fourth LIGO-Virgo-KAGRA observing run, we observed gravitational waves from merging binaries roughly every three days. While this routine detection promises exciting results, it is becoming a significant challenge to analyze all events using our most sophisticated theoretical models. In this talk, I will describe how to overcome these challenges using deep learning techniques for rapid, amortized Bayesian inference. This approach uses simulated data to train neural networks (such as normalizing flows) to represent the Bayesian posterior. Once trained, sampling becomes extremely fast. I will also describe how to establish full confidence in results using importance sampling, as well as initial results on population inference and future prospects to treat realistic noise.