Simulation-based Inference for Gravitational Wave Astronomy

November 17, 2021
Recorded 17 November 2021. Kyle Cranmer of New York University presents "Simulation-based Inference for Gravitational Wave Astronomy" at IPAM's Workshop III: Source inference and parameter estimation in Gravitational Wave Astronomy. Abstract: I will briefly review the taxonomy of simulation-based inference techniques developed in the last few years. The taxonomy will include branching on frequentist vs. Bayesian, learned or engineered summary statistics, densities vs. density ratios, and amortized vs. sequential approaches. I will then try to place the concerns of gravitational wave astronomy in this framing. I will also discuss briefly the role of inductive bias (e.g. symmetries) for approaches based on neural networks. Learn more online at: