Accelerating Bayesian Inference with Machine Learning
Bayesian inference methods are in widespread use in astronomy to infer knowledge of a physical model, often from limited amounts of data. Accurate and reproducible results are essential when the analysis may decide the fate of the universe! However, bayesian methods such as nested sampling and MCMC are notoriously slow, especially when the model is complex, or the dataset large. In this talk I will describe Nessai, our solution to this problem, which uses normalising flows to accelerate the nested sampling method. We show the potential to achieve orders of magnitude speedup over while being a drop-in replacement for state of the art bayesian tools. I will show our method applied to the analysis of gravitational waves from colliding black holes.
