Gravity seminar - Stephen Green Seminar

Abstract: In this talk, I will describe the use of normalizing flows for rapid likelihood-free inference of binary black hole system parameters from gravitational-wave data with deep neural networks. A normalizing flow is an invertible mapping on a sample space that can be used to induce a transformation from a simple probability distribution to a more complex one; if the simple distribution can be rapidly sampled and its density evaluated, then so can the complex distribution. By conditioning this flow on detector strain and training on simulated data, it can learn a mapping from a multivariate standard normal distribution into a gravitational-wave posterior. This then allows for rapid parameter estimation. Next, I describe a more powerful latent variable model by incorporating normalizing flows into a variational autoencoder. This model has performance comparable to Markov chain Monte Carlo, and in particular it can model multimodal posteriors. I will demonstrate that on an 8-dimensional parameter space (masses, aligned spins, time and phase of coalescence, luminosity distance, and inclination), all parameters and degeneracies are well-recovered. For these models, sampling is extremely fast, requiring less than two seconds to draw 10^4 samples.