Gravity Seminar: Machine learning for gravitational wave parameter estimation: towards waveform systematics, Sama Al-Shammari, Cardiff University, U.K.

When performing parameter estimation for gravitational-wave (GW) data, Bayesian techniques provide a powerful framework to describe said parameters. These techniques are consistent with both our prior knowledge as well as empirical data. However, one of the key quantities needed for statistical inference, the likelihood of the observed data, is usually intractable. Additionally, waveform systematics stand as yet another hurdle when performing data analysis of gravitational wave signals. Simulation-based inference (SBI) is a class of methods that utilise observable data in a simulator to obtain latent values and parameters required for parameter estimation. In this work, we use SBI machine-learning methods to simulate data from parameters obtained by sampling from the prior in conjunction with two different, marginalised, waveform models. We then make use of deep neural networks to learn statistical inference from the previously simulated data. Finally, we apply the learned neural network to injected gravitational-wave data and derive the posterior distribution. We compare our results to current Bayesian inference methods used by the LVK collaboration and show that our method provides similar results in a fraction of the time.

We will view the zoom in Aula 03b in Edifici Antoni Maria Alcover i Sureda