Fast Bayesian gravitational wave parameter estimation using convolutional neural networks

TL;DR

This paper introduces a convolutional neural network method for rapid Bayesian gravitational wave parameter estimation, significantly reducing computation time while providing compatible posterior distributions and useful initial parameter indications.

Contribution

The paper presents a novel CNN-based approach that accelerates gravitational wave parameter estimation, enabling real-time analysis and initial parameter insights, which is a significant advancement over traditional Bayesian methods.

Findings

CNN produces posterior distributions compatible with traditional methods.

Inference speed is reduced by orders of magnitude.

Provides valuable initial parameter estimates for multi-messenger astronomy.

Abstract

The determination of the physical parameters of gravitational wave events is a fundamental pillar in the analysis of the signals observed by the current ground-based interferometers. Typically, this is done using Bayesian inference approaches which, albeit very accurate, are very computationally expensive. We propose a convolutional neural network approach to perform this task. The convolutional neural network is trained using simulated signals injected in a Gaussian noise. We verify the correctness of the neural network's output distribution and compare its estimates with the posterior distributions obtained from traditional Bayesian inference methods for some real events. The results demonstrate the convolutional neural network's ability to produce posterior distributions that are compatible with the traditional methods. Moreover, it achieves a remarkable inference speed, lowering by orders of magnitude the times of Bayesian inference methods, enabling real-time analysis of gravitational wave signals. Despite the observed reduced accuracy in the parameters, the neural network provides valuable initial indications of key parameters of the event such as the sky location, facilitating a multi-messenger approach.