Generalised gravitational burst generation with Generative Adversarial Networks

TL;DR

This paper presents a novel use of conditional GANs to generate diverse gravitational wave burst signals, improving detection classifiers by training on a broader, interpolated signal space.

Contribution

The study introduces a conditional GAN framework for generating generalized gravitational wave burst signals, enabling interpolation and class mixing for enhanced detection.

Findings

GAN-generated signals replicate standard classes accurately

Interpolated signals improve classifier detection efficiency

Conditional GAN enables diverse signal generation

Abstract

We introduce the use of conditional generative adversarial networks forgeneralised gravitational wave burst generation in the time domain.Generativeadversarial networks are generative machine learning models that produce new databased on the features of the training data set. We condition the network on fiveclasses of time-series signals that are often used to characterise gravitational waveburst searches: sine-Gaussian, ringdown, white noise burst, Gaussian pulse and binaryblack hole merger. We show that the model can replicate the features of these standardsignal classes and, in addition, produce generalised burst signals through interpolationand class mixing. We also present an example application where a convolutional neuralnetwork classifier is trained on burst signals generated by our conditional generativeadversarial network. We show that a convolutional neural network classifier trainedonly on the standard five signal classes has a poorer detection efficiency than aconvolutional neural network classifier trained on a population of generalised burstsignals drawn from the combined signal class space.