cDVGAN: One Flexible Model for Multi-class Gravitational Wave Signal and Glitch Generation

TL;DR

The paper addresses generating time-domain gravitational wave signals and detector glitches for data analysis. It introduces cDVGAN, a conditional GAN with an auxiliary discriminator that processes first-order derivatives to improve realism, plus an extended variant cDVGAN2 for second-order derivatives. The model conditions on three classes (blip and tomte glitches, and BBH signals), and supports interpolation to create hybrid samples. Experiments show improved CNN detection performance and a high fidelity to reference waveforms, with significant data augmentation benefits and fast generation on GPU.

Abstract

Simulating realistic time-domain observations of gravitational waves (GWs) and GW detector glitches can help in advancing GW data analysis. Simulated data can be used in downstream tasks by augmenting datasets for signal searches, balancing data sets for machine learning, and validating detection schemes. In this work, we present Conditional Derivative GAN (cDVGAN), a novel conditional model in the Generative Adversarial Network framework for simulating multiple classes of time-domain observations that represent gravitational waves (GWs) and detector glitches. cDVGAN can also generate generalized hybrid samples that span the variation between classes through interpolation in the conditioned class vector. cDVGAN introduces an additional player into the typical 2-player adversarial game of GANs, where an auxiliary discriminator analyzes the first-order derivative time-series. Our results show that this provides synthetic data that better captures the features of the original data. cDVGAN conditions on three classes, two denoised from LIGO blip and tomte glitch events from its 3rd observing run (O3), and the third representing binary black hole (BBH) mergers. Our proposed cDVGAN outperforms 4 different baseline GAN models in replicating the features of the three classes. Specifically, our experiments show that training convolutional neural networks (CNNs) with our cDVGAN-generated data improves the detection of samples embedded in detector noise beyond the synthetic data from other state-of-the-art GAN models. Our best synthetic dataset yields as much as a 4.2% increase in area-under-the-curve (AUC) performance compared to synthetic datasets from baseline GANs. Moreover, training the CNN with hybrid samples from our cDVGAN outperforms CNNs trained only on the standard classes, when identifying real samples embedded in LIGO detector background (4% AUC improvement for cDVGAN).

Figures (17)

• Figure 1: Diagrams of a typical cGAN architecture (left), comprising one discriminator, and cDVGAN (right), comprising two discriminators. Class vectors $c$ (real) and $\hat{c}$ (fake), are fed to all model components in both cases. An intermediate derivative calculation is observed in the cDVGAN plot, where the derivative of the synthetic sample is calculated. cDVGAN2 includes yet another discriminator applied to second-order derivatives. In cDVGAN and cDVGAN2, the total generator loss is calculated as a linear combination of the discriminator losses applied to synthetic samples.
• Figure 2: A comparison of the discriminators from the original cGAN paper (used for McGANn and McDVGANn) and the projection discriminator (used for cWGAN, cDVGAN, cDVGAN2).
• Figure 3: Examples of blip (top), tomte (middle) and BBH signals (bottom) used to train GAN models
• Figure 4: Visualizations of the preprocessing steps applied to a blip glitch event.
• Figure 5: A plot of the first 3 principal components of the original samples. The separability of the three classes in this compressed representation indicates the diversity of the data.