New Zealand Statistical Association 2024 Conference

Tarin Eccleston

The University of Auckland

Variational autoencoders for stellar core-collapse gravitational waves

This is joint work with Matt Edwards

We present work towards a rapid stellar core-collapse waveform emulator using a variational autoencoder (VAE) – a follow-up from our previous work on using deep convolutional generative adversarial networks (DCGANs). The main advantage of using VAEs over DCGANs is that they provide a smoother and well-structured latent space representation by assuming a specific prior distribution. VAEs also allow us to perform variational inference and are less prone to mode collapse and unstable training. The pre-trained VAE will be used in match-filtering analysis to detect gravitational wave signals from stellar core-collapse events.