Deep learning anomaly detector for numerical relativistic waveforms

Abstract

Gravitational Wave Astronomy is an emerging field revealing hidden information from Astrophysics and Cosmology. The increasing volume of observational data and Numerical Relativity simulations has promoted several analyzes and modeling of compact binaries’ gravitational waves. Especially, Machine Learning has become a great support to boost research. In this project, we developed a U-Net Deep Learning model that detects possible anomalous waveforms in a Numerical Relativity catalog. We use binary black hole simulations with varying masses and spins. We categorized seven different anomaly types during the coalescence stages with a dataset of dominant and higher modes waveforms.