Extract non-Gaussian Features in Gravitational Wave Observation Data Using Self-Supervised Learning

Abstract

We propose a self-supervised learning model to denoise gravitational wave (GW) signals in the time series strain data without relying on waveform information. Denoising GW data is a crucial intermediate process for machine-learning-based data analysis techniques, as it can simplify the model for downstream tasks such as detections and parameter estimations. We use the blind-spot neural network and train it with whitened strain data with GW signals injected as both input data and target. Under the assumption of a Gaussian noise model, our model successfully denoises 38% of GW signals from binary black hole mergers in H1 data and 49% of signals in L1 data detected in the O1, O2, and O3 observation runs with an overlap greater than 0.5. We also test the model's potential to extract glitch features, loud inspiral compact binary coalescence signals a few seconds before the merger, and unseen CCSN signals during training.