Extracting Gravitational Wave Features Using Self-Supervised Learning Denoiser

Yu-Chiung Lin^1*, Albert Kong¹

¹Institute of Astronomy, National Tsing Hua University, Hsinchu, Taiwan

* Presenter:Yu-Chiung Lin, email:yuchiung.lin@mx.nthu.edu.tw

We propose a self-supervised learning model to denoise gravitational wave signals in the time series strain data of GW detectors, without the knowledge of GW templates. Denoising of GW data is a crucial intermediate process for machine-learning-based data analysis techniques. Denoised data can simplify the model for downstream tasks such as detections and parameter estimations. We use the blind-spot neural network and train with whitened strain data having GW signal injected as both input data and target. With the Gaussian noise model assumption, our model successfully denoises GW signals from binary black hole mergers detected in the O1, O2, and O3a observation run. Our model also has the potential to denoise GW signals of compact binary inspiral before the merger when the signal is loud.

Keywords: gravitational waves, deep learning, self-supervised learning