Incorporating machine learning algorithms for compact binary coalescences detection: a low-latency and explainable model approach

Zhi-Wei Chen³, Yu Chiung Lin^1*, Albert Kong¹, Chun Hsiang Chan²

¹Institute of Astronomy, National Tsing Hua University, Hsinchu, Taiwan
²Department of Geography, National Taiwan Normal University, Taipei, Taiwan
³Taipei First Girls High School, Taipei, Taiwan

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

Low-latency identification of gravitational wave (GW) signals from compact binary coalescences (CBC) is desirable in the era of multi-messenger astronomy. In this work, we incorporated several machine learning algorithms to detect CBC signals based on the morphological features in the Q-transform spectrogram of whitened strain data. By breaking the spectrogram down into several statistical quantities, we can train the machine learning model to identify the time location of the GW signals, including both the inspiral and merger parts. To enhance the explainability of the machine learning model, we further conducted explainable intelligence to investigate the impact of each statistical quantity on each prediction result. Our method can not only detect CBC mergers but is also promising to perform early-warning detection of inspiral signals.

Keywords: gravitational wave, machine learning, compact binary coalescences, explainable artificial intelligence