Deep learning for multimessenger core-collapse supernova detection

TL;DR

This paper introduces a machine learning method using a neural network to detect gravitational waves from core-collapse supernovae, demonstrating robustness with real noise data and potential for future detections.

Contribution

The study develops a novel Mini-Inception Resnet neural network for multimessenger supernova detection, trained on simulated signals and tested on real LIGO-Virgo noise data.

Findings

Detection efficiency exceeds 70% for signals with SNR > 15

Can detect signals from 1 kpc distance in O2 noise conditions

Potential to detect supernova signals up to 14 kpc

Abstract

The detection of gravitational waves from core-collapse supernova (CCSN) explosions is a challenging task, yet to be achieved, in which it is key the connection between multiple messengers, including neutrinos and electromagnetic signals. In this work, we present a method for detecting these kind of signals based on machine learning techniques. We tested its robustness by injecting signals in the real noise data taken by the Advanced LIGO-Virgo network during the second observation run, O2. We trained a newly developed Mini-Inception Resnet neural network using time-frequency images corresponding to injections of simulated phenomenological signals, which mimic the waveforms obtained in 3D numerical simulations of CCSNe. With this algorithm we were able to identify signals from both our phenomenological template bank and from actual numerical 3D simulations of CCSNe. We computed the detection efficiency versus the source distance, obtaining that, for signal to noise ratio higher than 15, the detection efficiency is 70 % at a false alarm rate lower than 5%. We notice also that, in the case of O2 run, it would have been possible to detect signals emitted at 1 kpc of distance, whilst lowering down the efficiency to 60%, the event distance reaches values up to 14 kpc.