TY - JOUR

T1 - Training strategies for deep learning gravitational-wave searches

AU - Schäfer, Marlin B.

AU - Zelenka, Ondřej

AU - Nitz, Alexander H.

AU - Ohme, F.

AU - Brügmann, Bernd

N1 - Funding Information: We acknowledge the Max Planck Gesellschaft and the Atlas cluster computing team at Albert-Einstein Institut (AEI) Hannover for support, as well as the ARA cluster team at the URZ Jena. F. O. was supported by the Max Planck Society’s Independent Research Group Programme. O. Z. thanks the Carl Zeiss Foundation for the financial support within the scope of the program line “Breakthroughs”.

PY - 2022/2/8

Y1 - 2022/2/8

N2 - Compact binary systems emit gravitational radiation which is potentially detectable by current Earth bound detectors. Extracting these signals from the instruments' background noise is a complex problem and the computational cost of most current searches depends on the complexity of the source model. Deep learning may be capable of finding signals where current algorithms hit computational limits. Here we restrict our analysis to signals from non-spinning binary black holes and systematically test different strategies by which training data is presented to the networks. To assess the impact of the training strategies, we re-analyze the first published networks and directly compare them to an equivalent matched-filter search. We find that the deep learning algorithms can generalize low signal-to-noise ratio (SNR) signals to high SNR ones but not vice versa. As such, it is not beneficial to provide high SNR signals during training, and fastest convergence is achieved when low SNR samples are provided early on. During testing we found that the networks are sometimes unable to recover any signals when a false alarm probability

AB - Compact binary systems emit gravitational radiation which is potentially detectable by current Earth bound detectors. Extracting these signals from the instruments' background noise is a complex problem and the computational cost of most current searches depends on the complexity of the source model. Deep learning may be capable of finding signals where current algorithms hit computational limits. Here we restrict our analysis to signals from non-spinning binary black holes and systematically test different strategies by which training data is presented to the networks. To assess the impact of the training strategies, we re-analyze the first published networks and directly compare them to an equivalent matched-filter search. We find that the deep learning algorithms can generalize low signal-to-noise ratio (SNR) signals to high SNR ones but not vice versa. As such, it is not beneficial to provide high SNR signals during training, and fastest convergence is achieved when low SNR samples are provided early on. During testing we found that the networks are sometimes unable to recover any signals when a false alarm probability

KW - astro-ph.IM

KW - cs.LG

KW - gr-qc

UR - http://www.scopus.com/inward/record.url?scp=85125220745&partnerID=8YFLogxK

U2 - 10.1103/PhysRevD.105.043002

DO - 10.1103/PhysRevD.105.043002

M3 - Article

VL - 105

JO - Physical Review D

JF - Physical Review D

SN - 2470-0010

IS - 4

M1 - 043002

ER -