SMuK 2021 – wissenschaftliches Programm
Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
GR: Fachverband Gravitation und Relativitätstheorie
GR 4: Gravitational waves
GR 4.3: Vortrag
Dienstag, 31. August 2021, 17:30–17:45, H6
Training Strategies for Deep Learning Gravitational-Wave Searches — Marlin Benedikt Schäfer1,2, •Ondřej Zelenka3,4, Alexander Harvey Nitz1,2, Frank Ohme1,2, and Bernd Brügmann3,4 — 1Max-Planck-Institut für Gravitationsphysik, Albert-Einstein-Institut, D-30167 Hannover, Germany — 2Leibniz Universität Hannover, D-30167 Hannover, Germany — 3Friedrich-Schiller-Universität Jena, D-07743 Jena, Germany — 4Michael Stifel Center Jena, D-07743 Jena, Germany
Deep learning may be capable of finding gravitational wave signals where current algorithms hit computational limits. 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 their impact, 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. We found that the networks are sometimes unable to recover any signals when a false alarm probability <10−3 is required. We resolve this by applying a modification we call unbounded Softmax replacement (USR) after training. With this alteration we find that the machine learning search retains ≥ 97.5% of the sensitivity of the matched-filter search down to a false-alarm rate of 1 per month.