SMuK 2023 – wissenschaftliches Programm
Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
GR: Fachverband Gravitation und Relativitätstheorie
GR 13: Relativity and Data Analysis
GR 13.2: Vortrag
Donnerstag, 23. März 2023, 16:20–16:40, ZEU/0260
Noise transients in machine-learning based gravitational-wave searches — •Ondřej Zelenka1,2, Bernd Brügmann1,2, and Frank Ohme3,4 — 1Friedrich-Schiller-Universität Jena, D-07743 Jena, Germany — 2Michael Stifel Center Jena, D-07743 Jena, Germany — 3Max-Planck-Institut für Gravitationsphysik, Albert-Einstein-Institut, D-30167 Hannover, Germany — 4Leibniz Universität Hannover, D-30167 Hannover, Germany
In the recent past, machine-learning based approaches have been proposed as a solution to some problems in gravitational-wave data analysis. One of these are noise transients, which significantly complicate detection of gravitational waves. Contemporary matched-filtering based searches as well as unmodeled searches employ systems which flag likely noise transients and reject potential false alarms. It is possible that machine-learning based algorithms can learn to distinguish noise transients from signals with astrophysical sources.
In this contribution, we present a machine-learning based gravitational-wave detection algorithm focused on binary black holes, which has been submitted to the MLGWSC-1 mock data challenge. Furthermore, we describe an issue which arose when the model encountered non-Gaussian background noise, and present its solution. In doing so, we demonstrate that a machine-learning based algorithm with a suitable training method is capable of distinguishing false alarms due to transients from binary black hole injections.