English
 
Help Privacy Policy Disclaimer
  Advanced SearchBrowse

Item

ITEM ACTIONSEXPORT
  Training Strategies for Deep Learning Gravitational-Wave Searches

Schäfer, M., Zelenka, O., Nitz, A. H., Ohme, F., & Brügmann, B. (2022). Training Strategies for Deep Learning Gravitational-Wave Searches. Physical Review D, 105: 043002. doi:10.1103/PhysRevD.105.043002.

Item is

Basic

show hide
Genre: Journal Article

Files

show Files
hide Files
:
2106.03741.pdf (Preprint), 3MB
Name:
2106.03741.pdf
Description:
File downloaded from arXiv at 2021-06-10 07:38
Visibility:
Public
MIME-Type / Checksum:
application/pdf / [MD5]
Technical Metadata:
Copyright Date:
-
Copyright Info:
-
:
PhysRevD.105.043002.pdf (Publisher version), 4MB
Name:
PhysRevD.105.043002.pdf
Description:
Open Access
Visibility:
Public
MIME-Type / Checksum:
application/pdf / [MD5]
Technical Metadata:
Copyright Date:
-
Copyright Info:
-

Locators

show

Creators

show
hide
 Creators:
Schäfer, Marlin1, Author              
Zelenka, Ondřej, Author
Nitz, Alexander H.2, Author              
Ohme, Frank1, Author              
Brügmann, Bernd, Author
Affiliations:
1Binary Merger Observations and Numerical Relativity, AEI-Hannover, MPI for Gravitational Physics, Max Planck Society, ou_2461691              
2Observational Relativity and Cosmology, AEI-Hannover, MPI for Gravitational Physics, Max Planck Society, ou_24011              

Content

show
hide
Free keywords: Astrophysics, Instrumentation and Methods for Astrophysics, astro-ph.IM,Computer Science, Learning, cs.LG,General Relativity and Quantum Cosmology, gr-qc
 Abstract: 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 $<10^{-3}$ is required. We resolve this restriction by applying a modification we call unbounded Softmax replacement (USR) after training. With this alteration we find that the machine learning search retains $\geq 97.5\%$ of the sensitivity of the matched-filter search down to a false-alarm rate of 1 per month.

Details

show
hide
Language(s):
 Dates: 2021-06-072022
 Publication Status: Published in print
 Pages: 17 pages, 11 figures, 3 tables, supplemental materials at https://github.com/gwastro/ml-training-strategies
 Publishing info: -
 Table of Contents: -
 Rev. Type: -
 Identifiers: arXiv: 2106.03741
DOI: 10.1103/PhysRevD.105.043002
 Degree: -

Event

show

Legal Case

show

Project information

show

Source 1

show
hide
Title: Physical Review D
  Other : Phys. Rev. D.
Source Genre: Journal
 Creator(s):
Affiliations:
Publ. Info: Lancaster, Pa. : American Physical Society
Pages: - Volume / Issue: 105 Sequence Number: 043002 Start / End Page: - Identifier: ISSN: 0556-2821
CoNE: https://pure.mpg.de/cone/journals/resource/111088197762258