ConvWave: Searching for Gravitational Waves with Fully Convolutional Neural Nets

Timothy Gebhard*, Niki Kilbertus*, Giambattista Parascandolo, Ian Harry, Bernhard Schölkopf

Accepted at the Deep Learning for Physical Sciences workshop at NeurIPS 2017, December 2017

PDF BibTeX

Cite this paper

@Article{         Gebhard_2017,
  title         = {ConvWave: Searching for Gravitational Waves with Fully
                  Convolutional Neural Nets},
  author        = {Gebhard, T. and Kilbertus, N. and Parascandolo, G. and
                  Harry, I. and Sch{\"o}lkopf, B.},
  year          = {2017},
  month         = {12},
  booktitle     = {Workshop on Deep Learning for Physical Sciences (DLPS) at
                  the 31st Conference on Neural Information Processing
                  Systems (NeurIPS)},
  url           = {https://dl4physicalsciences.github.io/files/nips_dlps_2017_13.pdf}
}

Code Poster

Abstract:

The first detection of gravitational waves (GWs) from a binary black hole merger in 2015 was a milestone in modern physics, and just recently awarded with the Nobel Prize. However, despite the unparalleled sensitivity of the LIGO detectors, there still exist challenges in the analysis of the recorded data. We apply CONVWAVE, a dilated, fully convolutional neural net directly on the time series strain data to identify simulated GW signals from black hole mergers in real, non-Gaussian background measurements from the LIGO detectors. CONVWAVE performs well on simulated signals with masses and distances chosen from ranges that contain the estimated parameters of all previously detected real events. It efficiently runs on strain data of arbitrary length from any number of detectors in real time. Through our proposed evaluation approach, it has the potential to develop into a complementary trigger generator in the existing LIGO search pipeline.