Deep learning for real-time gravitational wave detection

Lead Research Organisation: Cardiff University
Department Name: School of Physics and Astronomy

Abstract

Joint observations of the binary neutron star merger GW170817 by the LIGO-Virgo gravitational-wave detectors and electromagnetic telescopes produced a wealth of information not accessible to gravitational waves alone, such as evidence for the origin of heavy elements and a direct measurement of the cosmic expansion. Future joint observations of systems such as supernovae, long gamma-ray bursts, or as-yet unknown phenomena could produce equally important insights. Such observations rely on rapid (minute-latency) analysis of the gravitational-wave data to identify signals, in order to direct follow-up observations with telescopes before the phenomenon fades. Convolutional neural networks (CNNs) are a promising technology for the gravitational-wave analysis; preliminary studies indicate they are capable of robustly detecting well-understood signal morphologies with very low processing time. To date, however, CNNs have been demonstrated to detect gravitational-wave signals for which we have an accurate mathematical model that can be used to train the CNN, leaving us unable to use them for the detection of new or unexpected signal types, and none have been deployed for real-time analysis. The key objective of this project is to construct, characterise, and deploy a data analysis pipeline based on convolutional neural networks that can analyse data from the LIGO-Virgo gravitational-wave detector network and identify generic, unmodelled gravitational-wave transient signals with minute-scale latencies. This will be done by training a CNN to recognise multi-detector coherence due to a common signal, rather than matching single-detector data to mathematical signal models, allowing us to detect generic transient signals. We will focus on the 10Hz-2000Hz frequency band, where the LIGO-Virgo network has best sensitivity, and signal durations on the millisecond to second scale. This work will involve gravitational physics, astrophysics, the statistical theory of signal detection, advanced data mining techniques, and large-scale computing.

Publications

10 25 50

Studentship Projects

Project Reference Relationship Related To Start End Student Name
EP/S023992/1 01/04/2019 30/09/2027
2269647 Studentship EP/S023992/1 01/10/2019 30/09/2023 Michael Norman