A3D3 Seminar: Jess McIver

Monday 24 Feb 2025, 12:00 → 13:00 US/Pacific

Title: Deep learning algorithms in GW astrophysics: explaining their success

 

Abstract:  The new era of gravitational wave (GW) astrophysics is underway, with roughly 90 GW signals confirmed and hundreds of new candidates from the current LIGO-Virgo-KAGRA observing run (O4). Gravitational wave candidates that are potentially electromagnetically bright (e.g. neutron star mergers) are high-value targets for telescope follow-up, however, a high rate of transient GW detector noise events (“glitches”) can cause false alarm signals that waste valuable telescope time in following up noise events. To meet this challenge, the GWSkyNet and GSpyNetTree algorithms leverage convolutional neural networks to quickly and accurately vet GW signal candidates, however, as with other deep learning methods, it is often difficult for human users to interpret their decisions. I’ll report on the development of a new framework for probing explainability to interrogate the good performance of these algorithms by the GWSkyNet team, in collaboration with computer scientists, neuroscientists, cosmologists, and data scientists (the ML-ESTEEM network). I'll close with prospects for the future and our lessons learned for developing and testing explainable deep learning algorithms that human researchers can more easily trust and interpret.

 

Jess McIver is an Associate Professor and Tier 2 Canada Research Chair in Gravitational Wave Astrophysics at the University of British Columbia's Department of Physics and Astronomy. She is a leading researcher in gravitational wave astrophysics, having been a member of the LIGO Scientific Collaboration since 2007. McIver contributed to the detection of the first binary neutron star merger and was part of the team awarded the Science 2017 Breakthrough of the Year. Her work focuses on detector noise characterization, calibration, and multi-messenger astronomy, advancing our understanding of merging black holes and neutron stars. McIver holds degrees from Syracuse University and the University of Massachusetts Amherst, and previously held a postdoctoral fellowship at Caltech.

 

The A3D3 Seminar is a monthly lecture series that hosts scholars working across applied areas of artificial intelligence, such as hardware algorithm co-development, high energy physics, multi-messenger astrophysics,  and neuroscience. Our presenters come from all four domain fields and include occasional external speakers beyond the A3D3 science areas, governmental agencies and industry. The seminar will be recorded and published in YouTube. To receive future event updates, subscribe here.