GRASP colloquium by dr. John Veitch, University of Glasgow

Powering up gravitational wave inference with machine learning

Advanced LIGO and Virgo are now detecting several compact binary mergers per week, producing a population of signals that shed light on astrophysical formation, lensing, cosmology and nuclear physics. All of these investigations rely on bayesian analysis, stochastic sampling, and ultimately the quality of the model of the data. Current techniques have been able to keep up with the data, but challenges lie ahead, especially with next generation and space-based observatories.

Thankfully, rapid developments in machine learning are providing enhancements in multiple directions: lowering the latency of analyses, the complexity of the models that can be analysed, and the duration of the signals. I will give an overview of how normalising flows have proved useful in all these areas, with specific applications to next generation signals and glitch mitigation.