Get our free extension to see links to code for papers anywhere online!

Chrome logo Add to Chrome

Firefox logo Add to Firefox


Deep Learning at Scale for Gravitational Wave Parameter Estimation of Binary Black Hole Mergers

Mar 05, 2019
Hongyu Shen, E. A. Huerta, Zhizhen Zhao


Share this with someone who'll enjoy it:


We present the first application of deep learning at scale to do gravitational wave parameter estimation of binary black hole mergers that describe a 4-D signal manifold, i.e., black holes whose spins are aligned or anti-aligned, and which evolve on quasi-circular orbits. We densely sample this 4-D signal manifold using over three hundred thousand simulated waveforms. In order to cover a broad range of astrophysically motivated scenarios, we synthetically enhance this waveform dataset to ensure that our deep learning algorithms can process waveforms located at any point in the data stream of gravitational wave detectors (time invariance) for a broad range of signal-to-noise ratios (scale invariance), which in turn means that our neural network models are trained with over $10^{7}$ waveform signals. We then apply these neural network models to estimate the astrophysical parameters of black hole mergers, and their corresponding black hole remnants, including the final spin and the gravitational wave quasi-normal frequencies. These neural network models represent the first time deep learning is used to provide point-parameter estimation calculations endowed with statistical errors. For each binary black hole merger that ground-based gravitational wave detectors have observed, our deep learning algorithms can reconstruct its parameters within 2 milliseconds using a single Tesla V100 GPU. We show that this new approach produces parameter estimation results that are consistent with Bayesian analyses that have been used to reconstruct the parameters of the catalog of binary black hole mergers observed by the advanced LIGO and Virgo detectors.

* 8 pages, 4 figures and 4 tables 


   Access Paper Source



Share this with someone who'll enjoy it: