Conditional noise deep learning for parameter estimation of gravitational wave events

We construct a Bayesian inference deep learning machine for parameter estimation of gravitational wave events of binaries of black hole coalescence. The structure of our deep Bayesian machine adopts the conditional variational autoencoder scheme by conditioning on both the gravitational wave strains and the variations of the amplitude spectral density (ASD) of the detector noise. We show that our deep Bayesian machine is capable of yielding posteriors compatible with the ones from the nested sampling method and better than the one without conditioning on the ASD. Our result implies that the process of parameter estimation can be accelerated significantly by deep learning even with large ASD drifting/variation. We also apply our deep Bayesian machine to the LIGO/Virgo O3 events, the result is compatible with the one by the traditional Bayesian inference method for the gravitational wave events with signal-to-noise ratios higher than typical threshold value. We use one GPU device, NVIDIA RTX3090, to train the deep learning machines, which takes about 12 hours. After training, it takes less than one second to generate the posterior of a gravitational wave event and is far faster than the conventional nested sampling method.

Automated summary

We propose a Bayesian inference deep learning machine for parameter estimation of gravitational wave events. The machine is capable of accelerating the estimation process and generating posterior probabilities compatible with traditional methods.