Gravitational-wave background detection using machine learning

TL;DR

This paper introduces a hybrid deep learning and Bayesian inference method to detect and analyze the gravitational-wave background more efficiently than traditional techniques, demonstrating high-confidence detection and component separation in simulated data.

Contribution

The paper presents a novel multi-scale multi-headed autoencoder combined with Bayesian inference for rapid GWB detection and component disentanglement, advancing current analysis methods.

Findings

Detects GWB from binary black holes with high confidence (log Bayes factor 3) at $ imes 10^{-9}$ energy density.

Simultaneously measures faint cosmological GWB component at $ imes 10^{-10}$ energy density.

Effective in simulated LIGO-Virgo-KAGRA network data at design sensitivity.

Abstract

Extracting the faint gravitational-wave background (GWB) signal from dominant detector noise and disentangling its %diverse astrophysical and cosmological components remain significant challenges for traditional methods like cross-correlation analysis. We propose a novel hybrid approach that combines deep learning with Bayesian inference to identify and characterize the GWB more rapidly than current techniques. Our method utilizes a custom-designed multi-scale multi-headed autoencoder (MSMHAutoencoder) architecture to separate GWB signals from detector noise, and subsequently Marcov Chain Monte Carlo parameter estimation to disentangle the GWB components. Using simulated data representative of the LIGO-Virgo-KAGRA network at design sensitivity, we show that our MSMHAutoencoder can detect with high confidence (log noise Bayes factor of 3) a GWB from binary black hole mergers with fractional energy density $\Omega_{\text{BBH}} \approx 10^{-9}$ at 25 Hz. In the presence of such an astrophysical GWB, we can simultaneously measure a cosmological component as faint as $\Omega_{\text{Cosmo}} \approx 1.3 \times 10^{-10}$ using 47.4 days of training data.