Rapid Quality Assessment of Gravitational-Wave Glitch Reconstructions Using Machine Learning

Gravitational-wave detector data consists of non-stationary detector noise and transient bursts of noise known as glitches. These glitches impact the ability to both observe and characterize incoming gravitational-wave signals. Thus, it is imperative that we study these glitch populations to improve our sensitivity to real signals and to provide feedback to instrumentalists. In this work, we perform a large-scale reconstruction analysis on glitch time-domain waveforms from glitchy data from LIGO's third observation run. We introduce a tool to rapidly assess the quality of the glitch time-domain reconstructions by utilizing a machine-learning model built on time-domain Gaussianity tests of glitch-subtracted residual data. Using this framework, we demonstrate how large-scale time-domain datasets of real, noise-free detector glitches can be rapidly produced and assessed, paving the way for improved glitch population studies and future developments in classification and simulation tools.