Title: Reduction of transient noise artifacts in gravitational-wave data using deep learning

Abstract: Excess transient noise artifacts, or glitches impact the data quality of ground-based gravitational-wave (GW) detectors and impair the detection of signals produced by astrophysical sources. Mitigation of glitches is crucial for improving GW signal detectability. However, glitches are the product of short-lived linear and non-linear couplings among the interrelated detector-control systems that include optic alignment systems and mitigation of seismic disturbances, generally making it difficult to model noise couplings. Hence, typically time periods containing glitches are vetoed to mitigate the effect of glitches in GW searches at the cost of reduction of the analyzable data. To increase the available data period and improve the detectability for both model and unmodeled GW signals, we present a new machine learning based method which uses on-site sensors/system-controls monitoring the instrumental and environmental conditions to model noise couplings and subtract glitches in GW detector data. We find that our method reduces 20-70% of the excess power in the data due to glitches. By injecting software simulated signals into the data and recovering them with one of the unmodeled GW detection pipelines, we address the robustness of the glitch-reduction technique to efficiently remove glitches with no unintended effects on GW signals.