Simulating Realistic Gravitational-wave Detector Glitch Populations using Generative Models across Multiple Glitch Classes

In the emerging field of gravitational-wave (GW) astronomy, the data collected by ground-based GW detectors such as LIGO is key to understanding the universe. In addition to detector noise and potential astrophysical signals, detector data consists of various different types of artifacts that hinder our ability to detect GW signals. These artifacts, known as glitches, are non-stationary, transient bursts of noise consisting of environmental and instrumental sources. These glitches impact the ability to both observe and characterize incoming GW signals by mimicking or overlapping with real GW signals. To address these issues, we present the first large-scale training data set extractor of gravitational-wave glitches. Glitches are extracted from noisy detector data using a Bayesian multi-component modeling method. Then they are tested for quality using a non-Gaussianity statistical test, to create a large-scale glitch waveforms dataset. We train a glitch-generator generative adversarial network (Gengli) to generate glitches for  multiple glitch classes and develop metrics to assess the quality of our generations.