Search for gravitational wave signals in LIGO-Virgo-KAGRA data using machine learning methods

The aim of this work is to explore the possibility of using neural network classifiers to identify gravitational-wave signals in the LIGO-Virgo-KAGRA data. This task is complicated by the non-Gaussian, non-stationary noise in the detectors, in particular in periods when only one detector takes data. One detrimental source of noise is the so-called glitches, non-Gaussian transient noise that can mimic astrophysical signals. In general, temporal coincidence between two or more detectors is used to mitigate contamination, but when a single detector is in operation, coincidence is impossible. In this context, machine learning strategies could help complement more traditional approaches. One problem often faced in machine learning applications is the need for a large amount of training data. It is therefore useful to have a way to simulate glitches and gravitational-wave signals. While techniques for simulating gravitational-wave signals are well established, the simulation of glitches is less common because of their variability and unknown origin. In this work, a conjecture about the possible nature of some glitch classes is used to simulate them.