Mitigation of Instrumental Glitches in Gravitational Wave Data Analysis

The development of a deep learning-based methodology for the accurate and fast classification and reconstruction of noise transients, called glitches, in gravitational-wave detectors is underway. Glitches, which originate from environmental disturbances or instrumental artifacts, appear in different unmodelled morphologies and represent one of the major challenges in gravitational-wave data analysis: beyond reducing detection confidence, they can significantly compromise the extraction of astrophysical information from observed signals. To date, approximately 25% of detected events require some form of glitch mitigation due to an overlap between the gravitational-wave signal and transient noise contamination. Improving the identification and subtraction of glitches is therefore essential for maximising the scientific output of present and future observing runs of the LIGO–Virgo–KAGRA collaboration and next generation detectors. The goal of this work is first to validate the methodology based on the recently developed tool called DeepExtractor on real interferometer data, ensuring physically consistent reconstructions and compatibility with existing gravitational-wave data-analysis pipelines. Results of the validation study mainly highlight great reconstruction capabilities of DeepExtractor, especially for certain glitch classes, and limitations of the glitchfind method employed in the analysis. In addition, the work includes the development of a dedicated Deep Extractor model specifically tailored to the Virgo detector. Such a model will account for the unique instrumental configuration, environmental conditions, and detector-specific noise characteristics of Virgo.