Speaker
Description
Glitches are short-duration, transient noises that can affect data quality and mask astrophysical signals. Therefore, it is extremely important to characterize them to understand their origin and mitigate them. A key step in this process is to characterize the different glitch families, that are thought to be linked to their different production mechanisms.
Glitches can be classified according to their morphology when represented as spectrograms. Convolutional Neural Networks (CNNs) have been proven to be a great tool in classifying 2D data such as spectrograms. In this work, we use CNNs to carry on a systematic study of glitches in Advanced Virgo O3 data.
We trained our deep learning model on ∼170k glitches starting from Advanced LIGO O3a glitches, labeled by the citizen science project GravitySpy, and used this model to label Advanced Virgo O3a data. We then use this data to build a new, custom model to label the remaining Virgo glitches, producing a catalog of families of Virgo glitches in O3, a key step toward a systematic real-time characterization of glitches in Advanced Virgo data.
