Speaker
Guillem Fernández-Rodríguez
(Universitat de València)
Description
This work explores feature selection for glitch classification in gravitational wave detectors, where transient noise events are grouped by their time-frequency morphology. Gravity Spy is a citizen-science project that provides a standard dataset built from four Q-transform views of glitches. Follow-up studies introduced attention-based multi-view models, transfer learning, and examined the effect of the Q-transform quality factor. These results point to spectrogram parameters as a key factor in model generalisation and bias. We run a random search over Q-transform and time window settings, train several computer vision models, and evaluate their behaviour using Virgo data.
Primary author
Guillem Fernández-Rodríguez
(Universitat de València)
Co-author
Isabel Cordero Carrión
(University of Valencia)
