Deep Active Learning applied to gravitational waves

Abstract

The detection of gravitational waves (GW) has opened a new window to progress in our understanding of astrophysical events and objects. The instruments used for direct detection (interferometers) require high sensitivity due to the tiny signals these GWs generate. In addition, proper characterisation of the detector is crucial for identifying noise sources and enhancing the performance. This thesis explores the implementation of Deep Active Learning (DAL) to identify and characterise short duration transient noise in the GW signal stream. We employ a convolutional neural network (CNN) combined with the DBSCAN clustering algorithm to classify glitches detected by interferometers. Moreover, an Attention Layer is implemented to highlight the relevant areas of the images for the final classification. Our approach recognises patterns similar to previously identified signals and detects anomalous ones that could correspond to previously unseen phenomena.