• Gravitational waves
• Machine Learning
• Detector Characterization
• Advanced Virgo
• Noise Hunting
• Convolutional neural network

Gravitational wave interferometers, such as Advanced Virgo (AdV), are sensitive and complex detectors that produce large amounts of data. In fact, detectors have, in addition to the output strain, hundreds of thousands of auxiliary channels that give information about the detector status. Machine learning is a promising technique for analyzing detector data quickly and thus detecting GWs and investigating noise sources. The purpose of this thesis is to apply a supervised deep learning algorithm to characterize noise in Advanced Virgo, in particular, in order to identify drifting lines starting from their time frequency evolution represented as images. In fact, detectors are affected by various sources of noise that decrease their sensitivity and stability. Drifting lines are a continuous noise source of a non-stationary nature. Their source is often unknown and they may be caused by the detector itself or by environmental sources. Statistical tools have been developed to characterize drifting lines: algorithms that track their time evolution and coherence tools dedicated to locating their origin. However, these systems take long time and must be accompanied by continuous manual inspections. We are looking for a fast and automatic way to detect them and deep learning is very promising. In this regard, I developed, trained, and tested a convolutional neural network (CNN) with an appropriate data set of simulated noise lines. In particular, we want to train CNN to classify noise lines into drifting and stationary lines. To generate drifting lines similar to those found in Advanced Virgo I developed a custom simulation by looking at drifting lines data from AdV, that models their frequency evolution in time. The network achieves high levels of classification accuracy (≥ 99%). As a proof of concept, this thesis applies convolutional neural networks to drifting lines analysis. A new approach for studying drifting lines, and a first step
toward applying it to real data collected by Advanced Virgo.