Mémoire

Abstract

Gravitational waves are oscillations of spacetime itself. They are produced by the most powerful and extreme events in the Universe. Predicted by the theory of general relativity, these waves were first detected in September 2015. The merger of two massive black holes generated a spacetime deformation that was detected by the LIGO interferometers.

Other sources, such as core-collapse supernovae, are also believed to produce gravitational waves. The collapse of a massive star's core could generate signals that last less than a second, with waveforms that are not accurately known. As a result, traditional detection techniques, which rely on a good understanding of the targeted source's waveform, are ineffective. Deep learning techniques have been proposed as an alternative for detecting GW-generated power excess in time-frequency representations.

In this research project, we develop adaptations of the algorithm \textit{ALBUS} for the detection of gravitational wave signals from core-collapse supernovae. ALBUS, which stands for Anomaly detection for Long-duration BUrst Searches, was originally designed for the detection of minute-long transient gravitational waves. It generates a time-frequency map highlighting pixels identified as potential GW signals. This work demonstrates that short-duration signal detection is possible by training a neural network algorithm, thereby opening up new possibilities for developing GW detection pipelines that leverage the speed and accuracy of neural networks.