17–20 Jun 2024
Hannover, Germany
Europe/Berlin timezone

A novel neural-network architecture for continuous-wave all-sky searches

20 Jun 2024, 12:10
25m
Hannover, Germany

Hannover, Germany

Speaker

Prasanna Joshi (Max Planck Institute for Gravitational Physics, Hannover)

Description

Continuous gravitational waves (CWs) are long-lasting gravitational waves emitted by rapidly spinning neutron stars that can be seen in the LIGO band. The most sensitive classical search method, the coherent matched filter search for continuous waves is not computationally feasible. Instead, a semi-coherent method is used for the search because it has a higher senesitivity than the coherent matched filter search, but at a reasonable computational cost. We present an alternative, new search method based on Deep Learning. In our study, we focus on training a Deep Neural Network (DNN) to perform a blind search for CWs emitted by isolated neutron stars over the whole sky. We have trained multiple DNNs with a convolutional neural network architecture to detect signals with a wide range of signal parameters. We highlight our specific architectural choices that have yielded good results after performing several experiments. We show that such a trained DNN can achieve a very high sensitivity on an all-sky search for continuous waves at a lower computational cost compared to more classical searches.

Primary authors

Prasanna Joshi (Max Planck Institute for Gravitational Physics, Hannover) Reinhard Prix (Albert-Einstein-Institut Hannover)

Presentation materials