What are the sources of ultra-high energy cosmic rays, those particles which rain down from the cosmos with colossal energies, and whose origin remains mysterious? The GRAND project (Giant Radio Array for Neutrino Detection) aims to answer this question, using a network of 200,000 radio antennas that will eventually be deployed by 2030 across the world. A first prototype of this project, GRANDProto300, a network made up of 300 antennas is being deployed in China.
The objective of this thesis is to develop machine learning algorithms to separate the signals of astrophysical origin from the radio background noise (stationary and transient), but also to reconstruct the properties (type, energy, direction) of particles arriving from the cosmos. This will involve developing new methods for time series signals in real time and in continuous flow, in a constrained environment (inference time, hardware consumption, limited communications). This thesis will therefore be an opportunity to develop new methods for model reduction and optimization of deep neural networks for applications less studied in the literature than vision aspects.
To carry out this work, the doctoral student will be able to rely on the simulation expertise of the GRAND team of the Laboratoire de Physique Nucléaire et des Hautes Énergies (CNRS IN2P3) to generate synthetic data sets and the data. by GRANDProto300. The student will also be associated with the GRAND collaboration.