Neutrinos are the lightest matter particles we know about. They come from the deep Universe and carry information about astrophysical sources and the evolution of our matter-dominated Cosmos. To discriminate these cosmic neutrinos from other particles interacting in our detectors we need to accurately predict their properties, such as the neutrino energy. This is one of the most important and challenging tasks in neutrino telescopes (such as KM3NeT) as it can provide direct information on the mechanisms that produce these particles in the Universe and lead to new discoveries.
The aim of this project is to employ Graph Neural Networks (GNNs) to accurately predict the neutrino energy deposited in the KM3NeT detector volume. The goal is to interpret the GNN predictions with respect to physics observables and experimental data, investigate possible biases and scale the initial training to expanding datasets (corresponding to larger detector volumes). This necessity for scalable models comes from the fact that the KM3NeT detectors are under construction with their volume expanding fast.
Background:
KM3NeT is a research infrastructure housing two underwater Cherenkov telescopes located in the Mediterranean Sea. It consists of two configurations which are currently under construction: ARCA with 230 detection units corresponding to 1 cubic kilometre of instrumented water volume and ORCA with 115 detection units corresponding to a mass of 7 Mton. The ARCA (Astroparticle Research with Cosmics in the Abyss) detector aims at studying neutrinos with energies in the TeV-PeV range coming from distant astrophysical sources, while the ORCA (Oscillation Research with Cosmics in the Abyss) detector is optimised for atmospheric neutrino oscillation studies at energies of a few GeV. Despite being currently under construction ARCA and ORCA are already taking data. This project aims to use simulated and experimental data from the ARCA detector.
Artificial intelligence is increasingly used in KM3NeT for data processing and analysis, aiming to provide a better performance on event reconstruction and significantly faster inference times compared to traditional reconstruction techniques. GNNs have been successfully employed for event classification and neutrino property regression tasks, such as the energy reconstruction. Despite the very promising results of the GNNs we are still lacking a deep understanding of the biases and systematic uncertainties that they inherit to their predictions. Another issue is the requirement for large datasets in order to train such algorithms which is not always possible. As ARCA is a detector under development its volume is progressively increasing.
The project is conducted at the National Centre of Scientific Research Demokritos and involves the synergy of Software and Knowledge Engineering Lab (SKEL | The AI lab) of the Institute of Informatics Telecommunications and the Molecular Thermodynamics and the Astroparticle team (APP) of the Institute of Nuclear and Particle Physics.
Since the project can be either an internship (at least 2 months full time), or a BSc/MSc thesis, the exact assignment will be decided on a case-by-case basis with the interested student.
The thesis will be jointly supervised by Dr. Evangelia Drakopoulou, drakopoulou@inp.demokritos.gr
Relevant literature:
[1] F. Filippini et al., Data reconstruction and classification with Graph neural networks in KM3NeT/ARCA, (2023) DOI: 10.22323/1.444.1194
[2] D. Guderian, PhD thesis, Development of detector calibration and graph neural network-based selection and reconstruction algorithms for the measurement of oscillation parameters with KM3NeT/ORCA, (2022)