• GNN
• HEP
• Machine Learning
• NA62

In the last years Graph Neural Networks (GNNs) have been developed as a new kind of
Deep Learning technique that is based on exploiting the unordered nature of data. They have
been extensively analyzed and tested to perform energy reconstruction in calorimeters with
promising results. Inspired from this line of research, this work explores the feasibility of using
GNN models to reconstruct events with multiple tracks in the final state, using the information
coming from the STRAW spectrometer of the NA62 experiment at CERN.
The performance of the GNN model as multi-track trigger algorithm is evaluated on K3π data, in order to propose a possible improvement with respect to the present trigger condition.
A possible online selection for the dark neutrino channel is studied using a GNN model previously trained. Finally, future perspectives to explore and improve these techniques are presented.