Speaker
Description
The NA62 experiment at the Super Proton Synchrotron at CERN is designed to measure the branching ratio of the ultra-rare channel $K^+ \rightarrow \pi^+\nu \bar{\nu}$ with BR = $(8.6 \pm 0.42)\times 10^{-11}$. The GigaTracker, a silicon pixel detector, measures the momentum and direction of incoming hadron beam particles at a rate of 750 MHz.
The success of the experiment relies on effectively managing pile-up, temporal matching, and vertex reconstruction between beam kaons and decay pions, a challenging task due to the combinatorial track-building approach currently adopted. This research is focused on integrating machine learning best practices in the current modeling strategies applied in the NA62 experiment. In this study, we present an extensive evaluation of different deep learning architectures for particle tracking, addressing the limitations of inefficient methods and showing the most effective approaches. In particular, we designed and implemented three distinct approaches based on Multi-Layer Perceptron, Transformer and Graph Neural Network. All the methods were evaluated on the basis of the efficiency, the purity and the fake tracks. The best results showed high efficiency and purity levels with a low fake tracks ratio.
