MACHINE LEARNING BASED EVENT RECONSTRUCTION FOR THE MUON E EXPERIMENT - DNN算法原型实现与验证

Aug 3, 2026

Upload by Violet

A proof-of-concept machine learning solution is implemented and tested within the MUonE experiment, targeting New Physics searches through the muon anomalous magnetic moment sector. Results from a deep neural network (DNN) based reconstruction method are comparable to classical track-and-vertex reconstruction while dramatically reducing execution time in the pattern recognition stage. The implementation satisfies classical reconstruction requirements and provides a strong foundation for further studies on higher-precision and faster event reconstruction in high-energy physics workflows.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages22

Reading Time55 min

AccessFree to download

Introduction

Track and vertex reconstruction

Particle track reconstruction in High Energy Physics experiments

State of the art

Computer Science • 25(1) 2024 [https://doi.org/10.7494/csci.2024.25.1.5690](https://doi.org/10.7494/csci.2024.25.1.5690) Abstract Keywords Citation Copyright Milosz Zdybal Marcin Kucharczyk Marcin Wolter MACHINE LEARNING BASED EVENT RECONSTRUCTION FOR THE MUONE EXPERIMENT A proof-of-concept solution based on the machine learning techniques has been implemented and tested within the MUonE experiment designed to search for New Physics in the sector of anomalous magnetic moment of a muon. The results of the DNN based algorithm are comparable to the classical reconstruction, reducing enormously the execution time for the pattern recognition phase. The present implementation meets the conditions of classical reconstruction, providing an advantageous basis for further studies. machine learning, artificial neural networks, track reconstruction, high energy physics Computer Science 25(1) 2024: 25–46 © 2024 Author(s) . This is an open access publication, which can be used, distributed and reproduced in any medium according to the Creative Commons CC-BY 4.0 License. 1. Introduction Significant developments have been applied during the last decades in the field of High Energy Physics (HEP) experiments, including computing technologies. Searches for New Physics phenomena, being an expansion of the so-called Standard Model, i.e. current incomplete theoretical knowledge about the basic behavior of the fundamental constituents of nature and the interactions between them, lead to experimental studies carried out at ever increasing energies. The number of particles created by the interaction of two particles (collision event) is generally increasing with the collision energy. As a consequence a huge number of charged particles have to be reconstructed ([e.g. in](e.g. in) proton-proton collisions), resulting in much more complex event patterns. A typical event in proton-proton collision showing the tracks of multiple particles passing through the detector is presented in Fig. 1, where the particles leave energy deposits (hits) in consecutive detector layers, being a basis for further track reconstruction. In order to enable the search for rare interesting collision events immersed in a huge background of events exhibiting well-known physics, the data rates related to the detector luminosity 1 have increased enormously (e.g. , 40 MHz readout rate in LHC) . It has to be reduced online by more than five orders of magnitude before the information from an event is written on mass storage for further analysis. Figure 1 . Example of an event in High Energy Physics experiment, showing tracks of multiple particles passing through the detector [31] . This paper aims to review the machine learning based approach applied in crucial stages of the data analysis process in HEP experiments, i.e. the procedure to determine basic kinematic parameters of charged particles at their point of production and the procedure to establish the location of these production points. They are commonly called track and vertex reconstruction. High density of tracks in a single 1 Luminosity translates to the number of collisions per second and it is related to track density. collision event (detector occupancy) in operating and planned high-energy physics experiments results in a large combinatorics of hits in the event pattern recognition. Therefore, a novel machine learning based event reconstruction algorithms have been developed and tested within a framework of the MUonE experiment [21] in order to maximize the statistical power of the final physics measurement. The results of the DNN based algorithm are comparable to the classical reconstruction, allowing not only to reduce execution time of the pattern recognition phase, but also to improve the precision and efficiency of the track and vertex reconstruction. 2. Particle track reconstruction in High Energy Physics experiments 2.1. State of the art In High Energy Physics experiments

FAQs

What problem does the document address in the MUonE experiment?

It addresses efficient reconstruction of particle tracks and production vertices under high event occupancy, where hit combinatorics makes pattern recognition computationally expensive.

How does the proposed method compare with classical reconstruction?

The DNN-based algorithm yields results comparable to classical reconstruction while greatly reducing execution time for the pattern recognition phase.

What is the role of track and vertex reconstruction in the analysis pipeline?

Track and vertex reconstruction determine key charged-particle kinematic parameters and interaction-point locations, which then support later event-level physical measurements.

Aug 3, 2026

Upload by Violet