Machine Learning Power Week 2023: Clustering in Hadronic Calorimeters

M. Al Halabi, M. Bajdel, J.P. Bormans, H. Bossi (ed), M. Calmon Behling, F. Ehmann, N. Götz, J. Jung, R. Kavak, M. Krüger, R. Lakos, A. Lauterbach, P. Mccormack (ed), A. Mithran, D. Murnane (ed), M. Nolte, T.S. Rogoschinski, J. Scharf, O. Tyagi, L. Våge (ed), M. Valialshchikov, S. Wagner · 2025

Read Machine Learning Power Week 2023: Clustering in Hadronic Calorimeters by M. Al Halabi, M. Bajdel, J.P. Bormans, H. Bossi (ed), M. Calmon Behling, F. Ehmann, N. Götz, J. Jung, R. Kavak, M. Krüger, R. Lakos, A. Lauterbach, P. Mccormack (ed), A. Mithran, D. Murnane (ed), M. Nolte, T.S. Rogoschinski, J. Scharf, O. Tyagi, L. Våge (ed), M. Valialshchikov, S. Wagner (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2508-09938.txt.

Introduction

Muaz Al Halabi and a large student team report a one-week machine learning workshop that tackled a real detector problem. The challenge was sorting energy deposits in the forward hadronic calorimeter planned for the ePIC detector at the Electron-Ion Collider. Graduate students learned machine learning while producing candidate solutions.

The paper describes the different approaches the groups built during that single week and compares their performance. Clustering shower signals correctly matters for reading future collision data. The authors also gather feedback on the workshop format itself. Their recommendations aim to help others run similar events that teach students while crowd-sourcing hard problems.

Catalog record

CREATORM. Al Halabi, M. Bajdel, J.P. Bormans, H. Bossi (ed), M. Calmon Behling, F. Ehmann, N. Götz, J. Jung, R. Kavak, M. Krüger, R. Lakos, A. Lauterbach, P. Mccormack (ed), A. Mithran, D. Murnane (ed), M. Nolte, T.S. Rogoschinski, J. Scharf, O. Tyagi, L. Våge (ed), M. Valialshchikov, S. Wagner
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE50 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2508-09938

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Machine Learning Power Week 2023: Clustering in Hadronic Calorimeters M. Al Halabi, M. Bajdel, J.P. Bormans, H. Bossi (ed), M. Calmon Behling, F. Ehmann, N. Götz, J. Jung, R. Kavak, M. Krüger, R. Lakos, A. Lauterbach, P. Mccormack (ed), A. Mithran, D. Murnane (ed), M. Nolte, T.S. Rogoschinski, J. Scharf, O. Tyagi, L. Våge (ed), M. Valialshchikov, S. Wagner Abstract Abstract In both high-energy physics and industry applications, a crowd-sourced approach to difficult problems is becoming increasingly common. These innovative approaches are ideal for the development of future facilities where the simulations can be publicly distributed, such as the Electron-Ion Collider (EIC). In this paper, we discuss a so-called “Power Week” where graduate students were able to learn about machine learning while also contributing to an unsolved problem at a future facility. Here, the problem of interest was the clustering of the forward hadronic calorimeter in the foreseen electron-proton/ion collider experiment (ePIC) detector at the EIC. The different possible approaches, developed over the course of a single week, and their performance are detailed and summarised. Feedback on the format of the week and recommendations for future similar programs are provided in the hopes to inspire future learning opportunities for students that also serve as a crowd-sourced approaches to unsolved problems. 1 Introduction In recent years, both private companies and scientific collaborations have turned to crowd-sourcing platforms in order to implement real solutions to problems that are challenging both computationally and scientifically [ 1 , 2 , 3 , 4 , 5 ] . Such approaches have proven successful at brainstorming new, innovative solutions to long-standing problems. One barrier to such crowd-sourcing being widely adopted in high-energy physics (HEP) contexts is the difficulty of releasing data publicly. Future facilities where details of detector configurations, technologies, resolution, etc., are not yet finalised, offer one potential context within HEP that avoids this difficulty. In addition, crowd-sourcing such approaches allows for innovative and creative solutions to long-standing problems. The Electron-Ion Collider (EIC) is a planned particle accelerator facility that will collide electrons of various energies with protons and nuclei [ 6 , 7 ] . The facility will be constructed at Brookhaven National Laboratory (BNL) on Long Island in the United States and will make use of the existing ion ring from the Relativistic Heavy Ion Collider (RHIC) facility. There is one full-acceptance general-purpose detector, referred to as the electron/Proton Ion Collider Experiment (ePIC), currently planned for the EIC. Some physics goals of the EIC include; determining how the properties of nucleons emerge from partons and their interactions, how partons are distributed within nucleons, how hadrons are formed from the confinement of partons, and more [ 7 ] . In order to accomplish these physics goals, the EIC will be constructed as one of the most sophisticated facilities to date. Towards this aim, there are large efforts in place in order to incorporate advanced techniques in the design, operation, reconstruction simulation, and physics analysis at the EIC [ 8 ] . As a part of these efforts, an ML Power Week was organised as a part of the Helmholtz Graduate School for Hadron and Ion Research, which took place from July 18th - 22nd, 2023, in Schmitten, Germany. This “Power Week” focused on the development of a clustering algorithm for the ePIC forward hadronic calorimeter. The Power Week was organised as a private Kaggle Competition [ 9 ] where participants would compete in groups in order to create the clustering algorithm with the best performance (see Section 2.4 for more details). In this paper, the setup and results of this competition are summarised. In Section 2 , the data set and setup of the challenge will be described. In Section 3 , previous approaches to clustering-based problems will be presented. In Section 4 , the various approaches that were attempted to solve this problem will be discussed. In Section 5 the results will be discussed in context and some recommendations for the format of future workshops will be provided. Finally, in Section 6 , the Power Week and its main takeaways will be summarised. 2 Dataset and Challenge For this Power Week, a simulation of a so-called “particle gun” hitting the ePIC forward hadronic calorimeter was used. In the following sections, the details of the calorimeter (Section 2.1 ), the simulation used (Section 2.2 ), as well as the setup of the competition (Section 2.3 ) and the performance metric used (Section 2.4 ) will be discussed. 2.1 The ePIC Forward Hadronic Calorimeter The forward electromagnetic and hadronic calorimeters in the hadron-going endcap of the ePIC detector are essential subdetectors for accomplishing its physics goals. This is primarily due to the fact that the incoming proton/nuclei will have significantly higher kinetic energy than the electron, and therefore, a majority of hadrons will occur in the forward region. The measurement of particles in the forward region will additionally rely on the calorimeter system as tracking momentum and resolution worsens at very forward values of the pseudorapidity. In this paper, the design of the forward hadronic calorimeter of the ePIC detector will refer to the design as it existed as of the R&D proposal # 107 1 1 1 See https://wiki.bnl.gov/conferences/images/8/8f/ERD_107_2023_submitted.pdf for details. that was submitted in July 2023. These conditions are subject to change as detector conditions evolve over time in conjunction with developments in the other subdetectors of ePIC. The detector part of interest is commonly referred to as the longitudinally segmented forward hadronic calorimeter (LFHCal) and is a steel-scintillating sampling calorimeter with its first segment made out of tungsten and scintillators, serving as a collimator. The LFHCal consists of two movable half-discs such that the calorimeter can be wheeled out of the way to give access to the internal detector components. This disc, one half of which is shown in the left-hand side of Figure 1 , has a radius of 270 cm, an active depth of 130 cm, and 8916 towers with a 5 cm X 5 cm front face. Each tower consists of so-called 8M modules arranged in a Lego-like structure. One such module is shown in the right-hand side of Figure 1 , and comprises 8 towers in a 4 by 2 tower arrangement, as seen in the top left of Figure 1 displaying a single 8M module layer and its segmentation. Each tower consists of 7 different readout channels - forming 7 independent layers in the z-direction. The asymmetric structures immediately surrounding the beam pipe have been added in order to increase the coverage of the calorimeter. See the eRD # 107 linked for more details. 2 2 2 For the updated LFHCal design see https://wiki.bnl.gov/EPIC/index.php?title=Forward_Hcal . Figure 1 : 3D rendering of the design of the LFHCal (left) and a zoomed-in rendering of a single 8M module (bottom). Individual components of the 8M module are shown in the top right corner, as well as the stacking concept (middle right). Please note that this is the design as of R&D proposal # 107, submitted July 2023, and as a result and has since changed. See https://wiki.bnl.gov/EPIC/index.php?title=Forward_Hcal for the most up to date design. Note that both of these renderings originally appeared in the R&D proposal # 107. 2.2 The

Read the complete text