R&D of the EM Calorimeter Energy Calibration with Machine Learning based on the low-level features of the Cluster

Aug 5, 2026

Upload by Miles

Developed a machine-learning energy calibration method for the ILC electromagnetic (EM) calorimeter (ECAL), a silicon-tungsten sampling calorimeter. A deep neural network performs regression to determine the energy of incident EM particles and improves ECAL energy resolution. The network architecture uses cluster hit information as low-level features. The paper presents the R&D status and reports results on energy calibration accuracy using SiD ECAL simulation data and a single-particle injected cluster construction.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages5

Reading Time13 min

AccessFree to download

Abstract

Introduction

Energy Calibration Using Machine Learning

Architecture of the Neural Network

R&D of the EM Calorimeter Energy Calibration with Machine Learning based on the low-level features of the Cluster Suzuna Morimasa 1 , ∗ , Masako Iwasaki 1,2,3 , ∗∗ , Taikan Suehara4 , Junichi Tanaka4 , Masahiko Saito4 , Hajime Nagahara3 , Yuta Nakashima3 , Noriko Takemura5 , and Takashi Nakano2 1 Osaka Metropolitan University Graduate School of Science, Osaka, Japan 2 Osaka University Research Center for Nuclear Physics (RCNP), Osaka, Japan 3 Osaka University Institute for Datability Science (IDS), Osaka, Japan 4The University of Tokyo, International Center for Elementary Partilce Physics (ICEPP), Tokyo Japan 5 Kyushu Institute of Technology, Fukuoka Japan Abstract. We have developed an energy calibration method using machine learning for the ILC electromagnetic (EM) calorimeter (ECAL), a sampling calorimeter consisting of Silicon-Tungsten layers. In this method, we use a deep neural network (DNN) for a regression to determine the energy of incident EM particles, improving the energy calibration resolution of the ECAL. The DNN architecture takes cluster hit data as low-level features of the cluster. In this paper, we report the status of our R&D and present results on energy calibration accuracy. 1 Introduction In the high energy colliding experiments, the precise energy measurement of the particle clusters detected with the electromagnetic (EM) and hadron calorimeters are crucial for the physics analyses. We have developed an energy calibration method using machine learning for the ILC EM calorimeter (ECAL), which is a sampling calorimeter that measures the energy of the particle cluster produced by the incident EM particle (electron or photon), for the precise energy determination. In our energy calibration method, we treat the energy calibration as a regression problem and use a deep neural network (DNN) architecture. Cluster hit data are used as low-level features in the DNN model. This paper reports the current status of the research and development (R&D) efforts. Both ILD and SiD detector designs employ the sampling type calorimeter for ECAL to measure the energies of incident electrons and photons for ILC experiments. In this study we use the ECAL detector simulation data for SiD. The SiD ECAL is the Silicon-Tungsten sampling type calorimeter, where absorbing Tungsten layers alternate with sensitive Silicon layers with 20 thin (2.5mm) Tungsten layers followed by 10 thick (5.0mm) layers for a total of 26X0. The design value of the energy resolution is ∆E/E = 0. 17/ √E ⊕ 0.01. The SiDECAL design is described in the ILC Technical Design Report (TDR)[1] . ∗ e-mail: [sd23697f@st.omu.ac.jp](sd23697f@st.omu.ac.jp) ∗∗[e-mail: masako.iwasaki@omu.ac.jp](e-mail: masako.iwasaki@omu.ac.jp) © The Authors, published by EDP Sciences. This is an open access article distributed under the terms of the Creative Commons Attribution License 4.0 ([https://creativecommons.org/licenses/by/4.0/](https://creativecommons.org/licenses/by/4.0/)). In this study, we use the MC data for SiD ECAL, here single EM particle is injected to the detector, and ECAL hits originating from the same incident particle are gathered as a cluster. Figure 1 shows the diagram of the ECAL which consists of twelve trapezoidal modules, with coordinate (left) and the schematic drawing of the hit production from an incident EM particle(right) . Figure 1. Diagram of the ECAL with coordinate (left) and the schematic drawing of the hit production from an incident EM particle(right), used in the study. In the conventional energy calibration, the energy of a cluster, which is regarded as an incident particle energy, is obtained by summing all hits in a cluster, and multiply by a single coefficient. We call the method the simple reconstruction. In the simple reconstruction, there exist several problems of : 1. Non-linearity between the true incident energy and the determined cluster energy due to the shower energy leakage in the high energy

FAQs

What is the proposed energy calibration method for the ILC EM calorimeter?

A regression deep neural network is used to determine incident EM particle energy, using cluster hit data as low-level features to improve calibration resolution.

What are the low-level input features used by the neural network?

The model directly takes all hit data in a cluster, where each hit includes energy, position coordinates (x, y, z), and the layer number.

What problems exist in conventional energy calibration?

Conventional “simple reconstruction” suffers from non-linearity at high energies due to shower leakage, particle-species dependence between electrons and photons, and angular dependence from detector geometry.

Aug 5, 2026

Upload by Miles