A new Machine Learning-based method for identiﬁcation of time-correlated events at tagged photon facilities - Abstract

Aug 3, 2026

Upload by Himbo

A new machine-learning multivariate analysis method is introduced for selecting time-correlated hits in tagging systems used at bremsstrahlung-based tagged-photon facilities. The approach replaces the commonly used sampling and subtraction of time-uncorrelated background, which can significantly increase uncertainties in high-precision experiments. Machine-learning classification also preserves correlations among kinematic variables in the final state, benefiting subsequent phenomenological analyses of experimental outcomes. The method is motivated by measurement limitations caused by random background at high rates.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages12

Reading Time30 min

AccessFree to download

Introduction

Selection of time-correlated hits with a conventional approach

New analysis concept

Application of the Machine Learning-based method

Preparation of the input data

Application of ensemble learning with boosted decision trees using CatBoost

Summary

arXiv:2307.04776v1 [[physics.data-an](physics.data-an)] 9 Jul 2023 Prepared for submission to JINST A new Machine Learning-based method for identiﬁcation of time-correlated events at tagged photon facilities V. Sokhoyan,􀀰, 1 E. Mornacchi 􀀰 􀀰 Institut für Kernphysik, University of Mainz, Johann-Joachim-Becher-Weg 45, D-55099 Mainz, Germany E-mail: [sokhoyan@uni-mainz.de](sokhoyan@uni-mainz.de) Abstract: We present a new Machine Learning-based multivariate analysis method for the selection of time-correlated hits in the tagging system and devices used to detect particles in the ﬁnal state atthe bremsstrahlung-based tagged photon facilities. This method can be applied instead of the widely used sampling and subtraction of the time-uncorrelated background, in particular at experiments aiming for high precision, where the subtraction of the time-uncorrelated background leads to increased uncertainties. Moreover, the identiﬁcation of events with Machine Learning algorithms allows to preserve the information about correlations of kinematic variables in the ﬁnal state, which can be advantageous for further phenomenological analyses of the experimental results. Keywords: Particle identiﬁcation methods, Pattern recognition, Analysis and statistical methods, Timing detectors 1Corresponding author. Contents 1 Introduction 1 2 Selection of time-correlated hits with a conventional approach 2 3 New analysis concept 4 4 Application of the Machine Learning-based method 4 4.1 Preparation of the input data 5 4.2 Application of ensemble learning with boosted decision trees using CatBoost 6 5 Summary 10 1 Introduction The accuracy of the measurements performed at electron accelerators with bremsstrahlung-based tagged photons is limited, among other factors, by the presence of the time-uncorrelated background in the tagging system at high rates, complicating unambiguous matching of the initial electron and the bremsstrahlung photon inducing the reaction in the target. The widely used sampling and subtraction of time-correlated events leads to a reduction in the measurement accuracy, depending on the degree of contamination of the data sample with time-uncorrelated background. The method presented in this paper allows for the identiﬁcation of the time-correlated events without subtraction of the random (uncorrelated) background. To illustrate the application of the method, we have chosen the reaction 􀁗 􀀿 → 􀀿􀁣0 for a selected photon beam energy range of 240 − 260 MeV, where no other reactions are expected to contribute to the sample with two photons and one proton candidate in the ﬁnal state. In addition, kinematic cuts were applied for reaction identiﬁcation and background suppression (see section 4.1) . After the initial event selection, the time-correlated (signal) events in the experimental data are expected to have the properties of a simulated 􀁗 􀀿 → 􀀿􀁣0 reaction, under the assumption that the simulation reasonably describes the data. Afterwards, the samples obtained from Monte Carlo simulation of the signal and experimental measurement of the time-uncorrelated (random) background are used to create Machine Learning (ML) models, which are applied to distinguish between time-correlated and uncorrelated events in the experimental data. This paper is organized as follows. Section 2 introduces the experimental setup and the conventional method for the subtraction of the random background coincidences in the tagging system. Section 3 explains the concept of the new method, while the application of the ML models, including the comparison of the new method with the conventional subtraction approach, is discussed in section 4. The outcome of this work is summarized in section 5. Figure 1 . Schematic setup of the Crystal Ball/TAPS experiment at MAMI [3] . 2 Selection of time-correlated hits with a conventional approach In this paper, the application of a new method for the identiﬁcation of time-correlated events

FAQs

为什么需要识别 time-correlated events？

在 bremsstrahlung-based tagged photon 实验中，高速率下存在 time-uncorrelated background，导致初始电子与诱发反应的刹那难以无歧义匹配，从而限制测量精度。

传统方法是如何处理 time-uncorrelated background 的？

传统做法通过对 time-correlated 事件进行采样，并减去 time-uncorrelated background。然而，背景污染程度越高，减法带来的不确定度就越大。

新的机器学习方法带来哪些优势？

新方法不需要对随机（不相关）背景进行减法即可区分 time-correlated 与 uncorrelated 事件，并能保留终态中动量学变量的相关性，从而有利于后续现象学分析。

Aug 3, 2026

Upload by Himbo