数据与计算发展前沿 ›› 2024, Vol. 6 ›› Issue (3): 108-115.

CSTR: 32002.14.jfdc.CN10-1649/TP.2024.03.012

doi: 10.11871/jfdc.issn.2096-742X.2024.03.012

• 会议论文 • 上一篇    下一篇

CEPC上基于DeepSets模型的喷注标记算法研究

廖立波1,2(),王书栋3,4,宋维民5,张兆领5,李刚3,黄永盛1,*()   

  1. 1.中山大学，理学院，广东 深圳 518107
    2.梧州学院，广西机器视觉与智能控制重点实验室，广西 梧州 54300
    3.中国科学院高能物理研究所，北京 100049
    4.中国科学院大学，北京 100049
    5.吉林大学，物理学院，吉林 长春 130000
  • 收稿日期:2023-10-31 出版日期:2024-06-20 发布日期:2024-06-21
  • 通讯作者: *黄永盛（E-mail: huangysh59@mail.sysu.edu.cn）
  • 作者简介:廖立波，中山大学理学院，在读博士，主要研究方向为粒子物理实验。
    负责论文内容的研究和撰写。
    LIAO Libo, School of Science, Shenzhen campus of Sun Yat-sen University, Ph.D student, mainly research is particle physics experiment.
    He is responsible for research and writing of the article.
    E-mail：liaolibo@ihep.ac.cn|黄永盛，中山大学理学院，教授，博士，主要研究方向为高亮度激光逆康普顿光源及其应用开拓、CEPC束流能量标定、极化测量，CEPC同步辐射光源探测及其应用开拓、γγ对撞机设计与分析、γ与中子计量资质获取，激光等离子体相互作用产生强THz波，新加速原理加速正电子以及正负谬子束，暗光子理论研究等。
    本文提供研究指导和支持。
    HUANG Yongshen, School of Science, Shenzhen campus of Sun Yat-sen University, professor, doctor, mainly research is high luminosity laser inverse Compton light source and its application, calibration of beam energy in CEPC, polarize measurement, detection of synchrotron radiation light source and its application in CEPC, gamma-gamma collider design and analysis, metrological qualification acquisition of gamma and neutron, strong THz wave productive with laser plasma interaction, positron and muon beam acceleration with new acceleration theory, dark photon theory research.
    In this paper, he provides invaluable guidance and support throughout research。
    E-mail：huangysh59@mail.sysu.edu.cn
  • 基金资助:
    广西教育厅高校中青年教师科研基础能力提升项目“基于深度学习的喷注电荷鉴别研究”(2023KY0707)

The Study of Jet Tagging Algorithm Based on DeepSets at CEPC

LIAO Libo1,2(),WANG Shudong3,4,SONG Weimin5,ZHANG Zhaoling5,LI Gang3,HUANG Yongsheng1,*()   

  1. 1. School of Science, Sun Yat-Sen University, Shenzhen, Guangdong 518107, China
    2. Guangxi Key Laboratory of Machine Vision and Intelligent Control, Wuzhou University, Wuzhou, Guangxi 543000, China
    3. Institute of High Energy Physics, Chinese Academy of Sciences, Beijing 100049, China
    4. University of Chinese Academy of Sciences, Beijing 100049, China
    5. College of Physics, Jilin University, Changchun, Jilin 130000, China
  • Received:2023-10-31 Online:2024-06-20 Published:2024-06-21

RichHTML

2

PDF

145

摘要：

【目的】在希格斯玻色子被证实存在之后，精确测量希格斯玻色子性质以及检验电弱模型就成了高能物理领域今后一段时间内的重点工作，对喷注进行更加细致地研究，可以有效地提升测量和检验的精确度。【方法】本研究采用了新提出的、基于DeepSets的粒子流网络深度学习算法以及Z玻色子强衰变末态产生的喷注全模拟数据集，尝试对重味夸克喷注的标记算法进行研究。【结果】经过初步研究，与XGBoost等传统算法和DNN全连接网络相比，新算法性能提高约6%，使平均精度从80%提高到85%左右。【结论】这表明深度学习在高能物理实验中存在巨大的潜力，在粒子重建、事例分析等多方面有较好的应用场景，但还需更加深入地研究深度学习方法带来的影响，以提升深度学习的可信度和效果。

关键词: 深度学习, 粒子流网络, 喷注标记

Abstract:

[Purpose] After the existence of the Higgs boson was confirmed, precision measurements of its properties and searches for new physics beyond the Standard Model become the focus of the field of high energy physics in the future, and better detection of the jet can effectively improve the precision and sensitivity. [Method] In this study, a newly proposed DeepSets-based deep learning algorithm, the Particle Flow network, and a collection of full simulation data sample of jets generated by the hadronic decay of the Z boson are used to study the tagging algorithm of heavy flavor jets. [Results] Compared with the CEPC baseline algorithm, the performance of the flavor tagging is improved by about 6%, i.e., the average accuracy is increased from 80% to 85%. [Conclusion] This study shows that deep learning has great potential in high-energy physics experiments, and has wide application scenarios in particle reconstruction, physics analysis, and other aspects. However, more in-depth research on deep learning methods is needed to enhance the credibility and robustness.

Key words: deep learning, particle flow network, jet flavor tagging