Transforming jet flavour tagging at ATLAS

The ATLAS collaboration

doi:10.1038/s41467-025-65059-6

Transforming jet flavour tagging at ATLAS

The ATLAS collaboration

CPPM
University of Michigan, Ann Arbor
Nanjing University
Tsinghua University
SLAC National Accelerator Laboratory
Carleton University
University of Washington
Université Paris 11
Shandong University
University College London
University of Tokyo
Universite Paul Sabatier
Shefield University
Chinese Academy Of Sciences
European Organization for Nuclear Research
Shanghai Jiao Tong University
University of Glasgow
Deutsches Elektronen-Synchrotron (DESY)
University of Illinois
Argonne National Laboratory
Harvard University
The Hong Kong University of Science and Technology
Lawrence Berkeley National Laboratory
University Montreal
University of Geneva
Brandeis University
Ohio State University
The University of British Columbia
Department of Physics
Univ. Paris-Diderot
Campus UAB
ICREA
Academia Sinica, Institute Of Physics
University of Chinese Academy of Sciences
Sun Yat-Sen University
New York University
McGill University
CNRS Centre National de la Recherche Scientifique
University of Manchester
University of the Witwatersrand, Johannesburg
University of Pittsburgh
Department of Physics, Lancaster University
Osaka University
University of Bergen
Universität Würzburg
Université Clermont Auvergne
TU Dortmund University
University of Edinburgh
Michigan State University
University of Hong Kong
Simon Fraser University

Research output: Contribution to journal › Article › peer-review

4 Scopus citations

Abstract

Jet flavour tagging enables the identification of jets originating from heavy-flavour quarks in proton–proton collisions at the Large Hadron Collider, playing a critical role in its physics programmes. This paper presents GN2, a transformer-based flavour tagging algorithm deployed by the ATLAS Collaboration that represents a different methodology compared to previous approaches. Designed to classify jets based on the flavour of their constituent particles, GN2 processes low-level tracking information in an end-to-end architecture and incorporates physics-informed auxiliary training objectives to enhance both interpretability and performance. Its performance is validated in both simulation and collision data. The measured c-jet (light-jet) rejection in data is improved by a factor of 3.5 (1.8) for a 70% b-jet tagging efficiency, compared to the previous algorithm. GN2 provides substantial benefits for physics analyses involving heavy-flavour jets, such as measurements of Higgs boson pair production and the couplings of bottom and charm quarks to the Higgs boson, and demonstrates the impact of advanced machine learning methods in experimental particle physics.

Original language	British English
Article number	541
Journal	Nature Communications
Volume	17
Issue number	1
DOIs	https://doi.org/10.1038/s41467-025-65059-6
State	Published - Dec 2026

Access to Document

10.1038/s41467-025-65059-6

OpenUrl availability

Full text

Cite this

@article{e266892d0344462a9230f900c6ea3d9b,

title = "Transforming jet flavour tagging at ATLAS",

abstract = "Jet flavour tagging enables the identification of jets originating from heavy-flavour quarks in proton–proton collisions at the Large Hadron Collider, playing a critical role in its physics programmes. This paper presents GN2, a transformer-based flavour tagging algorithm deployed by the ATLAS Collaboration that represents a different methodology compared to previous approaches. Designed to classify jets based on the flavour of their constituent particles, GN2 processes low-level tracking information in an end-to-end architecture and incorporates physics-informed auxiliary training objectives to enhance both interpretability and performance. Its performance is validated in both simulation and collision data. The measured c-jet (light-jet) rejection in data is improved by a factor of 3.5 (1.8) for a 70\% b-jet tagging efficiency, compared to the previous algorithm. GN2 provides substantial benefits for physics analyses involving heavy-flavour jets, such as measurements of Higgs boson pair production and the couplings of bottom and charm quarks to the Higgs boson, and demonstrates the impact of advanced machine learning methods in experimental particle physics.",

author = "\{The ATLAS collaboration\} and J. Zhu and Y. Zhou and Y. Zhou and Y. Zhou and H. Zhou and B. Zhou and Z. Zheng and Y. Zhao and H. Zhao and Z. Zhang and Z. Zhang and Y. Zhang and Y. Zhang and Y. Zhang and Y. Zhang and T. Zhang and S. Zhang and Zhang, \{D. F.\} and Y. Yang and Y. Yang and X. Yang and X. Yang and Yang, \{H. J.\} and Yang, \{H. J.\} and S. Yan and Y. Xu and J. Wu and Z. Wang and Z. Wang and Y. Wang and Y. Wang and X. Wang and X. Wang and X. Wang and R. Wang and R. Wang and J. Wang and H. Wang and M. Usman and Y. Tian and R. Soualah and S. Singh and S. Singh and S. Shrestha and P. Sharma and Sharma, \{A. S.\} and Sharma, \{A. S.\} and Shah, \{A. H.\} and S. Parajuli and L. Panwar and J. Pan and M. Martinez and Martin, \{T. A.\} and Lu, \{Y. J.\} and Z. Liu and Liu, \{Y. W.\} and Liu, \{Y. W.\} and Liu, \{Y. W.\} and Q. Liu and P. Liu and Liu, \{J. K.K.\} and Z. Li and Z. Li and Z. Li and X. Li and T. Li and S. Li and S. Li and R. Li and L. Li and K. Li and H. Li and H. Li and H. Li and H. Li and B. Li and P. Kumari and M. Kumar and Khan, \{R. A.\} and Jones, \{R. W.L.\} and X. Jia and X. Huang and M. Hirose and S. Hassan and M. Haleem and A. Haddad and A. Gupta and Y. Guo and Y. Gao and Y. Fu and D. Francis and Fan, \{K. Y.\} and Didenko, \{A. R.\} and Das, \{S. J.\} and Z. Chen and J. Chen and J. Chen and H. Cai and M. Barakat and Anderson, \{A. C.\} and S. Ali",

note = "Publisher Copyright: {\textcopyright} The Author(s) 2025.",

year = "2026",

month = dec,

doi = "10.1038/s41467-025-65059-6",

language = "British English",

volume = "17",

journal = "Nature Communications",

issn = "2041-1723",

publisher = "Nature Research",

number = "1",

}

TY - JOUR

T1 - Transforming jet flavour tagging at ATLAS

AU - The ATLAS collaboration

AU - Zhu, J.

AU - Zhou, Y.

AU - Zhou, H.

AU - Zhou, B.

AU - Zheng, Z.

AU - Zhao, Y.

AU - Zhao, H.

AU - Zhang, Z.

AU - Zhang, Y.

AU - Zhang, T.

AU - Zhang, S.

AU - Zhang, D. F.

AU - Yang, Y.

AU - Yang, X.

AU - Yang, H. J.

AU - Yan, S.

AU - Xu, Y.

AU - Wu, J.

AU - Wang, Z.

AU - Wang, Y.

AU - Wang, X.

AU - Wang, R.

AU - Wang, J.

AU - Wang, H.

AU - Usman, M.

AU - Tian, Y.

AU - Soualah, R.

AU - Singh, S.

AU - Shrestha, S.

AU - Sharma, P.

AU - Sharma, A. S.

AU - Shah, A. H.

AU - Parajuli, S.

AU - Panwar, L.

AU - Pan, J.

AU - Martinez, M.

AU - Martin, T. A.

AU - Lu, Y. J.

AU - Liu, Z.

AU - Liu, Y. W.

AU - Liu, Q.

AU - Liu, P.

AU - Liu, J. K.K.

AU - Li, Z.

AU - Li, X.

AU - Li, T.

AU - Li, S.

AU - Li, R.

AU - Li, L.

AU - Li, K.

AU - Li, H.

AU - Li, B.

AU - Kumari, P.

AU - Kumar, M.

AU - Khan, R. A.

AU - Jones, R. W.L.

AU - Jia, X.

AU - Huang, X.

AU - Hirose, M.

AU - Hassan, S.

AU - Haleem, M.

AU - Haddad, A.

AU - Gupta, A.

AU - Guo, Y.

AU - Gao, Y.

AU - Fu, Y.

AU - Francis, D.

AU - Fan, K. Y.

AU - Didenko, A. R.

AU - Das, S. J.

AU - Chen, Z.

AU - Chen, J.

AU - Cai, H.

AU - Barakat, M.

AU - Anderson, A. C.

AU - Ali, S.

PY - 2026/12

Y1 - 2026/12

N2 - Jet flavour tagging enables the identification of jets originating from heavy-flavour quarks in proton–proton collisions at the Large Hadron Collider, playing a critical role in its physics programmes. This paper presents GN2, a transformer-based flavour tagging algorithm deployed by the ATLAS Collaboration that represents a different methodology compared to previous approaches. Designed to classify jets based on the flavour of their constituent particles, GN2 processes low-level tracking information in an end-to-end architecture and incorporates physics-informed auxiliary training objectives to enhance both interpretability and performance. Its performance is validated in both simulation and collision data. The measured c-jet (light-jet) rejection in data is improved by a factor of 3.5 (1.8) for a 70% b-jet tagging efficiency, compared to the previous algorithm. GN2 provides substantial benefits for physics analyses involving heavy-flavour jets, such as measurements of Higgs boson pair production and the couplings of bottom and charm quarks to the Higgs boson, and demonstrates the impact of advanced machine learning methods in experimental particle physics.

AB - Jet flavour tagging enables the identification of jets originating from heavy-flavour quarks in proton–proton collisions at the Large Hadron Collider, playing a critical role in its physics programmes. This paper presents GN2, a transformer-based flavour tagging algorithm deployed by the ATLAS Collaboration that represents a different methodology compared to previous approaches. Designed to classify jets based on the flavour of their constituent particles, GN2 processes low-level tracking information in an end-to-end architecture and incorporates physics-informed auxiliary training objectives to enhance both interpretability and performance. Its performance is validated in both simulation and collision data. The measured c-jet (light-jet) rejection in data is improved by a factor of 3.5 (1.8) for a 70% b-jet tagging efficiency, compared to the previous algorithm. GN2 provides substantial benefits for physics analyses involving heavy-flavour jets, such as measurements of Higgs boson pair production and the couplings of bottom and charm quarks to the Higgs boson, and demonstrates the impact of advanced machine learning methods in experimental particle physics.

UR - https://www.scopus.com/pages/publications/105027769260

U2 - 10.1038/s41467-025-65059-6

DO - 10.1038/s41467-025-65059-6

M3 - Article

C2 - 41535252

AN - SCOPUS:105027769260

SN - 2041-1723

VL - 17

JO - Nature Communications

JF - Nature Communications

IS - 1

M1 - 541

ER -

Transforming jet flavour tagging at ATLAS

Abstract

Access to Document

OpenUrl availability

Other files and links

Fingerprint

Cite this