Development of systematic uncertainty-aware neural network trainings for binned-likelihood analyses at the LHC

CMS Collaboration

doi:10.1140/epjc/s10052-025-14713-w

Development of systematic uncertainty-aware neural network trainings for binned-likelihood analyses at the LHC

CMS Collaboration

CERN
California Institute of Technology
Texas A&M University
Texas A&M University at Qatar
Academy of Sciences of the Republic of Uzbekistan
A. Alikhanian Yerevan Institute of Physics
Imperial College London
University of Wisconsin-Madison
Wayne State University
University of Virginia
Vanderbilt University
Texas Tech University
Kyungpook National University
Swiss Federal Institute of Technology Lausanne
University of Tennessee, Knoxville
Rutgers - The State University of New Jersey, New Brunswick
University of Rochester
Rice University
Purdue University Northwest
Purdue University
University of Puerto Rico
Princeton University
Ohio State University
University of Notre Dame
Northwestern University
Northeastern University
SUNY Buffalo
University of Nebraska-Lincoln
University of Minnesota Twin Cities
Massachusetts Institute of Technology
University of Maryland, College Park
Kansas State University
University of Kansas
Horia Hulubei National Institute of Physics and Nuclear Engineering
Johns Hopkins University
University of Iowa
Erciyes University
Sinop University
Georgian Technical University
Bingol University
University of Illinois at Chicago
Florida Institute of Technology
Ain Shams University
Florida State University
University of Florida
Fermi Natl. Accelerator Laboratory
United States Naval Academy
Cornell University
University of Colorado Boulder
Carnegie Mellon University
University of California at Santa Barbara
University of California at San Diego
University of California at Riverside
University of California at Los Angeles
University of California at Davis
Brown University
Karamanoglu Mehmetbey University
Bethel University
Boston University
University of Alabama
University of Turin
Catholic University of America
Baylor University
Brunel University London
Monash University
Durham University
Rutherford Appleton Laboratory
University of Southampton
University of Bristol
NASU - Kharkov Institute of Physics and Technology
NASU - Institute for Single Crystals
Yildiz Technical University
Istanbul University
Istanbul University - Cerrahpaşa
Erzincan University
Istanbul Technical University
Hacettepe University
Near East University
Okan University
Bogazici University
Kafkas University
Turkish National Defence University
Marmara University
Middle East Technical University
Bozok University
Cukurova University
Adiyaman University
Izmir Bakircay University
Konya Technical University
Université de Tunis El Manar
Chulalongkorn University
National Taiwan University
National Central University
Indian Institute of Science Bangalore
CNRS/IN2P3-Univ Mediterranee
University of Zurich
Austrian Academy of Sciences
ETH Zurich—Institute for Particle Physics and Astrophysics (IPA)
Paul Scherrer Institute
National and Kapodistrian University of Athens
Brandenburg University of Technology
University of Ruhuna
Saegis Campus
University of Colombo
Eastern University, Sri Lanka
Universidad de Cantabria
University of Oviedo
Universidad Autónoma de Madrid
CIEMAT
University of Belgrade
Laboratório de Instrumentação e Física Experimental de Partículas
Warsaw University of Technology
University of Warsaw
National Centre for Nuclear Research
AGH University of Krakow
Quaid-I-Azam University
University of Canterbury
University of Montenegro
Benemerita Universidad Autonoma de Puebla
Universidad Iberoamericana (UIA)
Centro de Investigacion y de Estudios Avanzados del Instituto Politécnico Nacional
Consejo Nacional de Ciencia y Tecnologia Mexico
Universidad de Sonora
University of Malaya
Universiti Kebangsaan Malaysia
Vilnius University
University of Latvia
Riga Technical University
Kuwait University
American University of the Middle East
Sungkyunkwan University
Yonsei University
University of Seoul
Seoul National University
Sejong University
Kyung Hee University
Korea University
Hanyang University
Institut de Physique des 2 Infinis de Lyon
Chonnam National University
Department of Mathematics and Physics - GWNU
University of Trieste
University of Rome La Sapienza
University of Pisa
Institute of Physics Bhubaneswar
University of Perugia
National Research Council of Italy
University of Pavia
University of Padua
Laboratori Nazionali di Legnaro
Luleå University of Technology
University of Naples Federico II
University of Milan - Bicocca
University of Genoa
National Institute for Nuclear Physics
Guglielmo Marconi University
University of Florence
University of Catania
Centro Siciliano di Fisica Nucleare e di Struttura Della Materia
University of Bologna
Agenzia nazionale per le nuove tecnologie, l'energia e lo sviluppo economico sostenibile
University of Bari
Helwan University
University College Dublin
Institute for Research for Fundamental Sciences
Arak University
University of Science and Technology of Mazandaran
Isfahan University of Technology
Sharif University of Technology
German Electron Synchrotron
Indian Institute of Science Education and Research Pune
University of Hyderabad
Homi Bhabha National Institute
Indian Institute of Technology Bhubaneswar
Tata Institute of Fundamental Research
Indian Institute of Technology Madras
Amity University, Noida
Saha Institute of Nuclear Physics
Visva-Bharati University
University of Delhi
Panjab University
Punjab Agricultural University
Hungarian University of Agriculture and Life Sciences
Wigner Research Centre for Physics
Institute for Nuclear Research
University of Debrecen
Eötvös Loránd University
Assiut University
Babes-Bolyai University
University of Ioannina
National Technical University of Athens
Demokritos National Centre for Scientific Research
Karlsruhe Institute of Technology
University of Hamburg
RWTH Aachen University
Jülich Research Centre
University of Wuppertal
Universidade do Estado do Amazonas
Istinye University
Institut Pluridisciplinaire - Hubert Curien (IPHC)
Université de Haute-Alsace
IN2P3-CNRS
Université Paris-Saclay
Lappeenranta-Lahti University of Technology
Helsinki Institute of Physics
University of Helsinki
National Institute of Chemical Physics and Biophysics, Tallinn
Al-Fayoum University
Academy of Scientific Research and Technology
Suez University
The British University in Egypt
Cairo University
Universidad San Francisco de Quito
Escuela Politécnica Nacional
Charles University
University of Cyprus
Ruder Boskovic Institute
University of Split
Universidad de Antioquia
Universidad de los Andes Colombia
Zhejiang University
CAS - Institute of Modern Physics
Nanjing Normal University
University of Shanghai for Science and Technology
Henan Normal University
University of Science and Technology of China
Sun Yat-Sen University
South China Normal University
Peking University
CAS - Institute of High Energy Physics
University of Chinese Academy of Sciences
China Spallation Neutron Source
China Center of Advanced Science and Technology World Laboratory
Tsinghua University
Beihang University
Universidad de Tarapacá
Sofia University St. Kliment Ohridski
Bulgarian Academy of Sciences
Universidade Estadual Paulista Júlio de Mesquita Filho
Universidade Federal do Rio Grande do Sul
Universidade do Estado do Rio de Janeiro
Universidade Estadual de Campinas
Centro Brasileiro de Pesquisas Físicas
FACAMP - Faculdades de Campinas
Université catholique de Louvain
Ghent University
Université libre de Bruxelles
Vrije Universiteit Brussel
University of Antwerp
TU Wien
Yerevan State University

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

Abstract

We propose a neural network training method capable of accounting for the effects of systematic variations of the data model in the training process and describe its extension towards neural network multiclass classification. The procedure is evaluated on the realistic case of the measurement of Higgs boson production via gluon fusion and vector boson fusion in the ττ decay channel at the CMS experiment. The neural network output functions are used to infer the signal strengths for inclusive production of Higgs bosons as well as for their production via gluon fusion and vector boson fusion. We observe improvements of 12 and 16% in the uncertainty in the signal strengths for gluon and vector-boson fusion, respectively, compared with a conventional neural network training based on cross-entropy.

Original language	English
Article number	1360
Journal	European Physical Journal C
Volume	85
Issue number	11
DOIs	https://doi.org/10.1140/epjc/s10052-025-14713-w
Publication status	Published - Nov 2025
Externally published	Yes

Access to Document

10.1140/epjc/s10052-025-14713-w

Cite this

@article{2972a310ffff467593a1c0a57bdec8f1,

title = "Development of systematic uncertainty-aware neural network trainings for binned-likelihood analyses at the LHC",

abstract = "We propose a neural network training method capable of accounting for the effects of systematic variations of the data model in the training process and describe its extension towards neural network multiclass classification. The procedure is evaluated on the realistic case of the measurement of Higgs boson production via gluon fusion and vector boson fusion in the ττ decay channel at the CMS experiment. The neural network output functions are used to infer the signal strengths for inclusive production of Higgs bosons as well as for their production via gluon fusion and vector boson fusion. We observe improvements of 12 and 16\% in the uncertainty in the signal strengths for gluon and vector-boson fusion, respectively, compared with a conventional neural network training based on cross-entropy.",

author = "CMS Collaboration and D. Druzhkin and V. Borshch and A. Babaev and A. Uzunian and S. Slabospitskii and V. Kachanov and Y. Skovpen and O. Radchenko and A. Kozyrev and T. Dimova and V. Blinov and P. Volkov and V. Savrin and M. Perfilov and V. Klyukhin and A. Gribushin and A. Ershov and L. Dudko and M. Dubinin and V. Bunichev and E. Boos and A. Terkulov and M. Kirakosyan and M. Azarkin and V. Andreev and S. Polikarpov and R. Chistov and M. Chadeeva and A. Zhokin and V. Popov and N. Lychkovskaya and V. Gavrilov and K. Ivanov and T. Aushev and A. Vorobyev and L. Uvarov and V. Sulimov and D. Sosnov and V. Oreshkin and V. Murzin and V. Kim and Y. Ivanov and V. Golovtcov and G. Gavrilov and A. Toropin and I. Tlisova and N. Krasnikov and M. Kirsanov and D. Kirpichnikov and O. Bouhali",

note = "Publisher Copyright: {\textcopyright} CERN for the benefit of the CMS Collaboration 2025.",

year = "2025",

month = nov,

doi = "10.1140/epjc/s10052-025-14713-w",

language = "English",

volume = "85",

journal = "European Physical Journal C",

issn = "1434-6044",

publisher = "Springer Nature",

number = "11",

}

TY - JOUR

T1 - Development of systematic uncertainty-aware neural network trainings for binned-likelihood analyses at the LHC

AU - CMS Collaboration

AU - Druzhkin, D.

AU - Borshch, V.

AU - Babaev, A.

AU - Uzunian, A.

AU - Slabospitskii, S.

AU - Kachanov, V.

AU - Skovpen, Y.

AU - Radchenko, O.

AU - Kozyrev, A.

AU - Dimova, T.

AU - Blinov, V.

AU - Volkov, P.

AU - Savrin, V.

AU - Perfilov, M.

AU - Klyukhin, V.

AU - Gribushin, A.

AU - Ershov, A.

AU - Dudko, L.

AU - Dubinin, M.

AU - Bunichev, V.

AU - Boos, E.

AU - Terkulov, A.

AU - Kirakosyan, M.

AU - Azarkin, M.

AU - Andreev, V.

AU - Polikarpov, S.

AU - Chistov, R.

AU - Chadeeva, M.

AU - Zhokin, A.

AU - Popov, V.

AU - Lychkovskaya, N.

AU - Gavrilov, V.

AU - Ivanov, K.

AU - Aushev, T.

AU - Vorobyev, A.

AU - Uvarov, L.

AU - Sulimov, V.

AU - Sosnov, D.

AU - Oreshkin, V.

AU - Murzin, V.

AU - Kim, V.

AU - Ivanov, Y.

AU - Golovtcov, V.

AU - Gavrilov, G.

AU - Toropin, A.

AU - Tlisova, I.

AU - Krasnikov, N.

AU - Kirsanov, M.

AU - Kirpichnikov, D.

AU - Bouhali, O.

PY - 2025/11

Y1 - 2025/11

N2 - We propose a neural network training method capable of accounting for the effects of systematic variations of the data model in the training process and describe its extension towards neural network multiclass classification. The procedure is evaluated on the realistic case of the measurement of Higgs boson production via gluon fusion and vector boson fusion in the ττ decay channel at the CMS experiment. The neural network output functions are used to infer the signal strengths for inclusive production of Higgs bosons as well as for their production via gluon fusion and vector boson fusion. We observe improvements of 12 and 16% in the uncertainty in the signal strengths for gluon and vector-boson fusion, respectively, compared with a conventional neural network training based on cross-entropy.

AB - We propose a neural network training method capable of accounting for the effects of systematic variations of the data model in the training process and describe its extension towards neural network multiclass classification. The procedure is evaluated on the realistic case of the measurement of Higgs boson production via gluon fusion and vector boson fusion in the ττ decay channel at the CMS experiment. The neural network output functions are used to infer the signal strengths for inclusive production of Higgs bosons as well as for their production via gluon fusion and vector boson fusion. We observe improvements of 12 and 16% in the uncertainty in the signal strengths for gluon and vector-boson fusion, respectively, compared with a conventional neural network training based on cross-entropy.

UR - https://www.scopus.com/pages/publications/105024350483

U2 - 10.1140/epjc/s10052-025-14713-w

DO - 10.1140/epjc/s10052-025-14713-w

M3 - Article

AN - SCOPUS:105024350483

SN - 1434-6044

VL - 85

JO - European Physical Journal C

JF - European Physical Journal C

IS - 11

M1 - 1360

ER -

Development of systematic uncertainty-aware neural network trainings for binned-likelihood analyses at the LHC

Abstract

Access to Document

Other files and links

Fingerprint

Cite this