The physics behind ML-based quark-gluon taggers, by Sophia Vent, Ramon Winterhalder, Tilman Plehn

SciPost Physics Home Authoring Refereeing Submit a manuscript About The physics behind ML-based quark-gluon taggers Sophia Vent, Ramon Winterhalder, Tilman Plehn SciPost Phys. 20, 084 (2026) · published 16 March 2026 doi: 10.21468/SciPostPhys.20.3.084 pdf BiBTeX RIS Submissions/Reports Abstract Jet taggers provide an ideal testbed for applying explainability techniques to powerful ML tools. For theoretically and experimentally challenging quark-gluon tagging, we first identify the leading latent features that correlate strongly with physics observables, both in a linear and a non-linear approach. Next, we show how Shapley values can assess feature importance, although the standard implementation assumes independent inputs and can lead to distorted attributions in the presence of correlations. Finally, we use symbolic regression to derive compact formulas to approximate the tagger output. × TY - JOURPB - SciPost FoundationDO - 10.21468/SciPostPhys.20.3.084TI - The physics behind ML-based quark-gluon taggersPY - 2026/03/16UR - https://scipost.org/SciPostPhys.20.3.084JF - SciPost PhysicsJA - SciPost Phys.VL - 20IS - 3SP - 084A1 - Vent, SophiaAU - Winterhalder, RamonAU - Plehn, TilmanAB - Jet taggers provide an ideal testbed for applying explainability techniques to powerful ML tools. For theoretically and experimentally challenging quark-gluon tagging, we first identify the leading latent features that correlate strongly with physics observables, both in a linear and a non-linear approach. Next, we show how Shapley values can assess feature importance, although the standard implementation assumes independent inputs and can lead to distorted attributions in the presence of correlations. Finally, we use symbolic regression to derive compact formulas to approximate the tagger output.ER - × @Article{10.21468/SciPostPhys.20.3.084, title={{The physics behind ML-based quark-gluon taggers}}, author={Sophia Vent and Ramon Winterhalder and Tilman Plehn}, journal={SciPost Phys.}, volume={20}, pages={084}, year={2026}, publisher={SciPost}, doi={10.21468/SciPostPhys.20.3.084}, url={https://scipost.org/10.21468/SciPostPhys.20.3.084},} Ontology / Topics See full Ontology or Topics database. Jets Machine learning (ML) Neural networks Quantum chromodynamics (QCD) Authors / Affiliations: mappings to Contributors and Organizations See all Organizations. 1 2 Sophia Vent, 3 4 Ramon Winterhalder, 1 Tilman Plehn 1 Ruprecht-Karls-Universität Heidelberg / Heidelberg University 2 Università di Bologna / University of Bologna [UNIBO] 3 Università degli Studi di Milano / University of Milan [UNIMI] 4 Istituto Nazionale di Fisica Nucleare Sezione di Milano / INFN Sezione di Milano Funders for the research work leading to this publication Baden-Württemberg Stiftung Deutsche Forschungsgemeinschaft / German Research FoundationDeutsche Forschungsgemeinschaft [DFG] Deutscher Akademischer Austauschdienst / German Academic Exchange Service [DAAD]