pith. sign in

arxiv: 2606.18460 · v2 · pith:2EKH7473new · submitted 2026-06-16 · ✦ hep-ex

ParticleTransformer is all you need for reconstructing hadronic tau leptons

Nalong-Norman Seeba , Laurits Tani , Torben Lange , Joosep Pata This is my paper

Pith reviewed 2026-06-29 05:07 UTC · model grok-4.3

classification ✦ hep-ex
keywords hadronic tau reconstructiontau identificationdecay mode classificationmachine learningParticleTransformerFCC-eekinematic regressioncharge reconstruction
0
0 comments X

The pith

Machine learning models using ParticleTransformer reconstruct hadronic tau leptons with per-mille misidentification and percent-level momentum resolution at FCC-ee.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper establishes that a fully machine-learned approach can handle the reconstruction of hadronically decaying tau leptons by breaking the task into identification, decay mode classification, charge reconstruction, and full four-momentum regression. This would matter for the TeraZ program at FCC-ee because the expected large sample of Z to tau tau events demands accurate reconstruction despite undetected neutrinos and varied decay topologies. The authors compare dedicated task-specific models against a single unified multi-task model and report that both reach per-mille-level misidentification rates at high efficiency, F1 scores up to 0.95 on dominant decay channels, and kinematic performance exceeding standard jet observables.

Core claim

The paper claims that ParticleTransformer models trained on fully simulated electron-positron collisions with the CLD detector achieve per-mille-level tau mis-identification at high signal efficiency, decay mode F1 scores up to 0.95, sub-per-mille charge mis-identification that beats a conventional jet-charge estimator by up to two orders of magnitude, and per-mille angular plus percent-level visible transverse momentum resolution that surpasses reconstruction-level jet observables, thereby supplying a complete high-performance solution for hadronic tau reconstruction at FCC-ee.

What carries the argument

The ParticleTransformer architecture that processes sets of reconstructed particles to produce joint predictions across identification, classification, charge, and regression tasks.

If this is right

  • Z to tau tau events can be used for precision Standard Model measurements and beyond-Standard-Model searches with reduced reconstruction systematics.
  • Decay mode classification at F1 scores near 0.95 enables detailed studies of individual tau decay channels.
  • Charge mis-identification below the per-mille level supports high-precision charge asymmetry measurements.
  • Full four-momentum regression at percent-level transverse momentum resolution improves kinematic reconstruction in multi-tau final states.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same transformer approach could be retrained on other collider datasets to address similar reconstruction problems without redesigning hand-crafted algorithms.
  • A single multi-task model might eventually replace multiple specialized tau and jet reconstruction tools, simplifying analysis pipelines.
  • If computational cost remains low, the models could be deployed in online triggering or event selection at future high-luminosity runs.

Load-bearing premise

Performance measured on fully simulated samples with realistic detector effects will translate directly to real experimental data collected at FCC-ee.

What would settle it

Training the models on simulation and then applying them to actual FCC-ee collision data and finding tau mis-identification rates above the per-mille level or momentum resolution no better than jet observables would falsify the performance claims.

Figures

Figures reproduced from arXiv: 2606.18460 by Joosep Pata, Laurits Tani, Nalong-Norman Seeba, Torben Lange.

Figure 1
Figure 1. Figure 1: Schematic of the model architectures of the two training approaches: ( [PITH_FULL_IMAGE:figures/full_fig_p009_1.png] view at source ↗
Figure 1
Figure 1. Figure 1: Illustration of τh reconstruction. Particle flow candidates forming a reconstructed jet are processed by ParticleTransformer-based models through an embedding network and layers of pairwise particle attention. Task-specific output heads produce four reconstruction targets: τh identification, decay mode classification, charge reconstruction, and full four-momentum regression. The resulting τh objects can be… view at source ↗
Figure 2
Figure 2. Figure 2: Evolution of the validation loss during training for the [PITH_FULL_IMAGE:figures/full_fig_p010_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: (Left) The jet mis-identification rate Pmisid as a function of the τh identification efficiency ϵτ is nearly identical for SingleParTau and MultiParTau. (Right) The jet mis-identification rate per generator-level background jet pT bin at a global average τh identification efficiency of 80%. The fake rate is O(10−4 ) in the 0–5 GeV bin, rising to O(10−2 ) above 5 GeV, with both models performing comparably … view at source ↗
Figure 4
Figure 4. Figure 4: (Left) While both MultiParTau and SingleParTau have a similar performance, the unified model has consistently a slightly better performance measured in terms of F1 score, the harmonic mean of precision and recall. (Right) τh class-wise classification performance normalized over generator-level τh decay modes for MultiParTau. In order to evaluate the performance on the kinematic reconstruction task, we use … view at source ↗
Figure 5
Figure 5. Figure 5: The dependence of the reconstructed τh median(∆R) (left) and p vis T (right) of the generator￾level visible τh pT [PITH_FULL_IMAGE:figures/full_fig_p013_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: True and predicted p µ components for the SingleParTau. The charge reconstruction performance is evaluated using the charge mis-identification rate, de￾fined as the fraction of matched generator-level τh candidates for which the predicted charge differs from the true charge. As a reference, we compare the ML models to a conventional jet charge observable [52] denoted as QKappa: Q κ = 1  p jet T κ X i qi … view at source ↗
Figure 7
Figure 7. Figure 7: Charge reconstruction performance for SingleParTau, MultiParTau, and the QKappa base [PITH_FULL_IMAGE:figures/full_fig_p015_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Charge reconstruction performance for SingleParTau, MultiParTau, and the QKappa base [PITH_FULL_IMAGE:figures/full_fig_p016_8.png] view at source ↗
read the original abstract

The large number of $Z \rightarrow \tau\tau$ events expected during the TeraZ program at FCC-ee will allow for precision measurements and searches for physics beyond the Standard Model, requiring accurate reconstruction of hadronically decaying tau leptons. This reconstruction is particularly challenging due to the presence of undetected neutrinos and the diverse topology of hadronic tau decays, making the design of robust heuristic reconstruction algorithms challenging. In this work, we present the first fully machine learned hadronic tau reconstruction approach tuned for FCC-ee studies. The reconstruction is formulated as a set of complementary tasks, including tau identification, decay mode classification, charge reconstruction, and full four-momentum regression. The algorithms are evaluated on fully simulated electron-positron collision samples with realistic detector effects using the CLD detector setup. We compare dedicated task-specific models with a unified multi-task model and quantify their performance in a granular manner across all reconstruction tasks. Both approaches achieve per-mille-level tau mis-identification rates at high signal efficiency, decay mode classification F1 scores of up to 0.95 for the dominant channels, and sub-per-mille charge mis-identification rates, outperforming a conventional jet-charge estimator by up to two orders of magnitude. For the full kinematic reconstruction, the models achieve per-mille-level angular resolution and percent-level visible transverse momentum resolution, exceeding the performance of reconstruction-level jet observables. The resulting models provide a realistic high-performance solution for hadronic tau reconstruction at FCC-ee, offering identification, charge discrimination, decay mode analysis and full kinematic reconstruction.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 2 minor

Summary. The manuscript presents the first fully machine-learned approach to hadronic tau reconstruction at FCC-ee using ParticleTransformer models. It formulates the problem as complementary tasks (identification, decay-mode classification, charge reconstruction, and four-momentum regression) and evaluates dedicated single-task and unified multi-task models on fully simulated e+e- events with the CLD detector. The central claims are per-mille-level mis-identification rates at high efficiency, F1 scores up to 0.95 for dominant decay modes, sub-per-mille charge mis-identification (outperforming a jet-charge baseline by up to two orders of magnitude), and per-mille angular / percent-level pT resolution that exceeds reconstruction-level jet observables.

Significance. If the reported metrics prove robust, the work supplies a practical, high-performance baseline for tau reconstruction in the TeraZ program, where large Z→ττ samples will drive precision measurements and BSM searches. The multi-task formulation and direct comparison to conventional estimators are useful contributions for future collider studies.

major comments (3)
  1. [Abstract / Results] Abstract and results section: performance metrics (mis-ID rates, F1 scores, resolutions) are stated without any accompanying information on training/validation splits, event statistics, loss-function weighting for the multi-task model, or statistical/systematic uncertainties on the quoted figures. This absence prevents assessment of whether the claimed per-mille and sub-per-mille levels are statistically supported.
  2. [Methods / Evaluation] Methods / evaluation: no description is given of how the conventional jet-charge baseline is implemented, how the ParticleTransformer input features are constructed from CLD objects, or whether any simulation-level validation (e.g., comparison of input distributions or control samples) was performed. These details are load-bearing for the claim that the ML models outperform the baseline by up to two orders of magnitude.
  3. [Discussion / Conclusions] The manuscript reports results exclusively on fully simulated samples but provides no discussion of how the models might be validated or calibrated on data, nor any estimate of simulation-to-data discrepancies that could affect the quoted resolutions and mis-identification rates.
minor comments (2)
  1. [Introduction / Methods] Notation for decay-mode labels and kinematic variables should be defined explicitly in a table or early section for clarity.
  2. [Figures] Figure captions should include the exact selection criteria and event counts used for each performance curve.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive feedback. We address each major comment below and will revise the manuscript to incorporate the requested details and clarifications.

read point-by-point responses

  1. Referee: [Abstract / Results] Abstract and results section: performance metrics (mis-ID rates, F1 scores, resolutions) are stated without any accompanying information on training/validation splits, event statistics, loss-function weighting for the multi-task model, or statistical/systematic uncertainties on the quoted figures. This absence prevents assessment of whether the claimed per-mille and sub-per-mille levels are statistically supported.

    Authors: We agree these details are necessary for a complete evaluation. In the revised manuscript we will add explicit information on the training/validation/test splits (including fractions and total event counts), the event statistics of the simulated samples, the loss-function weighting scheme employed for the multi-task model, and statistical uncertainties on all reported metrics. Where relevant, we will also discuss potential systematic uncertainties arising from the simulation. revision: yes

  2. Referee: [Methods / Evaluation] Methods / evaluation: no description is given of how the conventional jet-charge baseline is implemented, how the ParticleTransformer input features are constructed from CLD objects, or whether any simulation-level validation (e.g., comparison of input distributions or control samples) was performed. These details are load-bearing for the claim that the ML models outperform the baseline by up to two orders of magnitude.

    Authors: We will expand the Methods and Evaluation sections to include a precise description of the jet-charge baseline implementation (including the algorithm, track selection, and weighting), a complete list of input features derived from CLD objects, and any simulation-level validation steps performed (such as distribution comparisons between signal and background or control-sample checks). revision: yes

  3. Referee: [Discussion / Conclusions] The manuscript reports results exclusively on fully simulated samples but provides no discussion of how the models might be validated or calibrated on data, nor any estimate of simulation-to-data discrepancies that could affect the quoted resolutions and mis-identification rates.

    Authors: We will add a paragraph in the Discussion section addressing this point. Because FCC-ee has not yet collected data, direct validation on real data is not possible at present. We will discuss likely sources of simulation-to-data discrepancies (e.g., detector modeling, particle identification efficiencies) and outline calibration strategies that could be applied once data are available, such as the use of control samples from Z decays. revision: yes

Circularity Check

0 steps flagged

No significant circularity

full rationale

The paper reports empirical performance of ML models (ParticleTransformer variants) trained and evaluated on fully simulated FCC-ee events. No equations, derivations, or first-principles claims are present; all metrics (mis-ID rates, F1 scores, resolutions) are direct outputs of model inference on simulation samples. No fitted parameters are relabeled as predictions, no self-citation chains support load-bearing premises, and no ansatz or uniqueness theorems are invoked. The work is self-contained within its simulation domain.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on the assumption that ML models trained on simulation will perform similarly on real data, with no free parameters or new entities introduced.

axioms (1)
  • domain assumption Simulated data with realistic detector effects accurately represents real FCC-ee collisions
    Performance is evaluated only on simulation.

pith-pipeline@v0.9.1-grok · 6675 in / 1016 out tokens · 64895 ms · 2026-06-29T05:07:43.972685+00:00 · methodology

Review history (2 revisions) →

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

53 extracted references · 32 canonical work pages · 8 internal anchors

  1. [1]

    Alcaraz Maestre et al.,Tau Reconstruction with Full Simulation of the CLD Detector and prospects for the measurement of the tau polarization at FCC-ee, Mar

    J. Alcaraz Maestre et al.,Tau Reconstruction with Full Simulation of the CLD Detector and prospects for the measurement of the tau polarization at FCC-ee, Mar. 2025. [Online]. Available: https://doi.org/10.17181/69a4s-7vw37

    work page doi:10.17181/69a4s-7vw37 2025

  2. [2]

    A New Mass Reconstruction Technique for Resonances Decaying to di-tau

    A. Elagin et al., “A New Mass Reconstruction Technique for Resonances Decaying to di-tau,”Nucl. Instrum. Meth. A, vol. 654, pp. 481–489, 2011. arXiv:1012.4686 [hep-ex]

    work page internal anchor Pith review Pith/arXiv arXiv 2011

  3. [3]

    Stochastic mass-reconstruction: a new technique to reconstruct resonance masses of heavy particles decaying into tau lepton pairs

    S. Maruyama, “Stochastic mass-reconstruction: a new technique to reconstruct resonance masses of heavy particles decaying into tau lepton pairs,” Dec. 2015. arXiv:1512.04842 [hep-ex]

    work page internal anchor Pith review Pith/arXiv arXiv 2015

  4. [4]

    Evidence for the 125 GeV Higgs boson decaying to a pair of tau leptons

    S. Chatrchyan et al., “Evidence for the 125 GeV Higgs Boson Decaying to a Pair ofτLeptons,” JHEP, vol. 05, p. 104, 2014. arXiv:1401.5041 [hep-ex]

    work page internal anchor Pith review Pith/arXiv arXiv 2014

  5. [5]

    Tau identification algorithms and study of the CP structure of the Yukawa coupling between the Higgs boson and tau leptons in CMS,

    O. Filatov, “Tau identification algorithms and study of the CP structure of the Yukawa coupling between the Higgs boson and tau leptons in CMS,” Ph.D. dissertation, U. Hamburg (main), 2023, Hamburg U., Hamburg, 2023

    2023

  6. [6]

    CP violation angles from H→τ τdecays at FCC-ee,

    S. Giappichini et al., “CP violation angles from H→τ τdecays at FCC-ee,” Feb. 2026. arXiv: 2602.06635 [hep-ph]

    work page arXiv 2026

  7. [7]

    Performance ofτ-lepton reconstruction and identification in CMS,

    CMS Collaboration, “Performance ofτ-lepton reconstruction and identification in CMS,”Journal of Instrumentation, vol. 7, no. 01, P01001–P01001, 2012

    2012

  8. [8]

    Theτchallenges at FCC-ee,

    M. Dam, “Theτchallenges at FCC-ee,”Eur. Phys. J. Plus, vol. 136, no. 9, p. 963, 2021

    2021

  9. [9]

    Tau-lepton Physics at the FCC-ee circular e$^+$e$^-$ Collider

    M. Dam, “Tau-lepton Physics at the FCC-ee circular e +e− Collider,”SciPost Phys. Proc., vol. 1, p. 041, 2019. arXiv:1811.09408 [hep-ex]

    work page internal anchor Pith review Pith/arXiv arXiv 2019

  10. [10]

    Reconstruction and identification ofτlepton decays to hadrons andν τ at CMS,

    CMS Collaboration, “Reconstruction and identification ofτlepton decays to hadrons andν τ at CMS,”Journal of Instrumentation, vol. 11, no. 01, P01019–P01019, 2016

    2016

  11. [11]

    Giappichini et al.,Measurements ofH→τ τcross-section at FCC-ee, Mar

    S. Giappichini et al.,Measurements ofH→τ τcross-section at FCC-ee, Mar. 2025. [Online]. Available:https://doi.org/10.17181/d3y9x-8a819

    work page doi:10.17181/d3y9x-8a819 2025

  12. [12]

    Identification of hadronic tau lepton decays using a deep neural network,

    CMS Collaboration, “Identification of hadronic tau lepton decays using a deep neural network,” JINST, vol. 17, P07023, 2022. arXiv:2201.08458

    work page arXiv 2022

  13. [13]

    Recent advancements in the tau reconstruction and identification techniques in CMS,

    A. Cardini, “Recent advancements in the tau reconstruction and identification techniques in CMS,” in18th International Workshop on Tau Lepton Physics, Dec. 2025. arXiv:2512.07387 [hep-ex]

    work page arXiv 2025

  14. [14]

    ParticleNet: Jet Tagging via Particle Clouds,

    H. Qu and L. Gouskos, “ParticleNet: Jet Tagging via Particle Clouds,”Phys. Rev. D, vol. 101, no. 5, p. 056 019, 2020. arXiv:1902.08570 [hep-ph]

    work page arXiv 2020

  15. [15]

    A unified approach for jet tagging in Run 3 at √s=13.6 TeV in CMS,

    CMS Collaboration, “A unified approach for jet tagging in Run 3 at √s=13.6 TeV in CMS,” 2024. [Online]. Available:https://cds.cern.ch/record/2904702

    work page arXiv 2024

  16. [16]

    Comparison of the performance of tau reconstruction and identification algo- rithms in Run 3,

    CMS Collaboration, “Comparison of the performance of tau reconstruction and identification algo- rithms in Run 3,” 2025. [Online]. Available:https://cds.cern.ch/record/2946445

    work page arXiv 2025

  17. [17]

    Tau lepton identification in displaced topologies using machine learning at CMS,

    CMS Collaboration, “Tau lepton identification in displaced topologies using machine learning at CMS,” 2024. [Online]. Available:https://cds.cern.ch/record/2904366 18

    work page arXiv 2024

  18. [18]

    Performance of boosted tau lepton identification with DeepTau Framework (Boosted DeepTau),

    CMS Collaboration, “Performance of boosted tau lepton identification with DeepTau Framework (Boosted DeepTau),” 2025. [Online]. Available:https://cds.cern.ch/record/2941434

    work page arXiv 2025

  19. [19]

    Low transverse-momentum hadronic tau lepton reconstruction performance in the Run 3 Scouting dataset,

    CMS Collaboration, “Low transverse-momentum hadronic tau lepton reconstruction performance in the Run 3 Scouting dataset,” CERN, Geneva, Tech. Rep., 2024. [Online]. Available:https : //cds.cern.ch/record/2905110

    work page arXiv 2024

  20. [20]

    Reconstruction, Identification, and Calibration of hadronically decaying tau leptons with the ATLAS detector for the LHC Run 3 and reprocessed Run 2 data,

    ATLAS Collaboration, “Reconstruction, Identification, and Calibration of hadronically decaying tau leptons with the ATLAS detector for the LHC Run 3 and reprocessed Run 2 data,” CERN, Geneva, Tech. Rep., 2022. [Online]. Available:https://cds.cern.ch/record/2827111

    work page arXiv 2022

  21. [21]

    TauFinder: A Reconstruction Algorithm forτLeptons at Linear Colliders,

    A. Muennich, “TauFinder: A Reconstruction Algorithm forτLeptons at Linear Colliders,” 2010. [Online]. Available:https://cds.cern.ch/record/1443551

    work page arXiv 2010

  22. [22]

    Tau Lepton Reconstruction at the Muon Collider: Cross Section Measurement of the H→τ+τ−Process,

    M. Gallinaro et al., “Tau Lepton Reconstruction at the Muon Collider: Cross Section Measurement of the H→τ+τ−Process,”HiHEP, vol. 2, no. 1, p. 3, 2026. arXiv:2601.01674 [hep-ex]

    work page arXiv 2026

  23. [23]

    Particle Transformer for jet tagging,

    H. Qu et al., “Particle Transformer for jet tagging,” inInternational conference on machine learning, PMLR, 2022, p. 18 281. arXiv:2202.03772

    work page arXiv 2022

  24. [24]

    Tau lepton identification and reconstruction: A new frontier for jet-tagging ML algorithms,

    T. Lange et al., “Tau lepton identification and reconstruction: A new frontier for jet-tagging ML algorithms,”Comput. Phys. Commun., vol. 298, p. 109 095, 2024. arXiv:2307.07747 [hep-ex]

    work page arXiv 2024

  25. [25]

    A unified machine learning approach for reconstructing hadronically decaying tau leptons,

    L. Tani et al., “A unified machine learning approach for reconstructing hadronically decaying tau leptons,”Computer Physics Communications, vol. 307, p. 109 399, 2025. [Online]. Available:https: //www.sciencedirect.com/science/article/pii/S0010465524003229

    2025

  26. [26]

    Tau kinematics from impact parameters,

    J. K¨ uhn, “Tau kinematics from impact parameters,”Physics Letters B, vol. 313, no. 3-4, pp. 458– 460, 1993

    1993

  27. [27]

    A guide to hunting long-lived particles at the LHC,

    S. Knapen and S. Lowette, “A guide to hunting long-lived particles at the LHC,”Annual Review of Nuclear and Particle Science, vol. 73, no. 1, pp. 421–449, 2023

    2023

  28. [28]

    Collider searches for long-lived particles beyond the Standard Model,

    L. Lee et al., “Collider searches for long-lived particles beyond the Standard Model,”Progress in Particle and Nuclear Physics, vol. 106, pp. 210–255, 2019

    2019

  29. [29]

    Novel signature for long-lived particles at the LHC,

    S. Banerjee et al., “Novel signature for long-lived particles at the LHC,”Physical Review D, vol. 98, no. 11, p. 115 026, 2018

    2018

  30. [30]

    The present and future status of heavy neutral leptons,

    A. M. Abdullahi et al., “The present and future status of heavy neutral leptons,”Journal of Physics G: Nuclear and Particle Physics, vol. 50, no. 2, p. 020 501, 2023

    2023

  31. [31]

    Extending limits on neutral heavy leptons,

    M. Gronau et al., “Extending limits on neutral heavy leptons,”Physical Review D, vol. 29, no. 11, p. 2539, 1984

    1984

  32. [32]

    Tani et al.,Fuτure - dataset for studies, development, and training of algorithms for recon- structing and identifying hadronically decaying tau leptons (Version v3)

    L. Tani et al.,Fuτure - dataset for studies, development, and training of algorithms for recon- structing and identifying hadronically decaying tau leptons (Version v3). Zenodo, 2026. [Online]. Available:https://doi.org/10.5281/zenodo.20717190

    work page doi:10.5281/zenodo.20717190 2026

  33. [33]

    FCC-ee: The lepton collider: Future circular collider conceptual design report volume 2,

    FCC Collaboration et al., “FCC-ee: The lepton collider: Future circular collider conceptual design report volume 2,”Eur. Phys. J. Spec. Top., vol. 228, no. CERN-ACC-2018-0057, pp. 261–623, 2018

    2018

  34. [34]

    FCC physics opportunities: Future circular collider conceptual design report volume 1,

    FCC collaboration, “FCC physics opportunities: Future circular collider conceptual design report volume 1,”European Physical Journal C, vol. 79, no. 6, p. 474, 2019

    2019

  35. [35]

    CLD – A Detector Concept for the FCC-ee,

    N. Bacchetta et al., “CLD – A Detector Concept for the FCC-ee,” Nov. 2019. arXiv:1911.12230 [physics.ins-det]

    work page arXiv 2019

  36. [36]

    Tani et al.,HEP-KBFI/ml-tau-data: arXiv submission (Version v1)

    L. Tani et al.,HEP-KBFI/ml-tau-data: arXiv submission (Version v1). Zenodo, 2026. [Online]. Available:https://doi.org/10.5281/zenodo.20705217 19

    work page doi:10.5281/zenodo.20705217 2026

  37. [37]

    A comprehensive guide to the physics and usage of PYTHIA 8.3,

    C. Bierlich et al., “A comprehensive guide to the physics and usage of PYTHIA 8.3,”SciPost Phys. Codebases, p. 8, 2022. [Online]. Available:https://scipost.org/10.21468/SciPostPhysCodeb.8

    work page doi:10.21468/scipostphyscodeb.8 2022

  38. [38]

    Geant4: a simulation toolkit,

    GEANT4 Collaboration, “Geant4: a simulation toolkit,”Nucl. Instrum. Meth. A, vol. 506, p. 250, 2003

    2003

  39. [39]

    Marlin and LCCD: Software tools for the ILC,

    F. Gaede, “Marlin and LCCD: Software tools for the ILC,”Nucl. Instrum. Meth. A, vol. 559, J. Blumlein et al., Eds., p. 177, 2006

    2006

  40. [40]

    Key4hep, a framework for future HEP experiments and its use in FCC,

    G. Ganis et al., “Key4hep, a framework for future HEP experiments and its use in FCC,”Eur. Phys. J. Plus, vol. 137, p. 149, 2022. arXiv:2111.09874

    work page arXiv 2022

  41. [41]

    Conformal tracking for all-silicon trackers at future electron–positron colliders,

    E. Brondolin et al., “Conformal tracking for all-silicon trackers at future electron–positron colliders,” Nucl. Instrum. Meth. A, vol. 956, p. 163 304, 2020. arXiv:1908.00256 [physics.ins-det]

    work page arXiv 2020

  42. [42]

    EDM4hep - a common event data model for HEP experiments,

    F. Gaede et al., “EDM4hep - a common event data model for HEP experiments,”PoS, vol. ICHEP2022, p. 1237, Nov. 2022

    2022

  43. [43]

    Performance of Particle Flow Calorimetry at CLIC

    J. S. Marshall et al., “Performance of particle flow calorimetry at CLIC,”Nucl. Instrum. Meth. A, vol. 700, p. 153, 2013. arXiv:1209.4039

    work page internal anchor Pith review Pith/arXiv arXiv 2013

  44. [44]

    The Pandora Software Development Kit for Pattern Recognition

    J. S. Marshall and M. A. Thomson, “The Pandora software development kit for pattern recognition,” Eur. Phys. J. C, vol. 75, p. 439, 2015. arXiv:1506.05348

    work page internal anchor Pith review Pith/arXiv arXiv 2015

  45. [45]

    Reichenbach et al.,jpata/CLDConfig: MLPF and tau reconstruction samples, 2026 spring, ver- sion v1.2.5, Jun

    L. Reichenbach et al.,jpata/CLDConfig: MLPF and tau reconstruction samples, 2026 spring, ver- sion v1.2.5, Jun. 2026. [Online]. Available:https://doi.org/10.5281/zenodo.20541598

    work page doi:10.5281/zenodo.20541598 2026

  46. [46]

    Jet reconstruction at high-energy lepton colliders

    M. Boronat et al., “Jet reconstruction at high-energy electron–positron colliders,”Eur. Phys. J. C, vol. 78, p. 144, 2018. arXiv:1607.05039

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  47. [47]

    Track parameters in LCIO,

    T. Kramer, “Track parameters in LCIO,” Linear collider note. [Online]. Available:https://bib- pubdb1.desy.de/record/81214/files/LC-DET-2006-004%5B1%5D.pdf

    2006

  48. [48]

    Decoupled Weight Decay Regularization

    I. Loshchilov and F. Hutter, “Decoupled weight decay regularization,”arXiv preprint arXiv:1711.05101, 2017

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  49. [49]

    Circular interpretation of regression coefficients,

    J. Cremers et al., “Circular interpretation of regression coefficients,”British Journal of Mathemat- ical and Statistical Psychology, vol. 71, no. 1, pp. 75–95, 2018

    2018

  50. [50]

    Robust Estimation of a Location Parameter,

    P. J. Huber, “Robust Estimation of a Location Parameter,”The Annals of Mathematical Statistics, vol. 35, no. 1, pp. 73–101, 1964

    1964

  51. [51]

    Gradient surgery for multi-task learning,

    T. Yu et al., “Gradient surgery for multi-task learning,”Advances in neural information processing systems, vol. 33, pp. 5824–5836, 2020

    2020

  52. [52]

    Measurement of jet charge observables in dijet events at √s= 8 TeV,

    CMS Collaboration, “Measurement of jet charge observables in dijet events at √s= 8 TeV,” 2016

    2016

  53. [53]

    Tani et al.,HEP-KBFI/ml-tau-model: arXiv submission (Version v1)

    L. Tani et al.,HEP-KBFI/ml-tau-model: arXiv submission (Version v1). Zenodo, 2026. [Online]. Available:https://doi.org/10.5281/zenodo.20705279 20

    work page doi:10.5281/zenodo.20705279 2026