Pith. sign in

REVIEW 4 major objections 5 minor 49 references

Combining three particle-level deep jet taggers with event-level boosting improves projected CEPC Higgs precision for H→cc and H→gg by about 42% and 26%, and yields a quantitative ~1.3σ sensitivity benchmark for the unexplored H→ss channel.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · deepseek-v4-flash

2026-08-03 14:05 UTC pith:UOS2L3JA

load-bearing objection Useful fast-sim projection giving the first quantitative H->ss benchmark at CEPC, but the claimed precision gains over the CEPC benchmark are uncontrolled and the abstract disagrees with the body on every headline number. the 4 major comments →

arxiv 2512.21558 v2 pith:UOS2L3JA submitted 2025-12-25 hep-ph hep-ex

Deep-learning jet flavor tagging for precision hadronic Higgs measurements at future e^+e^- Higgs factories

Xinzhu Wang , Yifan Zhu , Chunxiang Zhu , Jianfeng Jiang , Manqi Ruan , Kun Wang , Haijun Yang , Yongfeng Zhu This is my paper
classification hep-ph hep-ex
keywords jet flavor taggingHiggs hadronic decaysstrange Yukawa couplinge+e- Higgs factoryparticle-level deep learningParticleNetXGBoostZ->nu nu channel
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper argues that at a future electron-positron Higgs factory, the most effective way to identify which quarks or gluons a Higgs decayed into is to let three particle-level deep networks—ParticleNet, Particle Transformer, and More-Interaction Particle Transformer—score each jet, then feed those scores together with event-level kinematic variables into an XGBoost classifier. Using the CEPC reference design at 240 GeV and 20 ab^-1, this two-stage recipe projects relative precisions on sigma(ZH) x Br(H->X) of 0.18% for H->bb, 1.07% for H->cc, 0.52% for H->gg, and 78% for H->ss in the Z->nu nu channel. Compared with the published CEPC benchmark, the cc and gg precisions improve by about 42% and 26%. The H->ss channel remains statistically limited, with a significance of about 1.3 sigma—the paper calls this a quantitative sensitivity estimate for a single channel. The motivation is that second-generation Higgs Yukawa couplings, especially the strange Yukawa, are among the least tested pieces of the Standard Model.

Core claim

On the paper's own terms, the central claim is that a two-stage classifier—per-jet deep-learning flavor scores plus event-level kinematics—extracts essentially all the flavor information available in simulated ZH, Z->nu nu events, and does so in a modular way that a holistic event-level network does not. Applied to the CEPC reference detector, the paper derives projected statistical precisions of 0.18% for H->bb, 1.07% for H->cc, 0.52% for H->gg, and 78% for H->ss at 20 ab^-1, with a corresponding ~1.3 sigma sensitivity for H->ss. The gains over the CEPC baseline are largest in the channels where jet flavor identification is hardest: H->cc and H->gg improve by about 42% and 26%, while H->bb

What carries the argument

The load-bearing object is the two-stage classification pipeline. Stage one: three particle-level jet taggers—ParticleNet (dynamic graph convolutions over particle clouds), Particle Transformer (transformer-style attention over particles), and More-Interaction Particle Transformer (a simplified attention variant)—assign each jet one of eleven labels using per-particle kinematics, PID flags, and track impact parameters. Stage two: XGBoost, a gradient-boosted decision tree, takes the per-jet flavor scores from all three taggers plus global event observables (single-jet kinematics, dijet mass, missing energy, jet-MET correlations) and classifies each event into H->bb, H->cc, H->ss, H->gg, 2f ba

Load-bearing premise

The H->ss projection depends on the Monte Carlo generator's production rate of strange hadrons, notably charged kaons: the paper uses Pythia6.4 to match the CEPC reference setup and notes Pythia8.313 would shift the quoted 78% precision by about 40%, so if real strangeness production differs from Pythia6.4, the central H->ss benchmark moves.

What would settle it

Re-run the identical CEPC simulation and classifier chain with Pythia8.313 hadronization instead of Pythia6.4, keeping the same detector card, training split, and analysis code; if the H->ss relative precision does not improve by roughly 40%—or moves in the opposite direction—the quoted 78% / 1.3 sigma result is generator-sensitive and the paper's central H->ss claim does not hold as stated. A simpler companion check is to compare the predicted charged-kaon rate in e+e- -> qq events at 240 GeV against available data or a generator tuned to data.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • The published CEPC benchmark for H->cc improves from 1.85% to 1.07%, and for H->gg from 0.70% to 0.52%, both at 20 ab^-1, making these channels significantly more accessible for precision Yukawa tests.
  • H->ss receives a first quantitative benchmark in the Z->nu nu channel alone: about 1.3 sigma at 20 ab^-1, meaning it is not yet observable but is within reach of combined Z-decay channels, higher luminosity, or updated hadronization models.
  • The same two-stage recipe can be transferred to other proposed e+e- Higgs factories by changing the detector response and luminosity assumptions, so the methodology is not tied to one collider design.
  • Because the jet taggers are reusable and the event-level classifier is interpretable—SHAP and gain analysis show heavy-flavor scores and missing-energy observables dominate—the pipeline can be calibrated mode by mode before systematics are included.
  • The modular structure supports a path to a global fit: once data-driven flavor-tagging calibrations and systematic nuisance parameters are available, they can be inserted without retraining the entire stack.

Where Pith is reading between the lines

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

  • Editorial inference: the H->ss number is the least robust of the four results. The paper itself notes that switching hadronization from Pythia6.4 to Pythia8.313 improves the H->ss precision by about 40% through a higher charged-kaon rate, so the 78% / 1.3 sigma figure should be read as tied to the Pythia6.4 assumption used to match the CEPC reference setup.
  • Editorial note: the abstract quotes slightly different numbers—0.17%, 1.06%, 0.50%, 68%, and about 1.5 sigma—while the body (Section 3.4 and Table 6) reports 0.18%, 1.07%, 0.52%, 78%, and about 1.3 sigma. The body numbers are the derived ones and are the safer anchor for any summary.
  • Editorial extension: the same modular recipe could be applied to Z->ll and Z->qq events without retraining the jet taggers, and the combination of all Z decay channels is a natural next step that should push H->ss beyond the 1.3 sigma single-channel benchmark.
  • Editorial extension: a cheaper and complementary test of the H->ss projection is to compare the predicted K+/- production rate in e+e- -> qq events at 240 GeV against data; this directly tests the generator assumption that drives the strange-jet tagging sensitivity.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The paper presents a two-stage analysis for measuring hadronic Higgs decays at a future e+e- Higgs factory, using CEPC at sqrt(s)=240 GeV and 20 ab^-1 as benchmark. Events e+e- -> ZH with Z->nu nubar and H->bb, cc, ss, gg are analyzed by combining three particle-level deep jet taggers (ParticleNet, Particle Transformer, MIParT) with an event-level XGBoost classifier that also uses global kinematic observables. The authors report projected statistical relative precisions on sigma(ZH)xBr(H->X) of 0.18% (bb), 1.07% (cc), 0.52% (gg), and 78% (ss), corresponding to a 1.3 sigma sensitivity for H->ss, and claim improvements over the published CEPC benchmark of about 42% for H->cc and 26% for H->gg. The paper includes jet-level confusion matrices, event-level classifier performance, SHAP interpretability, a comparison with a holistic event classifier, and a careful discussion of limitations including the absence of systematic uncertainties and the dependence on the hadronization model.

Significance. If the results are established, the paper demonstrates a concrete, modular recipe — deep jet taggers followed by event-level gradient boosting — that improves flavor identification for Higgs decays in the clean e+e- environment, particularly for the difficult cc/g and s/g separations. The work is useful as a benchmark for future Higgs factories and for the H->ss channel, which has received little quantitative study. Strengths include the explicit and reproducible MC setup (Whizard + Pythia6.4 + Delphes3 with the CEPC card), the multi-initialization stability checks, the SHAP-based physics interpretation, and the honest statement of limitations (no systematics, single Z decay channel, generator dependence). The central derivation is internally consistent for the ss channel (Table 5 yields reproduce 78% and 1.3 sigma). However, the headline improvement claims over the CEPC benchmark rest on an uncontrolled comparison, and the abstract disagrees with the body on essentially every numerical result.

major comments (4)
  1. [Sec. 3.4, Table 6] The claimed improvements over the CEPC benchmark (42% for H->cc, 26% for H->gg) are not controlled. The CEPC column is taken from Ref. [37], which likely uses different detector simulation (full simulation vs. Delphes fast simulation), event selection, tagger inputs, and possibly systematic uncertainties. The present analysis is explicitly statistical-only and uses Delphes with the CEPC card, while the text applies the 'different assumptions' caveat only to the FCC-ee comparison. To attribute the improvement to the deep-learning taggers, a same-framework baseline is required: rerun the CEPC reference tagger/selection in the identical Whizard+Pythia+Delphes+XGBoost chain, or at minimum quantify how much of the gain comes from selection, simulation, and statistical treatment.
  2. [Abstract vs. Sec. 3.4 and Table 6] The abstract and the body disagree on every headline number: abstract gives 0.17%/1.06%/0.50%/68% and 43%/29% improvement and 1.5 sigma, while the body and Table 6 give 0.18%/1.07%/0.52%/78%, 42%/26%, and 1.3 sigma. This is not a minor typo in one place; the abstract is the version most readers will rely on. The discrepancy must be resolved before publication.
  3. [Sec. 3.4, Tables 5 and 6] The manuscript does not specify how the quoted relative precisions are derived from the event yields in Table 5. For H->cc, Table 5 gives S=11747 and B=7047, yielding sqrt(S+B)/S = 1.17%; the profile-likelihood significance from Eq. (2.2) gives Z=115.6, i.e., 1/Z=0.87%. Neither equals the quoted 1.07%. Similar but smaller discrepancies exist for H->bb (0.19% vs. 0.18%) and H->gg (0.54% vs. 0.52%); only H->ss matches sqrt(S+B)/S. The statistical procedure (e.g., a likelihood fit with a signal-strength parameter, inclusion of auxiliary constraints, or a different definition of 'relative precision') must be stated explicitly so that the central projections are reproducible.
  4. [Sec. 3.4 and Sec. 4] The H->ss sensitivity is generator-dependent. The paper deliberately uses Pythia6.4 to match the CEPC-TDR setup, and cites Ref. [52] finding that Pythia8.313 improves the H->ss precision by about 40% through a higher K+- rate. This means the quoted 78% precision and 1.3 sigma significance are conditional on the hadronization model, and this caveat is not carried into the abstract or the conclusion. Given the paper's central message includes a quantitative H->ss benchmark, a dedicated cross-check with Pythia8 (even a simplified one) or a strong, prominent caveat in the abstract is needed. The current text acknowledges the issue in Sec. 3.4 but then presents the numbers as a concrete benchmark without this qualification.
minor comments (5)
  1. [Sec. 3.2] Typo: 'pronouced' should be 'pronounced'.
  2. [Appendix B.2] The sentence ending 'Table 8..' has a double period.
  3. [Sec. 2.1 / Table 1] The H->ss cross section is 0.01 fb, giving 200 events at 20 ab^-1; the table shows N_exp=0.2k. This is consistent but is worth a footnote to avoid appearing as an order-of-magnitude typo.
  4. [Appendix C / Table 9] The comparison with the holistic approach is useful, but the generator used for Ref. [52] is not specified in the table caption or text. Since the main text discusses Pythia8-induced 40% improvement for H->ss, clarify which generator is used in each column so the equal 29% values are not misinterpreted.
  5. [Sec. 4] The sentence 'The sensitivity of H->ss measurements can be enhanced via three key avenues...' would benefit from a forward reference to the quantitative Pythia8 comparison in Sec. 3.4, to connect the discussion.

Circularity Check

0 steps flagged

No significant circularity; the sensitivity projections are outputs of an explicit MC/ML pipeline and are not defined in terms of their own inputs.

full rationale

The central quantitative claims (projected precisions and the 42%/26% improvements over the CEPC benchmark) are outputs of the paper's own Monte Carlo pipeline. The pipeline fits no parameter to the claimed target: the jet taggers are trained and evaluated on statistically separated samples (25% reserved for tagger evaluation; event-level 6:2:2 split), the event-level XGBoost combines tagger scores with kinematical observables, and the significance Z is computed from S/B yields using Eq. (2.2). None of these steps defines a target in terms of an input or uses a fitted parameter as its own prediction. The comparison columns in Table 6 quote external benchmark values from Refs. [37,38], so the improvement figures are not self-referential. The H->ss sensitivity is stated as statistically limited (1.3 sigma), and the manuscript itself flags the main limitations: Sec. 3.4 says 'systematic uncertainties... have not been included' and Sec. 4 says the study 'considers only statistical uncertainties and a single Z decay mode.' These are modelling/robustness concerns, not circular reductions. The self-citations (MIParT [33], CEPC measurements [46,48], holistic study [52]) are algorithm descriptions and benchmark comparisons; none is invoked as a uniqueness theorem or unverified premise that forces the result. Appendix C even shows the two-stage result is comparable to, not better than, the holistic benchmark. Therefore no step reduces to its own inputs by construction. The possible uncontrolled comparison with the CEPC reference (different simulation, Delphes, selection) affects the validity of the improvement claim, but that is a correctness/external-validity concern, not circularity.

Axiom & Free-Parameter Ledger

4 free parameters · 5 axioms · 0 invented entities

The central claims rest on simulation and on the ML pipeline's own trained components. No external data are fitted: all 'data' are Monte Carlo. The hand-chosen and data-fitted elements that shape the quoted precisions are the per-channel thresholds, the rectangular preselection, the trained weights of the three taggers and the XGBoost hyperparameters. The principal domain assumptions are the fidelity of the Delphes fast simulation, the adequacy of Pythia6.4 strangeness production for s-tagging (flagged by the authors as ~40%-level generator-dependent), the neglect of H->WW*, and the unbiasedness of the test-set significance estimate.

free parameters (4)
  • Per-channel XGBoost score threshold
    Chosen by maximizing Z (Eq. 2.2) on the same test sample whose yields (Tables 4-5) are then quoted; an optimistic, data-fitted operating point.
  • Preselection cuts (leading-jet pT, pz; mjj; Ez,miss) = 15-100 GeV; +-95 GeV; 110-140 GeV; +-55 GeV
    Hand-chosen windows (Sec. 2.2) that keep ~80% of signal and reject >97% of backgrounds; their choice is not propagated as an uncertainty.
  • PN/ParT/MIParT network weights
    Each tagger is trained on 75% of the signal MC (Sec. 2.2); the flavor scores feeding XGBoost are outputs of supervised fitting to the Monte Carlo, so all downstream numbers inherit the training-sample dependence.
  • XGBoost hyperparameters (depth, learning-rate schedule, subsampling) = depth=3; eta=0.1/0.05/0.01; 1000 trees; subsample=colsample=0.8
    Tuned on the validation split (App. B.1); the five-run spread in Fig. 7 does not include this tuning choice.
axioms (5)
  • domain assumption Whizard1.95 + Pythia6.4 + Delphes3 with the official CEPC card adequately reproduce CEPC detector response (tracking, impact parameters, PID, particle flow) for flavor-tagging projections.
    Invoked in Sec. 2.1; all efficiencies and yields inherit the fast-simulation accuracy.
  • domain assumption Pythia6.4 strangeness production is an acceptable basis for the H->ss sensitivity estimate, despite the known ~40% improvement with Pythia8.313 (K+/- rates).
    Sec. 3.4 and Conclusions: the authors deliberately use Pythia6.4 'to remain aligned with the CEPC-TDR', citing Ref. [52] for the generator dependence.
  • standard math Sensitivity can be estimated by counting S and B events after threshold selection, with the threshold optimized on the same test sample (Eq. 2.2), and quoted precision taken as sqrt(S+B)/S.
    Eq. (2.2) and Sec. 3.4; note the reported Z~1.3 sigma matches S/sqrt(S+B) rather than Eq. (2.2) in the ss channel.
  • domain assumption H->WW* decays are negligible after the lepton veto.
    Sec. 2.1, stated as consistent with Ref. [48].
  • domain assumption Production cross sections and branching ratios from Refs. [46,47] are accurate.
    Used for all normalizations in Table 1.

pith-pipeline@v1.3.0-alltime-deepseek · 22392 in / 22899 out tokens · 220005 ms · 2026-08-03T14:05:07.573578+00:00 · methodology

0 comments
read the original abstract

Precise measurements of Higgs decays into quarks and gluons are essential for probing the Yukawa couplings of the Higgs boson and testing the flavor structure of the Standard Model. We investigate the process $e^+e^- \to ZH$ at $\sqrt{s}=240~\mathrm{GeV}$ at a future $e^+e^-$ Higgs factory, taking the CEPC design as a benchmark. The analysis focuses on events with $Z\to\nu\bar{\nu}$ and hadronic Higgs decays $H\to b\bar{b}$, $c\bar{c}$, $s\bar{s}$ and $gg$. Jet flavor is identified using state-of-the-art particle-level deep neural network taggers (ParticleNet, Particle Transformer and More-Interaction Particle Transformer), whose per-jet outputs are combined with global event observables in a two-stage analysis employing XGBoost classifiers to separate the four Higgs decay modes from the dominant two- and four-fermion Standard Model backgrounds. Assuming an integrated luminosity of $20~\mathrm{ab}^{-1}$, we obtain projected relative precision on $\sigma(ZH)\times\mathrm{Br}(H\to X)$ of 0.17% for $X=b\bar{b}$, 1.06% for $c\bar{c}$, 0.50% for $gg$ and 68% for $s\bar{s}$. Compared with the CEPC published results, the precisions for $H\to c\bar{c}$ and $H\to gg$ are improved by about 43% and 29%, respectively. For $H\to s\bar{s}$ we present a quantitative sensitivity estimation corresponding to a statistical significance of about $1.5\sigma$. These results highlight the potential of deep-learning-based jet flavor tagging for precision studies of Higgs decays at future $e^+e^-$ Higgs factories.

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

49 extracted references · 24 linked inside Pith

  1. [1]

    ATLAS Collaboration,A detailed map of higgs boson interactions by the atlas experiment ten years after the discovery,Nature607(2022) 52

    2022

  2. [2]

    CMS Collaboration,A portrait of the higgs boson by the cms experiment ten years after the discovery,Nature607(2022) 60

    2022

  3. [3]

    Novaes,Standard model: An Introduction, in10th Jorge Andre Swieca Summer School: Particle and Fields, pp

    S.F. Novaes,Standard model: An Introduction, in10th Jorge Andre Swieca Summer School: Particle and Fields, pp. 5–102, 1, 1999 [hep-ph/0001283]

    Pith/arXiv arXiv 1999

  4. [4]

    Higgs,Broken symmetries, massless particles and gauge fields,Phys

    P.W. Higgs,Broken symmetries, massless particles and gauge fields,Phys. Lett.12(1964) 132

    1964

  5. [5]

    Englert and R

    F. Englert and R. Brout,Broken Symmetry and the Mass of Gauge Vector Mesons,Phys. Rev. Lett.13(1964) 321

    1964

  6. [6]

    Guralnik, C.R

    G.S. Guralnik, C.R. Hagen and T.W.B. Kibble,Global Conservation Laws and Massless Particles,Phys. Rev. Lett.13(1964) 585

    1964

  7. [7]

    Kibble,Symmetry breaking in nonAbelian gauge theories,Phys

    T.W.B. Kibble,Symmetry breaking in nonAbelian gauge theories,Phys. Rev.155(1967) 1554

    1967

  8. [8]

    ATLAS Collaboration,Measurements of Higgs boson production and couplings in diboson final states with the ATLAS detector at the LHC,Phys. Lett. B726(2013) 88 [1307.1427]

    Pith/arXiv arXiv 2013

  9. [9]

    ATLAS Collaboration,Observation and measurement of Higgs boson decays to WW ∗ with the ATLAS detector,Phys. Rev. D92(2015) 012006 [1412.2641]

    Pith/arXiv arXiv 2015

  10. [10]

    CMS Collaboration,Measurements of higgs boson properties in the diphoton decay channel in proton-proton collisions at √s= 13tev,Journal of High Energy Physics2018(2018)

    2018

  11. [11]

    CMS Collaboration,Observation of Higgs boson decay to bottom quarks, Tech. Rep. CMS-PAS-HIG-18-016, CERN, Geneva (2018)

    2018

  12. [12]

    ATLAS Collaboration,Observation ofH→b ¯bdecays andV Hproduction with the ATLAS detector,Phys. Lett. B786(2018) 59 [1808.08238]

    Pith/arXiv arXiv 2018

  13. [13]

    CMS Collaboration,Observation of the higgs boson decay to a pair ofτleptons with the cms detector,Physics Letters B779(2018) 283–316

    2018

  14. [14]

    ATLAS Collaboration,Cross-section measurements of the Higgs boson decaying into a pair of τ-leptons in proton-proton collisions at √s= 13TeV with the ATLAS detector,Phys. Rev. D 99(2019) 072001 [1811.08856]

    Pith/arXiv arXiv 2019

  15. [15]

    [16]ATLAScollaboration,Evidence for the Dimuon Decay of the Higgs Boson in pp Collisions with the ATLAS Detector,Phys

    ATLAS Collaboration,Measurements of WH and ZH production with Higgs boson decays into bottom quarks and direct constraints on the charm Yukawa coupling in 13 TeV pp collisions with the ATLAS detector,JHEP04(2025) 075 [2410.19611]. [16]ATLAScollaboration,Evidence for the Dimuon Decay of the Higgs Boson in pp Collisions with the ATLAS Detector,Phys. Rev. Le...

    Pith/arXiv arXiv 2025

  16. [18]

    Skands,Introduction to QCD, inTheoretical Advanced Study Institute in Elementary Particle Physics: Searching for New Physics at Small and Large Scales, pp

    P. Skands,Introduction to QCD, inTheoretical Advanced Study Institute in Elementary Particle Physics: Searching for New Physics at Small and Large Scales, pp. 341–420, 2013, DOI [1207.2389]. – 28 –

    Pith/arXiv arXiv 2013

  17. [19]

    D’Ambrosio, G.F

    G. D’Ambrosio, G.F. Giudice, G. Isidori and A. Strumia,Minimal flavor violation: An Effective field theory approach,Nucl. Phys. B645(2002) 155 [hep-ph/0207036]

    Pith/arXiv arXiv 2002

  18. [20]

    Branco, P.M

    G.C. Branco, P.M. Ferreira, L. Lavoura, M.N. Rebelo, M. Sher and J.P. Silva,Theory and phenomenology of two-Higgs-doublet models,Phys. Rept.516(2012) 1 [1106.0034]

    Pith/arXiv arXiv 2012

  19. [21]

    Redi and A

    M. Redi and A. Weiler,Flavor and CP Invariant Composite Higgs Models,JHEP11(2011) 108 [1106.6357]

    Pith/arXiv arXiv 2011

  20. [22]

    Harnik, J

    R. Harnik, J. Kopp and J. Zupan,Flavor Violating Higgs Decays,JHEP03(2013) 026 [1209.1397]

    Pith/arXiv arXiv 2013

  21. [23]

    CEPC Study Group,CEPC Conceptual Design Report: Volume 1 - Accelerator,1809.00285

    Pith/arXiv arXiv

  22. [24]

    CEPC Study Group,CEPC Conceptual Design Report: Volume 2 - Physics & Detector, 1811.10545

    Pith/arXiv arXiv

  23. [25]

    CEPC Study Group,CEPC Technical Design Report: Accelerator,Radiation Detection Technology and Methods8(2024) 1 [2312.14363]

    arXiv 2024

  24. [26]

    CEPC Study Group,CEPC Technical Design Report - Reference Detector,2510.05260

    arXiv

  25. [27]

    Agapov et al.,Future Circular Lepton Collider FCC-ee: Overview and Status, in Snowmass 2021, 3, 2022 [2203.08310]

    I. Agapov et al.,Future Circular Lepton Collider FCC-ee: Overview and Status, in Snowmass 2021, 3, 2022 [2203.08310]

    Pith/arXiv arXiv 2021

  26. [28]

    BARISH and J.E

    B. BARISH and J.E. BRAU,The international linear collider,International Journal of Modern Physics A28(2013) 1330039. [29]ATLAScollaboration,Transforming jet flavour tagging at ATLAS,2505.19689

    arXiv 2013

  27. [30]

    E. Bols, J. Kieseler, M. Verzetti, M. Stoye and A. Stakia,Jet Flavour Classification Using DeepJet,JINST15(2020) P12012 [2008.10519]

    Pith/arXiv arXiv 2020

  28. [31]

    Qu and L

    H. Qu and L. Gouskos,Jet tagging via particle clouds,Physical Review D101(2020) 056019

    2020

  29. [32]

    H. Qu, C. Li and S. Qian,Particle Transformer for Jet Tagging,2202.03772

    Pith/arXiv arXiv

  30. [33]

    Y. Wu, K. Wang, C. Li, H. Qu and J. Zhu,Jet tagging with more-interaction particle transformer*,Chinese Physics C49(2025) 013110

    2025

  31. [34]

    Roe, H.-J

    B.P. Roe, H.-J. Yang, J. Zhu, Y. Liu, I. Stancu and G. McGregor,Boosted decision trees, an alternative to artificial neural networks,Nucl. Instrum. Meth. A543(2005) 577 [physics/0408124]

    Pith/arXiv arXiv 2005

  32. [35]

    Yang, B.P

    H.-J. Yang, B.P. Roe and J. Zhu,Studies of Boosted Decision Trees for MiniBooNE Particle Identification,Nucl. Instrum. Meth. A555(2005) 370 [physics/0508045]

    Pith/arXiv arXiv 2005

  33. [36]

    Chen and C

    T. Chen and C. Guestrin,Xgboost: A scalable tree boosting system, inProceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, KDD ’16, p. 785–794, ACM, Aug., 2016, DOI

    2016

  34. [37]

    An et al.,Precision higgs physics at the cepc,Chinese Physics C43(2019) 043002

    F. An et al.,Precision higgs physics at the cepc,Chinese Physics C43(2019) 043002

    2019

  35. [38]

    Del Vecchio, J

    A. Del Vecchio, J. Eysermans, L. Gouskos, G. Iakovidis, A. Maloizel, G. Marchiori et al., Precision Measurements of Higgs Hadronic Decay Modes at the FCC-ee,2511.23149

    Pith/arXiv arXiv

  36. [39]

    Yang,Status of the CEPC Project,PoSICHEP2024(2025) 821

    H. Yang,Status of the CEPC Project,PoSICHEP2024(2025) 821

    2025

  37. [40]

    Ai et al.,Flavor physics at the CEPC: a general perspective,Chin

    X. Ai et al.,Flavor physics at the CEPC: a general perspective,Chin. Phys.49(2025) 103003 [2412.19743]. – 29 –

    arXiv 2025

  38. [41]

    Li et al.,Beam test of a baseline vertex detector prototype for cepc,IEEE Transactions on Nuclear Science71(2024) 1319

    S. Li et al.,Beam test of a baseline vertex detector prototype for cepc,IEEE Transactions on Nuclear Science71(2024) 1319

    2024

  39. [42]

    S. Song, J. Chen, J. Liu, Y. Liu, B. Qi, Y. Shi et al.,Study of residual artificial neural network for particle identification in the CEPC high-granularity calorimeter prototype, JINST19(2024) P04033 [2310.09489]

    Pith/arXiv arXiv 2024

  40. [43]

    Kilian, T

    W. Kilian, T. Ohl and J. Reuter,Whizard—simulating multi-particle processes at lhc and ilc, The European Physical Journal C71(2011)

    2011

  41. [44]

    Sj¨ ostrand, S

    T. Sj¨ ostrand, S. Mrenna and P. Skands,Pythia 6.4 physics and manual,Journal of High Energy Physics2006(2006) 026–026. [45]DELPHES 3collaboration,DELPHES 3, A modular framework for fast simulation of a generic collider experiment,JHEP02(2014) 057 [1307.6346]

    Pith/arXiv arXiv 2006

  42. [46]

    Y. Zhu, H. Cui and M. Ruan,The higgs→b b,c c, gg measurement at cepc,Journal of High Energy Physics2022(2022)

    2022

  43. [47]

    Bagnaschi, L

    E. Bagnaschi, L. Biermann and M. Spira,Higgs-Boson Decays: Updates,2510.24658

    arXiv

  44. [48]

    Zhao, Y.-F

    H. Zhao, Y.-F. Zhu, C.-D. Fu, D. Yu and M.-Q. Ruan,The higgs signatures at the cepc cdr baseline,Chinese Physics C43(2019) 023001

    2019

  45. [49]

    Catani, Y

    S. Catani, Y. Dokshitzer, M. Olsson, G. Turnock and B. Webber,New clustering algorithm for multijet cross sections ine +e− annihilation,Physics Letters B269(1991) 432

    1991

  46. [50]

    Lundberg and S.-I

    S. Lundberg and S.-I. Lee,A unified approach to interpreting model predictions,1705.07874

    Pith/arXiv arXiv

  47. [51]

    Shlens,A tutorial on principal component analysis,1404.1100

    J. Shlens,A tutorial on principal component analysis,1404.1100

    Pith/arXiv arXiv

  48. [52]

    Y. Zhu, H. Liang, Y. Wang, Y. Che, H. Wang, C. Zhou et al.,Holistic approach and Advanced Color Singlet Identification for physics measurements at high energy frontier,2506.11783

    Pith/arXiv arXiv

  49. [53]

    Altmann et al.,ECF A Higgs, electroweak, and top Factory Study, vol

    J. Altmann et al.,ECF A Higgs, electroweak, and top Factory Study, vol. 5/2025 ofCERN Yellow Reports: Monographs(6, 2025), 10.23731/CYRM-2025-005, [2506.15390]. – 30 –

    arXiv 2025