Pith. sign in

REVIEW 5 major objections 7 minor 115 references

A multi-category classifier on mixed jet samples is bounded by a simplex whose vertices define operational jet flavors and recover their mixing fractions.

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 · grok-4.5

2026-07-31 05:47 UTC pith:KWHKCW6B

load-bearing objection Clean multi-topic generalization of operational jet flavor with a real pipeline; the physics demo is useful but leans on an untested universality assumption in tag-and-probe. the 5 major comments →

arxiv 2607.24921 v1 pith:KWHKCW6B submitted 2026-07-27 hep-ph cs.LGhep-ex

Simplex Demixing: Disentangling Multiple Light-Flavor Jets at Colliders

Gregorio de la Fuente , Jesse Thaler This is my paper
classification hep-ph cs.LGhep-ex
keywords simplex demixingjet flavoroperational definitiontopic modelingCWoLaquark-gluon discriminationhadron-level jetstag-and-probe
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.

Collider physicists have long needed a practical, hadron-level way to define more than two jet flavors without relying on parton-shower labels. This paper extends the earlier operational quark/gluon definition to any number of flavors by showing that a classifier trained on mixed samples lives inside a simplex whose vertices are the purest categories present in the data. From those vertices one can read off the mixing fractions and then reconstruct substructure distributions for each operational flavor. On synthetic three-flavor mixtures the method recovers the known fractions; on dijet events with high-luminosity statistics and perfect particle ID it finds seven operational topics, five of which align strongly with down, up, strange, anti-strange, and gluon jets. The result matters because it turns an ambiguous labeling problem into a geometric demixing problem that can, in principle, be run on real collider data.

Core claim

At the categorical cross-entropy minimum, a classifier trained on M mixtures that are convex combinations of T ≤ M mutually irreducible topics has convex hull equal to a (T−1)-simplex. The T vertices of that simplex determine the mixing fractions (up to permutation) whenever the fraction matrix has full column rank. The paper turns this geometry into a three-stage learning procedure—learn, shape, prune—called simplex demixing, and shows that it recovers multiple light-flavor operational topics from dijet mixtures.

What carries the argument

Simplex demixing: a classifier is parameterized so its outputs live in a learnable simplex inside the probability simplex; an edge-length loss pulls the vertices onto the data cloud and an L1 weight prunes excess vertices, after which the vertices invert to the mixing-fraction matrix.

Load-bearing premise

Each latent flavor must occupy a nonempty pure region in the measured jet features so the simplex vertices are actually reached in finite data; rare flavors and limited particle identification can leave those regions empty.

What would settle it

Train the demixer on the same tag-and-probe dijet mixtures but with particle-ID features removed or with substantially lower statistics; if the seven-vertex geometry collapses or the strong one-to-one match to down/up/strange/anti-strange/gluon disappears, the claimed identifiability fails.

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

If this is right

  • Mixing fractions read from simplex vertices let one reconstruct any jet observable’s distribution for each operational light flavor without parton labels.
  • The same geometric procedure applies to any continuous or set-valued features, not only jets, whenever mixed samples hide mutually irreducible topics.
  • With HL-LHC-scale dijet samples and full hadron information, five light flavors are strongly recoverable at the ensemble level even if single-jet tagging remains hard.
  • Operational topics extracted from different processes can be compared directly, testing how much “quark” or “gluon” depends on the surrounding event.

Where Pith is reading between the lines

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

  • Choosing the number of topics without domain knowledge will likely need stability or lasso-path criteria already common in sparse feature selection; the paper flags this but does not solve it.
  • If π/K separation is weak, strange-related vertices may merge with down-like ones, so the method’s reach is tied to detector PID more tightly than the idealized study shows.
  • The tag-and-probe construction still leans on a supervised Monte-Carlo tagger to build mixtures; a fully unsupervised route to diverse mixtures would remove that last label dependence.

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

5 major / 7 minor

Summary. The manuscript generalizes the operational quark/gluon jet definition from two mixtures to M jet samples and T mutually irreducible topics. It proves that, at the categorical cross-entropy optimum, classifier outputs lie in a (T−1)-simplex whose vertices determine the topic mixing fractions, subject to mutual irreducibility and a full-rank fraction matrix. A three-stage learn/shape/prune network implements the idea. Toy Pythia mixtures recover d/u/g structure and show the expected collapse to a line for two topics. In an HL-LHC-like dijet study, a Pythia-trained seven-flavor tagger and eta binning create 14 tag-and-probe mixtures; after enforcing T=7, five topics align strongly with d, u, s, anti-s, and gluon jets and two align weakly with anti-d and anti-u. Topic-weighted distributions of constituent multiplicity, 2-subjettiness, and jet charge generally reproduce the corresponding Pythia distributions.

Significance. If the universality assumptions are validated, this is a significant extension of data-driven jet-flavor definitions beyond quark/gluon separation and a useful bridge between topic modeling and collider measurements. Notable strengths are an explicit theorem with stated rank and mutual-irreducibility conditions, a practical architecture for continuous point-cloud jets, public code and toy data, bootstrap uncertainty estimates, falsifiable simplex geometry, and an unusually candid treatment of rare antiflavors, gluon contamination, and the idealized detector assumptions. At present, however, the physics study remains a proof of concept because its mixtures and topic count rely on Pythia information and its most important conditional-independence assumption is not tested.

major comments (5)
  1. [§5.2, Eqs. (5.1) and (5.4)] Theorem 1 applies to the physics study only if all 14 mixtures are convex combinations of the same topic distributions. Equation (5.1) assumes that the two jets’ hadron-level features are conditionally independent given their topics, and Eq. (5.4) then drops c(x'), eta, and tau from p(x|t,c(x'),eta,tau). Real dijets retain pT-balance, color, MPI/UE, and shower correlations, and the confidence cut can alter the probe features at fixed topic. If this universality fails, the learned vertices are selection-weighted effective topics and Eq. (5.12) inherits a systematic not visible to the bootstrap. Please test this directly, e.g. compare probe observables at fixed flavor/topic across tags, eta bins, and thresholds, and repeat the demixing under varied selections.
  2. [§5.1–§5.2; abstract and §1] The 14 mixtures are constructed with a supervised PFN trained on seven Pythia-labeled flavors and a 0.8 confidence cut. Thus the demixing stage is label-free, but the overall procedure is not yet a data-only extraction and can steer the learned simplex toward the generator’s own flavor taxonomy. The agreement in Figs. 7–9 is therefore partly a Pythia/tagger closure test rather than independent discovery. The abstract and introduction should narrow the claim to a tag-assisted proof of concept, and the paper should quantify tagger/generator dependence or explain a concrete deployment path that does not use truth-labeled simulation.
  3. [§5.3, stage three; Fig. 6] The L1 strength beta is explicitly chosen to keep exactly seven vertices because seven light flavors are expected. Consequently, Fig. 6 establishes that a stable seven-vertex representation can be learned after imposing T=7; it does not independently show that seven topics emerge from the data. Since the seven-topic conclusion is central, please add domain-agnostic evidence—e.g. validation CCE/Ledge versus T, active-vertex stability across bootstrap runs, and comparisons for T=6 and T=8—or state consistently that the result is conditional on the externally supplied topic count.
  4. [§3.5, Eqs. (3.35)–(3.38)] The equality p(x|t,c)=p_t(x) is asserted because the topics were constructed without the category labels, but absence from training does not imply conditional statistical independence. In general, conditioning on an overlapping truth category c reweights x within topic t. Hence p(t|c) need not be a nonnegative conditional probability even asymptotically; like p(c|t), it is generally a signed linear-overlap coefficient unless an additional screening-off assumption holds. This affects the interpretation of Figs. 4, 7, and 8 and the probability arguments in §5.5. Please state and justify the extra assumption or recast both coefficient matrices as quasi-probability/overlap matrices.
  5. [§4.2, §5.3, and §5.6] The reported 15%–85% intervals use a fixed-hyperparameter bootstrap from one Pythia sample after one supervised tagger and selection. As the text acknowledges, this omits retuning variance; it also omits vertex-number, threshold, eta-bin, tagger, generator, and broken-factorization systematics. Since Figs. 9–10 assess physical agreement against these bands, they should be labeled as internal statistical intervals and supplemented by at least the leading selection/model variations. In particular, bootstrap resampling cannot reveal a bias common to every resample, such as a violation of Eq. (5.1).
minor comments (7)
  1. [§5.2, dataset preparation] Please clarify whether the 70%/20%/10% split is performed by event or by jet. Both jets from one dijet enter the pooled probe sample, so a per-jet split could place correlated objects in training and validation/test sets; block bootstrap resampling alone would not remove that leakage.
  2. [Fig. 6] Only 10 of the 91 possible two-dimensional projections are shown. Please explain the selection criterion and provide either a supplementary full projection grid or a quantitative measure demonstrating that every retained vertex lies on the learned convex hull.
  3. [Table 2] The fractions are rounded to two decimals, while G_mc enters a pseudoinverse. Please state that unrounded fractions were used and clarify whether the quoted fractions were renormalized after the perfect heavy-flavor exclusion.
  4. [§4.2 and §5.3] The toy and physics studies use 20 and 34 bootstrap resamples, respectively. The 15% and 85% quantiles are then based on only a few tail samples; please report convergence of the intervals with resample count.
  5. [§3.4, Eq. (3.27)] The notation t,t' in the edge loss should make clear that the sum is over all M architectural vertices before pruning, even when the eventual physical topic count is smaller.
  6. [§5.4] Please define quantitative criteria for the terms “strongly identifiable” and “weakly identifiable,” rather than relying only on visual inspection of the coefficient matrices.
  7. [Code Availability] The code link is welcome. It would improve reproducibility to archive a versioned release, fixed configuration files, random seeds, and the trained supervised tagger used to construct the mixtures.

Circularity Check

1 steps flagged

Mostly non-circular: Theorem 1 is a self-contained derivation; mild definitional character of operational topics and domain-knowledge choice of T=7 do not force the Pythia-closure results.

specific steps
  1. self definitional [Sec. 2.1 (operational identification); Sec. 5.3–5.4 (T=7 prune)]
    "The key assumption of the operational definition of quark and gluon jets is that these two topics should be identified with the "quark" and "gluon" distributions up to permutation [62]. ... We pick β to keep only T=7 active vertices, using our domain knowledge that there should be seven light flavors in the samples."

    Operational topics are defined as the mutually irreducible simplex vertices (maximally separable categories), then identified with flavor names by assumption; T is set to the expected number of light flavors rather than selected by a data-only criterion. This makes the topic count and the label–topic dictionary partly definitional. It is mild: the measured p(t|c)/p(c|t) alignments and substructure shapes are still empirical and can (and do) fail for rare flavors.

full rationale

Theorem 1 (Sec. 3.2) derives the (T−1)-simplex geometry and recoverability of F_mt from the categorical cross-entropy stationary point plus mutual irreducibility and full column rank; the proof is internal and does not reduce to a fit or to an unverified self-citation. The operational definition intentionally equates topics with maximally separable (mutually irreducible) categories—this is transparent methodology, not a hidden claim that an independent external label was derived. Validation against Pythia uses separate truth labels via pseudoinverse quasi-probabilities and reports partial failure modes (weak d̄/ū, nonzero κ_qg), which is the opposite of a forced closure. Self-citations to the two-mixture CWoLa/jet-topics papers supply the special case being generalized; the multi-mixture theorem and three-stage architecture are new and proved/specified here. The only mild circularity-adjacent choices are (i) fixing T=7 from prior knowledge of seven light flavors when pruning, so the count of topics is not discovered, and (ii) building the 14 mixtures with a Pythia-supervised tagger, which injects composition diversity aligned with those labels—yet the probe-side demixing and substructure inversion remain nontrivial empirical tests. Assumption risks in the tag-and-probe factorization (Eq. 5.1) affect correctness, not circularity of the derivation chain. Score 2 reflects those minor design choices without elevating them to load-bearing circular reduction.

Axiom & Free-Parameter Ledger

6 free parameters · 6 axioms · 1 invented entities

The method rests on the classical mixture model plus mutual irreducibility (anchor regions), full column rank of the fraction matrix, and asymptotic CCE posterior learning. The practical pipeline adds tunable regularizers, a hand-chosen topic count, a supervised tagger and cuts that define the mixtures, and idealized detector/PID assumptions. No new physical entities are postulated; ‘operational topics’ are definitional outputs of the procedure.

free parameters (6)
  • edge-loss weight α = 5e-4 (toy); 5e-5 (physics)
    Tuned so L_edge pulls vertices to the cloud without crushing it; defaults ~5e-4 (toy) and 5e-5 (14-mixture).
  • L1 prune weight β = ≈0.0245 for T=7
    Scanned on a lasso path and chosen to retain exactly T=7 active vertices in most resamples.
  • L2 weight γ = 1e-4 (toy default); 0 (physics)
    Tuned for validation CCE plateau through stage three; set to 0 on the large physics training set.
  • topic count T = 7
    Fixed to 7 from prior knowledge of seven light flavors rather than a data-driven model-selection criterion (Sec. 5.3, 5.7).
  • tagger confidence threshold = 0.8
    Probe jets kept only if supervised softmax ≥0.8 for some flavor; controls purity vs statistics of the 14 mixtures.
  • η bin boundary = 0.75
    Low/high pseudorapidity split at |η|=0.75 to diversify quark/gluon fractions across mixtures.
axioms (6)
  • domain assumption Mixtures are row-stochastic convex combinations of T latent topic distributions (Eq. 3.8).
    Standard topic/mixture model; required for the classifier image to lie in a (T−1)-simplex.
  • domain assumption Topics are mutually irreducible: each has an anchor region where it is positive and all others vanish.
    Used in Theorem 1 to guarantee support at all T vertices; may fail for rare flavors or limited PID.
  • domain assumption Fraction matrix F has full column rank (vertices linearly independent).
    Needed for unique recovery of mixing weights from vertices (Eq. 3.18).
  • standard math At δL_CCE=0 the network outputs posterior mixture probabilities (Eq. 3.14).
    Standard calculus-of-variations result for categorical cross-entropy in the infinite-data limit.
  • ad hoc to paper Tag-and-probe mixtures from a supervised Pythia tagger plus η binning span the same universal operational topics as the dijet ensemble.
    Sec. 5.2 construction; assumes tagger-induced conditioning still yields full-rank F with identifiable anchors.
  • ad hoc to paper Perfect hadron-level particle ID (including π/K/p) and negligible untagged heavy flavor.
    Stated idealization in Sec. 5.1 and 5.7; authors note real ATLAS/CMS π/K separation is limited.
invented entities (1)
  • operational topics (simplex vertices as hadron-level jet flavors) no independent evidence
    purpose: Provide a data-driven, process-tied definition of multiple jet flavors without parton labels.
    Definitional outputs of the demixing procedure (extension of Ref. [62]), not new particles or forces; independent evidence would be stable substructure distributions on real collider data.

pith-pipeline@v1.2.0-grok45-kimik3 · 41167 in / 3846 out tokens · 85119 ms · 2026-07-31T05:47:31.384830+00:00 · methodology

0 comments
read the original abstract

Providing a practical and hadron-level definition of multiple jet flavors has been a long-standing challenge in collider physics. Previous work has introduced a data-driven, operational definition of quark and gluon jets, but no robust generalization beyond two jet categories presently exists. To address this, we introduce a machine-learning framework called "simplex demixing'' to extract $T$ jet flavors (or topics in the statistics literature) from $M$ data samples (or mixtures) with minimal constraints. Intuitively, our procedure identifies the maximally separable categories in the data, translating a multi-category classifier on the $M$ mixtures into a bounded geometric object with $T$ vertices. We first demonstrate our procedure on a toy problem to infer the truth-level fractions of down-quark, up-quark, and gluon jets from synthetic mixtures of the three pure samples. We then propose a tag-and-probe strategy to extract multiple light-flavor categories in a more realistic collider setting involving dijet production. As expected, the identifiability of jet flavors depends on their relative abundance in the samples and the hadron-level information available to the classifier architecture. Our work opens the door to data-driven extractions of multiple jet flavor properties at the Large Hadron Collider.

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

115 extracted references · 1 canonical work pages

  1. [1]

    Nilles and K.H

    H.P. Nilles and K.H. Streng,Quark - Gluon Separation in Three Jet Events,Phys. Rev. D 23(1981) 1944

    1981

  2. [2]

    Jones,Tests for Determining the Parton Ancestor of a Hadron Jet,Phys

    L.M. Jones,Tests for Determining the Parton Ancestor of a Hadron Jet,Phys. Rev. D39 (1989) 2550

    1989

  3. [3]

    Fodor,How to See the Differences Between Quark and Gluon Jets,Phys

    Z. Fodor,How to See the Differences Between Quark and Gluon Jets,Phys. Rev. D41 (1990) 1726

    1990

  4. [4]

    Jones,TOWARDS A SYSTEMATIC JET CLASSIFICATION,Phys

    L. Jones,TOWARDS A SYSTEMATIC JET CLASSIFICATION,Phys. Rev. D42(1990) 811

    1990

  5. [5]

    Lonnblad, C

    L. Lonnblad, C. Peterson and T. Rognvaldsson,Using neural networks to identify jets, Nucl. Phys. B349(1991) 675

    1991

  6. [6]

    Pumplin,How to tell quark jets from gluon jets,Phys

    J. Pumplin,How to tell quark jets from gluon jets,Phys. Rev. D44(1991) 2025

    1991

  7. [7]

    Gallicchio and M.D

    J. Gallicchio and M.D. Schwartz,Quark and Gluon Tagging at the LHC,Phys. Rev. Lett. 107(2011) 172001 [1106.3076]

    Pith/arXiv arXiv 2011

  8. [8]

    Gallicchio and M.D

    J. Gallicchio and M.D. Schwartz,Quark and Gluon Jet Substructure,JHEP04(2013) 090 [1211.7038]

    Pith/arXiv arXiv 2013

  9. [9]

    Bhattacherjee, S

    B. Bhattacherjee, S. Mukhopadhyay, M.M. Nojiri, Y. Sakaki and B.R. Webber,Associated jet and subjet rates in light-quark and gluon jet discrimination,JHEP04(2015) 131 [1501.04794]

    Pith/arXiv arXiv 2015

  10. [10]

    Ferreira de Lima, P

    D. Ferreira de Lima, P. Petrov, D. Soper and M. Spannowsky,Quark-Gluon tagging with Shower Deconstruction: Unearthing dark matter and Higgs couplings,Phys. Rev. D95 (2017) 034001 [1607.06031]

    Pith/arXiv arXiv 2017

  11. [11]

    Bhattacherjee, S

    B. Bhattacherjee, S. Mukhopadhyay, M.M. Nojiri, Y. Sakaki and B.R. Webber,Quark-gluon discrimination in the search for gluino pair production at the LHC,JHEP01(2017) 044 [1609.08781]

    Pith/arXiv arXiv 2017

  12. [12]

    Davighi and P

    J. Davighi and P. Harris,Fractal based observables to probe jet substructure of quarks and gluons,Eur. Phys. J. C78(2018) 334 [1703.00914]

    Pith/arXiv arXiv 2018

  13. [13]

    Larkoski and E.M

    A.J. Larkoski and E.M. Metodiev,A Theory of Quark vs. Gluon Discrimination,JHEP10 (2019) 014 [1906.01639]. [14]CMScollaboration,Search for direct production of supersymmetric partners of the top quark in the all-jets final state in proton-proton collisions at √s= 13TeV,JHEP10(2017) 005 [1707.03316]. – 39 – [15]CMScollaboration,Search for vectorlike light-...

    Pith/arXiv arXiv 2019

  14. [23]

    Salam,Towards Jetography,Eur

    G.P. Salam,Towards Jetography,Eur. Phys. J. C67(2010) 637 [0906.1833]

    Pith/arXiv arXiv 2010

  15. [24]

    Abdesselam et al.,Boosted Objects: A Probe of Beyond the Standard Model Physics, Eur

    A. Abdesselam et al.,Boosted Objects: A Probe of Beyond the Standard Model Physics, Eur. Phys. J. C71(2011) 1661 [1012.5412]

    Pith/arXiv arXiv 2011

  16. [25]

    Plehn and M

    T. Plehn and M. Spannowsky,Top Tagging,J. Phys. G39(2012) 083001 [1112.4441]

    Pith/arXiv arXiv 2012

  17. [26]

    Altheimer et al.,Jet Substructure at the Tevatron and LHC: New Results, New Tools, New Benchmarks,J

    A. Altheimer et al.,Jet Substructure at the Tevatron and LHC: New Results, New Tools, New Benchmarks,J. Phys. G39(2012) 063001 [1201.0008]

    Pith/arXiv arXiv 2012

  18. [27]

    Shelton,Jet Substructure, inTheoretical Advanced Study Institute in Elementary Particle Physics: Searching for New Physics at Small and Large Scales, pp

    J. Shelton,Jet Substructure, inTheoretical Advanced Study Institute in Elementary Particle Physics: Searching for New Physics at Small and Large Scales, pp. 303–340, 2013, DOI [1302.0260]

    Pith/arXiv arXiv 2013

  19. [28]

    Altheimer et al.,Boosted Objects and Jet Substructure at the LHC

    A. Altheimer et al.,Boosted Objects and Jet Substructure at the LHC. Report of BOOST2012, held at IFIC Valencia, 23rd-27th of July 2012,Eur. Phys. J. C74(2014) 2792 [1311.2708]

    Pith/arXiv arXiv 2012

  20. [29]

    Adams et al.,Towards an Understanding of the Correlations in Jet Substructure,Eur

    D. Adams et al.,Towards an Understanding of the Correlations in Jet Substructure,Eur. Phys. J. C75(2015) 409 [1504.00679]

    Pith/arXiv arXiv 2015

  21. [30]

    Cacciari,Phenomenological and theoretical developments in jet physics at the LHC,Int

    M. Cacciari,Phenomenological and theoretical developments in jet physics at the LHC,Int. J. Mod. Phys. A30(2015) 1546001 [1509.02272]

    Pith/arXiv arXiv 2015

  22. [31]

    Kogler et al.,Jet Substructure at the Large Hadron Collider: Experimental Review,Rev

    R. Kogler et al.,Jet Substructure at the Large Hadron Collider: Experimental Review,Rev. Mod. Phys.91(2019) 045003 [1803.06991]

    Pith/arXiv arXiv 2019

  23. [32]

    Marzani, G

    S. Marzani, G. Soyez and M. Spannowsky,Looking inside jets: an introduction to jet substructure and boosted-object phenomenology, vol. 958, Springer (2019), 10.1007/978-3-030-15709-8, [1901.10342]. – 40 –

    Pith/arXiv arXiv 2019

  24. [33]

    Larkoski, I

    A.J. Larkoski, I. Moult and B. Nachman,Jet Substructure at the Large Hadron Collider: A Review of Recent Advances in Theory and Machine Learning,Phys. Rept.841(2020) 1 [1709.04464]

    Pith/arXiv arXiv 2020

  25. [34]

    Larkoski,QCD masterclass lectures on jet physics and machine learning,Eur

    A.J. Larkoski,QCD masterclass lectures on jet physics and machine learning,Eur. Phys. J. C84(2024) 1117 [2407.04897]

    Pith/arXiv arXiv 2024

  26. [35]

    Guest, K

    D. Guest, K. Cranmer and D. Whiteson,Deep Learning and its Application to LHC Physics,Ann. Rev. Nucl. Part. Sci.68(2018) 161 [1806.11484]

    Pith/arXiv arXiv 2018

  27. [36]

    Albertsson et al.,Machine Learning in High Energy Physics Community White Paper,J

    K. Albertsson et al.,Machine Learning in High Energy Physics Community White Paper,J. Phys. Conf. Ser.1085(2018) 022008 [1807.02876]

    Pith/arXiv arXiv 2018

  28. [37]

    Radovic, M

    A. Radovic, M. Williams, D. Rousseau, M. Kagan, D. Bonacorsi, A. Himmel et al.,Machine learning at the energy and intensity frontiers of particle physics,Nature560(2018) 41

    2018

  29. [38]

    Carleo, I

    G. Carleo, I. Cirac, K. Cranmer, L. Daudet, M. Schuld, N. Tishby et al.,Machine learning and the physical sciences,Rev. Mod. Phys.91(2019) 045002 [1903.10563]

    Pith/arXiv arXiv 2019

  30. [39]

    Bourilkov,Machine and Deep Learning Applications in Particle Physics,Int

    D. Bourilkov,Machine and Deep Learning Applications in Particle Physics,Int. J. Mod. Phys. A34(2020) 1930019 [1912.08245]

    Pith/arXiv arXiv 2020

  31. [40]

    Schwartz,Modern Machine Learning and Particle Physics,2103.12226

    M.D. Schwartz,Modern Machine Learning and Particle Physics,2103.12226

    Pith/arXiv arXiv

  32. [41]

    Feickert and B

    M. Feickert and B. Nachman,A Living Review of Machine Learning for Particle Physics, 2102.02770

    Pith/arXiv arXiv

  33. [42]

    Boehnlein et al.,Colloquium: Machine learning in nuclear physics,Rev

    A. Boehnlein et al.,Colloquium: Machine learning in nuclear physics,Rev. Mod. Phys.94 (2022) 031003 [2112.02309]

    Pith/arXiv arXiv 2022

  34. [43]

    Karagiorgi, G

    G. Karagiorgi, G. Kasieczka, S. Kravitz, B. Nachman and D. Shih,Machine learning in the search for new fundamental physics,Nature Rev. Phys.4(2022) 399

    2022

  35. [44]

    Plehn, A

    T. Plehn, A. Butter, B. Dillon, T. Heimel, C. Krause and R. Winterhalder,Modern Machine Learning for LHC Physicists,2211.01421

    Pith/arXiv arXiv

  36. [45]

    Bonilla et al.,Jets and Jet Substructure at Future Colliders,Front

    J. Bonilla et al.,Jets and Jet Substructure at Future Colliders,Front. in Phys.10(2022) 897719 [2203.07462]

    Pith/arXiv arXiv 2022

  37. [46]

    DeZoort, P.W

    G. DeZoort, P.W. Battaglia, C. Biscarat and J.-R. Vlimant,Graph neural networks at the Large Hadron Collider,Nature Rev. Phys.5(2023) 281

    2023

  38. [47]

    K. Zhou, L. Wang, L.-G. Pang and S. Shi,Exploring QCD matter in extreme conditions with Machine Learning,Prog. Part. Nucl. Phys.135(2024) 104084 [2303.15136]

    Pith/arXiv arXiv 2024

  39. [48]

    Belis, P

    V. Belis, P. Odagiu and T.K. Aarrestad,Machine learning for anomaly detection in particle physics,Rev. Phys.12(2024) 100091 [2312.14190]

    Pith/arXiv arXiv 2024

  40. [49]

    Mondal and L

    S. Mondal and L. Mastrolorenzo,Machine learning in high energy physics: a review of heavy-flavor jet tagging at the LHC,Eur. Phys. J. ST233(2024) 2657 [2404.01071]

    Pith/arXiv arXiv 2024

  41. [50]

    P. Gras, S. H¨ oche, D. Kar, A. Larkoski, L. L¨ onnblad, S. Pl¨ atzer et al.,Systematics of quark/gluon tagging,JHEP07(2017) 091 [1704.03878]

    Pith/arXiv arXiv 2017

  42. [51]

    Banfi, G.P

    A. Banfi, G.P. Salam and G. Zanderighi,Infrared safe definition of jet flavor,Eur. Phys. J. C47(2006) 113 [hep-ph/0601139]

    Pith/arXiv arXiv 2006

  43. [52]

    Buckley and C

    A. Buckley and C. Pollard,QCD-aware partonic jet clustering for truth-jet flavour labelling, Eur. Phys. J. C76(2016) 71 [1507.00508]. – 41 –

    Pith/arXiv arXiv 2016

  44. [53]

    Caletti, A.J

    S. Caletti, A.J. Larkoski, S. Marzani and D. Reichelt,Practical jet flavour through NNLO, Eur. Phys. J. C82(2022) 632 [2205.01109]

    Pith/arXiv arXiv 2022

  45. [54]

    Caletti, A.J

    S. Caletti, A.J. Larkoski, S. Marzani and D. Reichelt,A fragmentation approach to jet flavor,JHEP10(2022) 158 [2205.01117]

    Pith/arXiv arXiv 2022

  46. [55]

    Czakon, A

    M. Czakon, A. Mitov and R. Poncelet,Infrared-safe flavoured anti-k T jets,JHEP04(2023) 138 [2205.11879]

    Pith/arXiv arXiv 2023

  47. [56]

    Gauld, A

    R. Gauld, A. Huss and G. Stagnitto,Flavor Identification of Reconstructed Hadronic Jets, Phys. Rev. Lett.130(2023) 161901 [2208.11138]

    Pith/arXiv arXiv 2023

  48. [57]

    Caola, R

    F. Caola, R. Grabarczyk, M.L. Hutt, G.P. Salam, L. Scyboz and J. Thaler,Flavored jets with exact anti-kt kinematics and tests of infrared and collinear safety,Phys. Rev. D108 (2023) 094010 [2306.07314]

    Pith/arXiv arXiv 2023

  49. [58]

    Behring et al.,Flavoured jet algorithms: a comparative study,JHEP09(2025) 149 [2506.13449]

    A. Behring et al.,Flavoured jet algorithms: a comparative study,JHEP09(2025) 149 [2506.13449]

    arXiv 2025

  50. [59]

    Gallicchio and M.D

    J. Gallicchio and M.D. Schwartz,Pure Samples of Quark and Gluon Jets at the LHC, JHEP10(2011) 103 [1104.1175]

    Pith/arXiv arXiv 2011

  51. [60]

    Frye, A.J

    C. Frye, A.J. Larkoski, M.D. Schwartz and K. Yan,Factorization for groomed jet substructure beyond the next-to-leading logarithm,JHEP07(2016) 064 [1603.09338]

    Pith/arXiv arXiv 2016

  52. [61]

    Frye, A.J

    C. Frye, A.J. Larkoski, M.D. Schwartz and K. Yan,Precision physics with pile-up insensitive observables,1603.06375

    Pith/arXiv arXiv

  53. [62]

    Komiske, E.M

    P.T. Komiske, E.M. Metodiev and J. Thaler,An operational definition of quark and gluon jets,JHEP11(2018) 059 [1809.01140]

    Pith/arXiv arXiv 2018

  54. [63]

    Stewart and X

    I.W. Stewart and X. Yao,Pure quark and gluon observables in collinear drop,JHEP09 (2022) 120 [2203.14980]

    Pith/arXiv arXiv 2022

  55. [64]

    Metodiev, B

    E.M. Metodiev, B. Nachman and J. Thaler,Classification without labels: Learning from mixed samples in high energy physics,JHEP10(2017) 174 [1708.02949]

    Pith/arXiv arXiv 2017

  56. [65]

    Metodiev and J

    E.M. Metodiev and J. Thaler,Jet Topics: Disentangling Quarks and Gluons at Colliders, Phys. Rev. Lett.120(2018) 241602 [1802.00008]

    Pith/arXiv arXiv 2018

  57. [66]

    Komiske, S

    P.T. Komiske, S. Kryhin and J. Thaler,Disentangling quarks and gluons in CMS open data, Phys. Rev. D106(2022) 094021 [2205.04459]

    Pith/arXiv arXiv 2022

  58. [67]

    Dolan, J

    M.J. Dolan, J. Gargalionis and A. Ore,Quark-versus-gluon tagging in CMS Open Data with CWoLa and TopicFlow,JHEP08(2025) 024 [2312.03434]. [68]ATLAScollaboration,Properties of jet fragmentation using charged particles measured with the ATLAS detector inppcollisions at √s= 13TeV,Phys. Rev. D100(2019) 052011 [1906.09254]. [69]ATLAS, CMScollaboration,Producti...

    Pith/arXiv arXiv 2025

  59. [72]

    Brewer, J

    J. Brewer, J. Thaler and A.P. Turner,Data-driven quark and gluon jet modification in heavy-ion collisions,Phys. Rev. C103(2021) L021901 [2008.08596]

    Pith/arXiv arXiv 2021

  60. [73]

    Bierlich et al.,A comprehensive guide to the physics and usage of PYTHIA 8.3,SciPost Phys

    C. Bierlich et al.,A comprehensive guide to the physics and usage of PYTHIA 8.3,SciPost Phys. Codeb.2022(2022) 8 [2203.11601]

    Pith/arXiv arXiv 2022

  61. [74]

    Dillon, D.A

    B.M. Dillon, D.A. Faroughy and J.F. Kamenik,Uncovering latent jet substructure,Phys. Rev. D100(2019) 056002 [1904.04200]

    Pith/arXiv arXiv 2019

  62. [75]

    Dillon, D.A

    B.M. Dillon, D.A. Faroughy, J.F. Kamenik and M. Szewc,Learning the latent structure of collider events,JHEP10(2020) 206 [2005.12319]

    Pith/arXiv arXiv 2020

  63. [76]

    Alvarez, M

    E. Alvarez, M. Spannowsky and M. Szewc,Unsupervised Quark/Gluon Jet Tagging With Poissonian Mixture Models,Front. Artif. Intell.5(2022) 852970 [2112.11352]

    Pith/arXiv arXiv 2022

  64. [77]

    LeBlanc, B

    M. LeBlanc, B. Nachman and C. Sauer,Going off topics to demix quark and gluon jets in αS extractions,JHEP02(2023) 150 [2206.10642]

    Pith/arXiv arXiv 2023

  65. [78]

    Blei, A.Y

    D.M. Blei, A.Y. Ng and M.I. Jordan,Latent dirichlet allocation,J. Mach. Learn. Res.3 (2003) 993–1022

    2003

  66. [79]

    Papadimitriou, H

    C.H. Papadimitriou, H. Tamaki, P. Raghavan and S. Vempala,Latent semantic indexing: a probabilistic analysis, inProceedings of the Seventeenth ACM SIGACT-SIGMOD-SIGART Symposium on Principles of Database Systems, PODS ’98, (New York, NY, USA), p. 159–168, Association for Computing Machinery, 1998, DOI

    1998

  67. [80]

    Arora, R

    S. Arora, R. Ge and A. Moitra,Learning topic models – going beyond svd, inProceedings of the 2012 IEEE 53rd Annual Symposium on Foundations of Computer Science, FOCS ’12, (USA), p. 1–10, IEEE Computer Society, 2012, DOI

    2012

  68. [81]

    Arora, R

    S. Arora, R. Ge, Y. Halpern, D. Mimno, A. Moitra, D. Sontag et al.,A practical algorithm for topic modeling with provable guarantees, inProceedings of the 30th International Conference on Machine Learning, S. Dasgupta and D. McAllester, eds., vol. 28 of Proceedings of Machine Learning Research, (Atlanta, Georgia, USA), pp. 280–288, PMLR, 17–19 Jun, 2013, ...

    2013

  69. [82]

    Bioucas-Dias, A

    J.M. Bioucas-Dias, A. Plaza, N. Dobigeon, M. Parente, Q. Du, P. Gader et al.,Hyperspectral unmixing overview: Geometrical, statistical, and sparse regression-based approaches,IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing5(2012) 354

    2012

  70. [83]

    Cutler and L

    A. Cutler and L. Breiman,Archetypal analysis,Technometrics36(1994) 338

    1994

  71. [84]

    Eugster and F

    M.J. Eugster and F. Leisch,From spider-man to hero—archetypal analysis in r,Journal of Statistical Software30(2009) 1

    2009

  72. [85]

    Suleman,Validation of archetypal analysis, in2017 IEEE International Conference on Fuzzy Systems (FUZZ-IEEE), pp

    A. Suleman,Validation of archetypal analysis, in2017 IEEE International Conference on Fuzzy Systems (FUZZ-IEEE), pp. 1–6, IEEE, 2017

    2017

  73. [86]

    Winter,N-FINDR: an algorithm for fast autonomous spectral end-member determination in hyperspectral data, inImaging Spectrometry V, M.R

    M.E. Winter,N-FINDR: an algorithm for fast autonomous spectral end-member determination in hyperspectral data, inImaging Spectrometry V, M.R. Descour and S.S. Shen, eds., vol. 3753, pp. 266 – 275, International Society for Optics and Photonics, SPIE, 1999, DOI

    1999

  74. [87]

    Li and J.M

    J. Li and J.M. Bioucas-Dias,Minimum volume simplex analysis: A fast algorithm to unmix hyperspectral data, inIGARSS 2008 - 2008 IEEE International Geoscience and Remote Sensing Symposium, vol. 3, pp. III – 250–III – 253, 2008, DOI. – 43 –

    2008

  75. [88]

    X. Li, T. Liu, B. Han, G. Niu and M. Sugiyama,Provably end-to-end label-noise learning without anchor points, inProceedings of the 38th International Conference on Machine Learning, M. Meila and T. Zhang, eds., vol. 139 ofProceedings of Machine Learning Research, pp. 6403–6413, PMLR, 2021, https://proceedings.mlr.press/v139/li21l.html

    2021

  76. [89]

    Katz-Samuels, G

    J. Katz-Samuels, G. Blanchard and C. Scott,Decontamination of mutual contamination models,J. Mach. Learn. Res.20(2019) 1521–1577

    2019

  77. [90]

    Bonnet-Guerrini, J

    R. Bonnet-Guerrini, J. Ioannou-Nikolaides, T. Petersen and V. Piuri,Multiclass classification without labels via posterior simplex geometry,to appear

  78. [91]

    Neyman and E.S

    J. Neyman and E.S. Pearson,On the Problem of the Most Efficient Tests of Statistical Hypotheses,Phil. Trans. Roy. Soc. Lond. A231(1933) 289

    1933

  79. [92]

    Scott, G

    C. Scott, G. Blanchard and G. Handy,Classification with asymmetric label noise: Consistency and maximal denoising, inProceedings of the 26th Annual Conference on Learning Theory, S. Shalev-Shwartz and I. Steinwart, eds., vol. 30 ofProceedings of Machine Learning Research, (Princeton, NJ, USA), pp. 489–511, PMLR, 12–14 Jun, 2013, https://proceedings.mlr.pr...

    2013

  80. [93]

    Mair and J

    S. Mair and J. Sj¨ olund,Archetypal analysis++: Rethinking the initialization strategy, Transactions on Machine Learning Research(2024)

    2024

Showing first 80 references.