pith. machine review for the scientific record. sign in

arxiv: 2604.20965 · v1 · submitted 2026-04-22 · ✦ hep-ph · hep-ex· physics.data-an

Recognition: unknown

Kitchen Sink Anomaly Detection

Alexander M\"uck, David Shih, Gregor Kasieczka, Louis Moureaux, Lukas Lang, Marie Hein, Michael Kr\"amer, Radha Mastandrea, Ranit Das

Authors on Pith no claims yet

Pith reviewed 2026-05-09 23:29 UTC · model grok-4.3

classification ✦ hep-ph hep-exphysics.data-an
keywords anomalydetectionkitchenobservableobservablessignalsinkagnostic
0
0 comments X

The pith

A combined kitchen sink observable set of Energy Flow Polynomials and subjettiness variables outperforms standard baselines in sensitivity to a wide range of resonant signals, with new public benchmarks released and an attribute bagging variant reducing training cost.

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

Particle physicists search for unexpected new signals in collision data by looking for anomalies that do not match known background processes. Earlier approaches either relied on a handful of carefully chosen jet features that might miss certain signals or tried to use the entire event data which is computationally heavy and less sensitive. This work creates several new simulated signal examples made available to the community and tests a large collection of high-level jet measurements including energy flow polynomials and subjettiness. The broad set proves more sensitive across many different possible signal types. They also test a method where each model in an ensemble sees only a random subset of the features, achieving similar performance at lower training cost.

Core claim

We find that our kitchen sink approach is the most sensitive to a broad range of signal types.

Load-bearing premise

That the chosen high-level observables remain sufficiently agnostic and performant when applied to real detector data and backgrounds rather than idealized simulations.

Figures

Figures reproduced from arXiv: 2604.20965 by Alexander M\"uck, David Shih, Gregor Kasieczka, Louis Moureaux, Lukas Lang, Marie Hein, Michael Kr\"amer, Radha Mastandrea, Ranit Das.

Figure 1
Figure 1. Figure 1: FIG. 1. Maximum significance improvement characteristic (SIC), eq. ( [PITH_FULL_IMAGE:figures/full_fig_p007_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: FIG. 2. The same as for Figure [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: FIG. 3. Minimal initial signal significance [PITH_FULL_IMAGE:figures/full_fig_p010_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: FIG. 4. We present the performance of a selection of feature sets in the same setting as Figure [PITH_FULL_IMAGE:figures/full_fig_p012_4.png] view at source ↗
read the original abstract

An enormous amount of R&D effort has resulted in many new resonant anomaly detection methods being proposed in recent years. However, the vast majority of previous R&D studies have suffered from two limitations: they have focused on a very small set of simulated signal benchmark models; and they have either used small sets of carefully crafted high-level jet substructure observables, which can be highly performant but are prone to model dependence, or the full collider event phase space, which is more agnostic but suffers from reduced sensitivity. In this work, we address both limitations: we formulate a number of new simulated signal benchmarks, which we make publicly available in a format fully compatible with the LHCO R&D benchmark; and we explore a high-level, yet highly agnostic, observable set consisting of Energy Flow Polynomials in addition to the usual subjettiness variables. We evaluate this "kitchen sink" observable set for both an idealized anomaly detector and the CWoLa hunting task, along with three baseline observable sets (the Baseline LHC Olympics set, subjettiness observables, and Energy Flow Polynomials). We find that our kitchen sink approach is the most sensitive to a broad range of signal types. Furthermore, we show that an attribute bagging variant, in which each ensemble member is trained on a random subset of substructure observables, yields comparable anomaly detection performance while significantly reducing training cost.

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.

Circularity Check

0 steps flagged

No significant circularity

full rationale

The paper is an empirical comparison study that introduces new public signal benchmarks and evaluates the sensitivity of a combined observable set (Energy Flow Polynomials plus subjettiness) against three baselines in both idealized anomaly detection and CWoLa settings. The central performance claims rest on direct numerical comparisons across these benchmarks rather than any mathematical derivation, fitted parameter, or self-referential definition. No load-bearing step reduces by construction to the paper's own inputs, self-citations, or renamed known results; the public release of the benchmarks supplies independent material for verification.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on the assumption that simulated events adequately proxy real collider data for evaluating anomaly detection sensitivity and that the selected high-level observables capture relevant jet substructure information without introducing excessive model dependence.

axioms (1)
  • domain assumption Simulated signal and background events sufficiently represent the statistical properties of real LHC data for the purpose of comparing anomaly detection methods.
    All evaluations and sensitivity claims are performed on simulated benchmarks.

pith-pipeline@v0.9.0 · 5563 in / 1200 out tokens · 27508 ms · 2026-05-09T23:29:09.346526+00:00 · methodology

discussion (0)

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

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Open LHC Monte Carlo Event Generation

    hep-ph 2026-05 unverdicted novelty 2.0

    A review of initiatives to make LHC Monte Carlo event generations available as open data to minimize redundant simulations and resource use.

Reference graph

Works this paper leans on

67 extracted references · 49 canonical work pages · cited by 1 Pith paper · 5 internal anchors

  1. [1]

    Of course, such an ideal template is not available in real data, but it can be generated for simulated data

    The Idealized Anomaly Detector The idealized anomaly detector (IAD) [26] uses a per- fect background template constructed by sampling from the true background distribution in the SR. Of course, such an ideal template is not available in real data, but it can be generated for simulated data. Therefore, the IAD simply provides an optimal benchmark for weakl...

  2. [2]

    The side- bands in our specific analysis are defined in Section IIID

    CWoLa hunting CWoLa hunting [23–25] is a fully data-driven weakly supervised anomaly detection method that uses the data in the sidebands (SBs) of the SR as the BT. The side- bands in our specific analysis are defined in Section IIID. CWoLa hunting relies on two crucial assumptions. The first is the standard bump hunt assumption that the new 3 physics res...

    2020

  3. [3]

    The first signalX→Y Y ′ →4qwithm Y = 100 GeVandm Y ′ = 500 GeVhas a2 + 2prong topology like the LHCO 2-prong signal. However, unlike in the LHCO signal where all resonantly pro- duced particles are vector bosons, theY ′ particle is a scalar in this model.Y ′ decays into a pair of bottom quarks whileYcan also decay into pairs of light quarks

  4. [4]

    The radion decays into twoWbosons

    The second signalW KK →W R→3Wwith mR = 500 GeVconsists of a heavy Kaluza-Klein vector bosonW KK decaying into aWboson and a scalar radionR[46, 47]. The radion decays into twoWbosons. We analyze the fully hadronic chan- nel, where allWbosons decay into two light quarks each, resulting in a2 + 4prong structure

  5. [5]

    TheT ′ par- ticles decay into a top quark and aZboson

    The third signalZ ′ →T ′T ′ →tZtZwithm T ′ = 400 GeVconsists of aZ ′ vector boson that decays into two vector-like quarksT′ [48, 49]. TheT ′ par- ticles decay into a top quark and aZboson. Again we only consider the fully hadronic channel, where all the intermediate vector bosons decay into light quarks, resulting in a5 + 5prong topology

  6. [6]

    kitchen sink

    The last signalG KK →HH→4twith mH = 400 GeVconsists of a heavy spin-2 Randall- Sundrum gravitonG KK that decays into two Higgs-like scalarsH[50]. The scalars decay into two top quarks, resulting in a6 +6prong structure in the fully hadronic channel which is considered here. We useMadGraph5_amc@nlo 3.6.2[51] at leading order to simulate the hard process, i...

  7. [7]

    Kasieczkaet al., Rept

    G. Kasieczkaet al., Rept. Prog. Phys.84, 124201 (2021), arXiv:2101.08320 [hep-ph]

    work page arXiv 2021

  8. [8]

    Aarrestadet al., SciPost Phys.12, 043 (2021), arXiv:2105.14027 [hep-ph]

    T. Aarrestadet al., SciPost Phys.12, 043 (2021), arXiv:2105.14027 [hep-ph]

    work page arXiv 2021

  9. [9]

    Karagiorgiet al., Nature Reviews Physics4, 399 (2022)

    G. Karagiorgiet al., Nature Reviews Physics4, 399 (2022)

    2022

  10. [10]

    Belis, P

    V. Belis, P. Odagiu, and T. K. Årrestad, Rev.Phys.12, 100091 (2023), arXiv:2312.14190 [physics.data-an]

    work page arXiv 2023

  11. [11]

    A Living Review of Machine Learning for Particle Physics,

    HEP ML Community, “A Living Review of Machine Learning for Particle Physics,”

  12. [12]

    ATLAS Collaboration, Phys. Rev. Lett.125, 131801 (2020), arXiv:2005.02983 [hep-ex]. 12 0 200 400 600 800 1000 Nsig 0 10 20 30 40 50 60 70max(SIC) LHCO 2-prong / IADCombined Random (EFP) EFP7 ( = 1) 2 0.0 0.44 0.87 1.31 1.74 2.18 S/ B 0 200 400 600 800 1000 Nsig 0 10 20 30 40 50 60 70max(SIC) GKK HH 4t / IAD 0.0 0.43 0.86 1.29 1.73 2.16 S/ B FIG. 4. We pre...

    work page arXiv 2020

  13. [13]

    CMS Collaboration, Rept. Prog. Phys.88, 067802 (2024), arXiv:2412.03747 [hep-ex]

    work page arXiv 2024

  14. [14]

    ATLAS Collaboration, Phys. Rev. D112, 072009 (2025), arXiv:2502.09770 [hep-ex]

    work page arXiv 2025

  15. [15]

    Gambhir, R

    R. Gambhiret al., Phys.Rev.Lett.135, 021902 (2025), arXiv:2502.14036 [hep-ph]

    work page arXiv 2025

  16. [16]

    CMS Collaboration, (2025), arXiv:2512.20395 [hep-ex]

    work page arXiv 2025

  17. [17]

    Buhmannet al., Phys

    E. Buhmannet al., Phys. Rev. D109, 055015 (2023), arXiv:2310.06897 [hep-ph]

    work page arXiv 2023

  18. [18]

    Sengupta, M

    D. Senguptaet al., JHEP04, 109 (2023), arXiv:2312.10130 [physics.data-an]

    work page arXiv 2023

  19. [19]

    Mikuni and B

    V. Mikuni and B. Nachman, Phys. Rev. D111, L051504 (2024), arXiv:2404.16091 [hep-ph]

    work page arXiv 2024

  20. [20]

    Thaler and K

    J. Thaler and K. Van Tilburg, JHEP03, 015 (2011), arXiv:1011.2268 [hep-ph]

    work page arXiv 2011

  21. [21]

    Thaler and K

    J. Thaler and K. Van Tilburg, JHEP02, 093 (2012), arXiv:1108.2701 [hep-ph]

    work page arXiv 2012

  22. [22]

    P. T. Komiske, E. M. Metodiev, and J. Thaler, JHEP 04, 013 (2018), arXiv:1712.07124 [hep-ph]

    work page arXiv 2018

  23. [23]

    R&d dataset for lhc olympics 2020 anomaly detection challenge,

    G. Kasieczka, B. Nachman, and D. Shih, “R&d dataset for lhc olympics 2020 anomaly detection challenge,” (2019)

    2020

  24. [24]

    Additional signal models for the lhco2020 r&d,

    R. Daset al., “Additional signal models for the lhco2020 r&d,” (2026)

    2026

  25. [25]

    M. Hein, B. Nachman, and D. Shih, (2025), arXiv:2512.13787 [hep-ph]

    work page arXiv 2025

  26. [26]

    Finkeet al., Phys

    T. Finkeet al., Phys. Rev. D109, 034033 (2023), arXiv:2309.13111 [hep-ph]

    work page arXiv 2023

  27. [27]

    Freytsis, M

    M. Freytsis, M. Perelstein, and Y. C. San, JHEP02, 220 (2023), arXiv:2310.13057 [hep-ph]

    work page arXiv 2023

  28. [28]

    Why do tree-based models still outperform deep learning on tabular data?arXiv preprint arXiv:2207.08815, 2022

    L. Grinsztajn, E. Oyallon, and G. Varoquaux, (2022), arXiv:2207.08815 [cs.LG]

    work page arXiv 2022

  29. [29]

    E. M. Metodiev, B. Nachman, and J. Thaler, JHEP10, 174 (2017), arXiv:1708.02949 [hep-ph]

    work page arXiv 2017

  30. [30]

    J. H. Collins, K. Howe, and B. Nachman, Phys. Rev. Lett.121, 241803 (2018), arXiv:1805.02664 [hep-ph]

    work page arXiv 2018

  31. [31]

    J. H. Collins, K. Howe, and B. Nachman, Phys. Rev. D99, 014038 (2019), arXiv:1902.02634 [hep-ph]

    work page arXiv 2019

  32. [32]

    Hallin, J

    A. Hallinet al., Phys. Rev. D106, 055006 (2021), arXiv:2109.00546 [hep-ph]

    work page arXiv 2021

  33. [33]

    J. A. Raineet al., Front. Big Data6, 899345 (2022), arXiv:2203.09470 [hep-ph]

    work page arXiv 2022

  34. [34]

    Hallinet al., Phys

    A. Hallinet al., Phys. Rev. D107, 114012 (2022), arXiv:2210.14924 [hep-ph]

    work page arXiv 2022

  35. [35]

    Gollinget al., Phys

    T. Gollinget al., Phys. Rev. D107, 096025 (2023), arXiv:2212.11285 [hep-ph]

    work page arXiv 2023

  36. [36]

    Gollinget al., Eur

    T. Gollinget al., Eur. Phys. J. C84, 241 (2023), arXiv:2307.11157 [hep-ph]

    work page arXiv 2023

  37. [37]

    Neyman and E

    J. Neyman and E. S. Pearson, Phil. Trans. Roy. Soc. Lond. A231, 289 (1933)

    1933

  38. [38]

    Nachman and D

    B. Nachman and D. Shih, Phys. Rev. D101, 075042 (2020), arXiv:2001.04990 [hep-ph]

    work page arXiv 2020

  39. [39]

    Andreassen, B

    A. Andreassen, B. Nachman, and D. Shih, Phys. Rev. D 101, 095004 (2020), arXiv:2001.05001 [hep-ph]

    work page arXiv 2020

  40. [40]

    Benkendorfer, L

    K. Benkendorfer, L. L. Pottier, and B. Nachman, Phys. Rev. D104, 035003 (2020), arXiv:2009.02205 [hep-ph]

    work page arXiv 2020

  41. [41]

    Das and D

    R. Das and D. Shih, Phys. Rev. D112, 074040 (2024), arXiv:2410.20537 [hep-ph]

    work page arXiv 2024

  42. [42]

    Leighet al., JHEP12, 105 (2025), arXiv:2407.19818 [hep-ph]

    M. Leighet al., JHEP12, 105 (2025), arXiv:2407.19818 [hep-ph]

    work page arXiv 2025

  43. [43]

    Oleksiyuk, S

    I. Oleksiyuk, S. Voloshynovskiy, and T. Golling, JHEP 07, 177 (2025), arXiv:2503.04342 [hep-ph]

    work page arXiv 2025

  44. [44]

    Pedregosaet al., Journal of Machine Learning Re- search12, 2825 (2011)

    F. Pedregosaet al., Journal of Machine Learning Re- search12, 2825 (2011)

    2011

  45. [45]

    Keet al., inNeural Information Processing Systems (2017)

    G. Keet al., inNeural Information Processing Systems (2017)

    2017

  46. [46]

    Heinet al., (2025), arXiv:2511.14832 [hep-ph]

    M. Heinet al., (2025), arXiv:2511.14832 [hep-ph]

    work page arXiv 2025

  47. [47]

    Asymptotic formulae for likelihood-based tests of new physics

    G. Cowanet al., Eur. Phys. J. C71, 1554 (2011), arXiv:1007.1727 [physics.data-an]

    work page internal anchor Pith review arXiv 2011

  48. [49]

    Bierlichet al., SciPost Phys

    C. Bierlichet al., SciPost Phys. Codebases , 8 (2022)

    2022

  49. [50]

    DELPHES 3, A modular framework for fast simulation of a generic collider experiment

    J. de Favereauet al.(DELPHES 3), JHEP02, 057 (2014), arXiv:1307.6346 [hep-ex]

    work page internal anchor Pith review arXiv 2014

  50. [51]

    Additional qcd background events for lhco2020 r&d (signal region only),

    D. Shih, “Additional qcd background events for lhco2020 r&d (signal region only),” (2021)

    2021

  51. [52]

    Agasheet al., JHEP01, 016 (2017), arXiv:1608.00526 [hep-ph]

    K. Agasheet al., JHEP01, 016 (2017), arXiv:1608.00526 [hep-ph]

    work page arXiv 2017

  52. [53]

    Agasheet al., Phys

    K. Agasheet al., Phys. Rev. D99, 075016 (2019), 13 arXiv:1711.09920 [hep-ph]

    work page arXiv 2019

  53. [54]

    LHC signatures of vector-like quarks

    Y. Okada and L. Panizzi, Adv. High Energy Phys.2013, 364936 (2013), arXiv:1207.5607 [hep-ph]

    work page Pith review arXiv 2013

  54. [55]

    Model Independent Framework for Searches of Top Partners

    M. Buchkremeret al., Nucl. Phys. B876, 376 (2013), arXiv:1305.4172 [hep-ph]

    work page Pith review arXiv 2013

  55. [56]

    Gravity particles from warped extra dimensions, predictions for LHC

    A. Carvalho, (2014), arXiv:1404.0102 [hep-ph]

    work page arXiv 2014

  56. [57]

    The automated computation of tree-level and next-to-leading order differential cross sections, and their matching to parton shower simulations

    J. Alwallet al., JHEP07, 079 (2014), arXiv:1405.0301 [hep-ph]

    work page internal anchor Pith review arXiv 2014

  57. [58]

    The anti-k_t jet clustering algorithm

    M. Cacciari, G. P. Salam, and G. Soyez, JHEP04, 063 (2008), arXiv:0802.1189 [hep-ph]

    work page internal anchor Pith review arXiv 2008

  58. [59]

    FastJet user manual

    M. Cacciari, G. P. Salam, and G. Soyez, Eur. Phys. J. C72, 1896 (2012), arXiv:1111.6097 [hep-ph]

    work page internal anchor Pith review arXiv 2012

  59. [60]

    P. T. Komiske, E. M. Metodiev, and J. Thaler, JHEP 01, 121 (2019), arXiv:1810.05165 [hep-ph]

    work page arXiv 2019

  60. [61]

    P. T. Komiske, E. M. Metodiev, and J. Thaler, Phys. Rev. Lett.123, 041801 (2019), arXiv:1902.02346 [hep- ph]

    work page arXiv 2019

  61. [62]

    P. T. Komiskeet al., Phys. Rev. D101, 034009 (2020), arXiv:1908.08542 [hep-ph]

    work page arXiv 2020

  62. [63]

    P. T. Komiske, E. M. Metodiev, and J. Thaler, Phys. Rev. D101, 036019 (2020), arXiv:1911.04491 [hep-ph]

    work page arXiv 2020

  63. [64]

    Andreassenet al., Phys

    A. Andreassenet al., Phys. Rev. Lett.124, 182001 (2020), arXiv:1911.09107 [hep-ph]

    work page arXiv 2020

  64. [65]

    P. T. Komiske, E. M. Metodiev, and J. Thaler, JHEP 07, 006 (2020), arXiv:2004.04159 [hep-ph]

    work page arXiv 2020

  65. [66]

    T. K. Ho, IEEE Transactions on Pattern Analysis and Machine Intelligence20, 832 (1998)

    1998

  66. [67]

    Bryll, R

    R. Bryll, R. Gutierrez-Osuna, and F. Quek, Pattern Recognition36, 1291 (2003)

    2003

  67. [68]

    Kitchen Sink Anomaly Detection Code,

    L. Lang, “Kitchen Sink Anomaly Detection Code,” (2026)

    2026