Pith. sign in

REVIEW 4 major objections 6 minor 43 references

PINCH: Pipeline-Informed Noise Characterization in LIGO's Third Observing Run

T0 review · 4 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read PINCH claims that an SVM trained on clean GstLAL triggers can identify glitch-induced triggers, and that this classification exposes distinct, time-varying fingerprints of glitch classes in O3b.

desk verdict PINCH is a genuinely new pipeline-informed glitch diagnostic with solid qualitative O3b findings, but the unvalidated SVM threshold means the central 'glitch-induced' labeling rests on an assumption the paper never tests. read the letter →

arxiv 2505.14949 v2 pith:JFLCSY6F submitted 2025-05-20 gr-qc astro-ph.IM

Zach Yarbrough , Andre Guimaraes , Prathamesh Joshi , Gabriela González , Andrew Valentini This is my paper
classification gr-qcastro-ph.IM
keywords gravitationalwavestransientnoiseglitchesSupportVectorMachineGstLALmatchedfilteringO3observingrunGravitySpy
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

PINCH claims that a support vector machine trained only on GstLAL triggers from quiet periods can identify which search-pipeline triggers were caused by transient noise rather than random detector fluctuation. The method scores ambiguous triggers that overlap Omicron-detected glitches using their matched-filter SNR and chi-squared values, and classifies triggers with score above one as glitch-induced. Applied to O3b data from the Hanford and Livingston interferometers, it reveals that different Gravity Spy glitch classes leave distinct, sometimes time-varying fingerprints in the pipeline's SNR, chi-squared, and mass parameter spaces. If the claim holds, detector characterization can be done through the pipeline's own response, ranking glitches by actual impact and detecting commissioning changes as shifts in those fingerprints.

What carries the argument

The load-bearing object is a Support Vector Machine—a supervised classifier that finds a boundary between classes—trained on clean GstLAL triggers in the two-dimensional space of matched-filter SNR rho and signal-consistency chi-squared, with one model per detector per O3b chunk. It learns the joint distribution of ordinary noise triggers; applying it to dirty triggers and inverting the score so that values above 1 mark deviations from clean behavior is what defines the glitch-induced population. Two summary statistics carry the analysis: the over-representation ratio, which compares a class's share of glitch-induced triggers with its share of all glitches, and the Bhattacharyya coefficient, which measures overlap between parameter distributions from different chunks.

What would settle it

Inject simulated glitches of known Gravity Spy classes into otherwise quiet O3b strain data and check whether the pipeline triggers they produce receive SVM scores above 1 and fall in the parameter-space regions the paper assigns to those classes.

Watch

Extended reading notes

Core claim

The central claim is that a one-class-style SVM trained per interferometer and per O3b chunk on 'clean' foreground triggers—those whose template duration overlaps no Omicron transient—can separate dirty triggers into glitch-induced (SVM score > 1) and merely coincident populations. Under this labeling, glitch classes show systematic structure in the GstLAL parameter space: Scattered Light triggers cluster at low SNR and high reduced chi-squared, Tomte triggers are stable and concentrated in mid-range SNR at Livingston, and Extremely Loud and Koi Fish dominate the highest SNR triggers at both detectors. The paper further claims that the over-representation ratio and the Bhattacharyya coefficient are the right metrics to quantify how disproportionately each class produces triggers and how consistent its fingerprint is over time, and that the RC tracking commissioning change at Hanford visibly shifted the Scattered Light fingerprint in the rho vs. chi-squared/rho-squared space.

Load-bearing premise

The clean/dirty split assumes Omicron detects every transient and that a trigger overlapping a glitch's template window was caused by that glitch, so any missed glitch or coincidental overlap contaminates both the training set and every reported fingerprint.

Editorial extensions

If this is right

  • Glitch classes can be ranked by per-occurrence impact: Koi Fish and Extremely Loud are over-represented among glitch-induced triggers, while Scattered Light, despite its frequency, is under-represented, so mitigation priority need not follow glitch counts.
  • Commissioning changes are visible from the pipeline side: the RC tracking shift in Scattered Light's rho and chi-squared/rho-squared distribution is a direct example of detector modifications changing noise fingerprints.
  • Stable classes such as Tomte at Livingston produce repeatable trigger parameter distributions, making them candidates for fixed vetoes or gating tailored to their SNR and mass regions.
  • The method is pipeline-agnostic and per-chunk, so the same procedure can be applied to O4 data or other matched-filter pipelines without retraining across runs.
  • Blip over-representation in the BBH search region and Fast Scattering over-representation in the BNS region identify where sensitivity loss is concentrated, pointing to targeted mitigation.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Going beyond the paper, the same per-chunk SVM design could be run in near-real time as data streams, turning the score-above-1 population into a live monitor of how each glitch class is coupling to the search.
  • The paper notes that visually distinct classes can overlap in pipeline space; a further step would be to cluster triggers by their SVM-measured pipeline response and ask whether those clusters correspond to unknown common physical sources.
  • A testable extension would be to apply PINCH to O4 and see whether the post-RC-tracking Scattered Light fingerprint persists, indicating whether that commissioning change removed a subpopulation or permanently changed the class's pipeline behavior.
  • Because the method is pipeline-agnostic, running the same clean-trained SVM on another matched-filter search pipeline would show which fingerprints are intrinsic to glitch physics and which depend on GstLAL's template bank.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 6 minor

Summary. The paper introduces PINCH, a method to classify GstLAL foreground triggers as glitch-induced or noise-like using Support Vector Machine (SVM) classifiers trained on triggers that do not overlap with Omicron-detected transients. For each O3b chunk and interferometer, an SVM is trained on the ρ and χ2 values of 'clean' triggers, and 'dirty' triggers (those overlapping Omicron glitches) are scored; triggers with SVM score above 1 are labeled as caused by glitches. The resulting glitch-induced triggers are then analyzed by Gravity Spy glitch class, in mass–mass space, ρ–χ2/ρ2 space, SNR distributions, over-representation ratios, and Bhattacharyya coefficients across chunks. The authors report class-specific fingerprints, time-varying behavior such as the Scattered Light shift after RC tracking at LHO, and discuss implications for detector characterization and pipeline mitigation.

Significance. If the classification step is valid, PINCH would provide a useful, pipeline-informed complement to existing glitch characterization tools, with the practical benefit of using production-pipeline triggers rather than synthetic glitch models. The paper is clearly written, the data handling is transparent, and the per-chunk, per-interferometer analysis is a strength. The method is also presented as pipeline-agnostic, which broadens its potential utility. However, the significance is conditional: the central classifier threshold is not validated against any ground truth, and the clean/dirty labeling rests on an untested completeness assumption about Omicron. The paper's main findings would be convincing only after these issues are addressed.

major comments (4)
  1. [§3.1.3, Fig. 5] The SVM score threshold of 1 is adopted by convention ('it is common to use a value of ±1') with no independent validation. Since the true outlier fraction in the evaluation set is unknown, as acknowledged in §3.1.2, there is no evidence that score > 1 separates glitch-induced triggers from triggers that merely coincide with a glitch. This threshold is load-bearing for every downstream result, including Tables 1–2 and Figures 7–13. The paper should provide a calibration study, for example using hardware injections, simulated glitches, or an independent set of human/veto-confirmed glitch-induced triggers, and report accuracy, completeness, and stability of the classification.
  2. [§3.1.1] The clean/dirty label is defined by temporal overlap with Omicron triggers, which assumes Omicron detects all relevant transients and that overlap implies causation. If Omicron misses low-SNR or non-stationary glitches, the clean training set is contaminated; if a pipeline trigger overlaps a glitch for unrelated reasons, the dirty set is mislabeled. This assumption propagates through all subsequent analysis. The authors should quantify Omicron's sensitivity in the relevant SNR and frequency range, and ideally compare with an independent transient finder or injection study to bound the contamination rate.
  3. [§4.2, Fig. 10] The ρ vs χ2/ρ2 distribution is the same feature space (up to normalization) used to train the SVM, and the glitch-induced triggers are selected precisely because they are outliers in that space. Therefore, the finding that different glitch classes occupy distinct regions of ρ vs χ2/ρ2 is partly a restatement of the classifier's input rather than an independent measurement. The class-conditional distributions are also selection-biased: glitch types that produce triggers resembling clean triggers in this space are excluded by construction. The paper should acknowledge this circularity explicitly and, if possible, present fingerprints in parameter spaces not used for classification, such as template masses or other GstLAL output quantities.
  4. [§4, Tables 1–2, Fig. 13] The results are presented without any uncertainty quantification. The percentages in Tables 1–2, the over-representation ratios in Figure 11, and the Bhattacharyya coefficients in Figure 13 have no error bars or statistical significance tests, despite the fact that the classification threshold and the Omicron-based labeling both introduce substantial systematic uncertainty. The paper should include a sensitivity analysis varying ν and the score threshold, and report bootstrap or other uncertainties on the key metrics.
minor comments (6)
  1. [§1 and §4] The introduction says O3b corresponds to chunks 24 through 40, while Section 4 says O3b consisted of chunks 23 through 40; Tables 1–2 begin at chunk 23. This inconsistency should be corrected.
  2. [§3.2.1] The heading 'Battacharyya Coefficient' is misspelled; it should be 'Bhattacharyya Coefficient'. Also, in the text following Eq. (1), 'varible' should be 'variable'.
  3. [§3.1.2] The description of the 10,000 training samples is incomplete: it should state how these samples are selected from the clean population (random subsample? balanced across the chunk? with or without replacement?) and whether the same samples are used across the ν sweep.
  4. [§3.1.3] The SVM formulation is described loosely: a binary SVM normally has two classes, but here the model is trained only on clean triggers. The paper should clarify explicitly that a one-class SVM is used (or justify the two-class interpretation), and define the decision function whose sign is inverted.
  5. [§4.1.1] The text says 'Figure 9 amd 7' where 'amd' should be 'and', and the sentence is grammatically incomplete. Also, Figure 7 caption describes top and bottom panels but the text references 'top panels' and 'bottom panels' in a way that is not fully consistent.
  6. [§4.3] The over-representation ratio formula in Eq. (2) is written as a product rather than a ratio of ratios; please rewrite it with explicit parentheses or a fraction to avoid ambiguity.

Circularity Check

1 steps flagged · score 4.0 of 10

Partial circularity: the glitch-induced trigger set is defined by SVM score in rho/chi2 space and then characterized in that same space, though class-specific fingerprints retain independent content.

  1. self definitional [Section 3.1.3 (Application of SVM) -> Section 4.2 (rho vs chi2/rho2; Figure 10 discussion)]
    "Thus, the dirty triggers that are evaluated by the SVM and return a score greater than 1 are considered to be caused by glitches. ... Figure 10, analogous to the previous mass-space comparison, shows that different glitch types occupy distinct regions of this parameter space."

    The SVM is trained on the rho and chi2 values of clean triggers and assigns high scores to dirty triggers that are unlike the clean distribution; score >1 is the operational definition of 'caused by glitches.' Section 4.2 then reports the distribution of exactly these triggers in the rho vs chi2/rho2 space as an empirical finding. That the selected population is non-clean-like in this space is true by construction, so claims that glitch-induced triggers occupy 'characteristic regions' of this space partly restate the classifier's decision boundary. The per-class differences (e.g., Scattered Light at low rho and high chi2) are not forced by the SVM because class labels enter only after selection, making the circularity partial rather than total.

full rationale

The central derivation is not a self-citation chain: references to prior GstLAL and detector-characterization work are background or corroborating, and the RC tracking citation is supported by the paper's own before/after data. The clean/dirty labeling depends on Omicron overlap, and the choice of SVM score > 1 as the glitch boundary is not validated against independent ground truth; these are correctness risks, not circularity. The main circular element is that the 'glitch-induced' population is defined by being anomalous in rho/chi2 space, and Section 4.2 then characterizes that population in the same space, partly restating the selection criterion. However, the mass-space distributions, SNR histograms, over-representation ratios, and Bhattacharyya-coefficient time evolution all use variables or aggregations not encoded in the SVM decision boundary, so the class-specific fingerprints and temporal findings have independent empirical content. A moderate score of 4 reflects this partial self-definitional character without treating the whole analysis as forced.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

No new physical entities are introduced; the over-representation ratio is a descriptive statistic, not an entity. The free parameters and assumptions listed are the main input choices that shape all reported results.

free parameters (3)
  • SVM nu hyperparameter = 0.01
    Chosen from a sweep over nu values on evaluation data to be conservative; directly controls the outlier fraction at large values and shapes the decision boundary.
  • SVM score threshold = 1
    Triggers with score greater than 1 are classified as glitch-induced; this is the standard plus-or-minus-1 SVM margin convention, but the threshold determines the counts used in every downstream metric.
  • Gravity Spy confidence threshold = 90%
    Only glitches with at least 90% Gravity Spy classification confidence are associated with pipeline triggers; this filters which triggers receive a glitch class label and affects all class-specific comparisons.
assumptions (4)
  • domain assumption Clean triggers, defined as those with no Omicron overlap, represent nominal detector behavior and are unaffected by glitches.
    Section 3.1.1, training data definition. If Omicron misses transients or clean times contain weak glitches, the SVM learns a contaminated clean distribution.
  • domain assumption Time overlap between a pipeline trigger's template duration and an Omicron glitch implies the glitch caused the trigger.
    Section 3.1.1, dirty labeling. This causal attribution is assumed from coincidence; no injection or veto study validates it.
  • domain assumption A one-class SVM in rho and chi2 space can separate glitch-induced triggers from random noise triggers despite having no labeled outliers.
    Section 3.1.2. The paper acknowledges that standard validation is impossible because the outlier fraction is unknown; the method's core assumption is that clean and dirty populations are separable in this 2D feature space.
  • domain assumption Gravity Spy classifications with at least 90% confidence are accurate labels for glitch morphologies.
    Sections 2.1 and 3.1.1. The analysis trusts Gravity Spy's classifier outputs to associate triggers with glitch classes.

how reviews work

0 comments
Cite this review

Pith. "Pith review of PINCH: Pipeline-Informed Noise Characterization in LIGO's Third Observing Run." pith.science (2026). https://pith.science/paper/JFLCSY6F

@misc{pith2026250514949,
  author       = {Pith},
  title        = {Pith review of: PINCH: Pipeline-Informed Noise Characterization in LIGO's Third Observing Run},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/JFLCSY6F}},
  note         = {Machine review of arXiv:2505.14949}
}
read the original abstract

We present a method to identify and categorize gravitational wave candidate triggers identified by matched filtering gravitational wave searches (pipelines) caused by transient noise (glitches) in gravitational wave detectors using Support Vector Machine (SVM) classifiers. Our approach involves training SVM models on pipeline triggers which occur outside periods of excess noise to distinguish between triggers caused by random noise and those induced by glitches. This method is applied independently to the triggers produced by the GstLAL search pipeline on data from the LIGO Hanford and Livingston observatories during the second half of the O3 observing run. The trained SVM models assign scores to ambiguous triggers, quantifying their similarity to triggers caused by random fluctuations, with triggers with scores above a defined threshold being classified as glitch-induced. Analysis of these triggers reveals the distinct impact of different glitch classes on the search pipeline, including their distribution in relevant parameter spaces. We use metrics such as the Bhattacharyya coefficient and an over-representation ratio to quantify the consistency and prevalence of glitch impacts over time and across parameter spaces. Our findings indicate that some glitch types consistently produce triggers in specific regions of the parameter space, while others generate triggers that are more widely distributed. We observe that Scattered Light glitches appear differently in the search pipeline before and after a commissioning change, demonstrating how such detector changes appear in the pipeline's response to certain glitch classes. This method provides a framework for understanding and mitigating the influence of non-Gaussian transients on gravitational wave search pipelines, with implications for improving detection sensitivity and better understanding noise populations.

Figures

Figures reproduced from arXiv: 2505.14949 by the authors.

Figure 1
Figure 1. Spectrogram of several relevant glitch classes. [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Possible cases for trigger template overlaps with a glitch of a given duration. [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. ρ vs χ 2/ρ 2 histograms for pipeline triggers at LLO in chunk 30 that do not overlap with transient noise and are labeled “clean” (left) and those that do overlap with transient noise and have been labeled “dirty” (right). Clean triggers are used as training data for the SVM, which is then used to evaluate the triggers in the dirty population. filtering on Gaussian noise and evaluated on a separate set of triggers p… view at source ↗
Figures from the paper (10 more)
Figure 4
Figure 4. Figure 4: Fraction of evaluation data identified as outliers (SVM score [PITH_FULL_IMAGE:figures/full_fig_p009_4.png]
Figure 5
Figure 5. Figure 5: Distribution of scores produced by the SVM model when used to evaluate [PITH_FULL_IMAGE:figures/full_fig_p010_5.png]
Figure 6
Figure 6. Figure 6: ρ vs χ 2/ρ 2 histograms for LHO (top) and LLO (bottom) showing the concentration of triggers categorized as dirty (left), and those which the SVM model determined to be caused by glitches (right) in the 30th chunk of O3b. The blank, “cut out” region in the right column…
Figure 7
Figure 7. Figure 7: Number of pipeline triggers caused by specific glitch types (top panels) and [PITH_FULL_IMAGE:figures/full_fig_p015_7.png]
Figure 8
Figure 8. Figure 8: Density distribution of pipeline triggers caused by Scattered Light glitches [PITH_FULL_IMAGE:figures/full_fig_p016_8.png]
Figure 9
Figure 9. Figure 9: Distribution of Mass 1 vs Mass 2 values for glitch-induced pipeline triggers [PITH_FULL_IMAGE:figures/full_fig_p018_9.png]
Figure 10
Figure 10. Figure 10: Distribution of ρ vs. χ 2/ρ2 values for glitch-induced pipeline triggers during chunk 28 of O3b at LHO (left) and LLO (right). On the top row, each point represents a single trigger, colored by the Gravity Spy classification of the glitch that caused it. The bottom ro…
Figure 11
Figure 11. Figure 11: Over-representation ratios by glitch type and search category. The left [PITH_FULL_IMAGE:figures/full_fig_p021_11.png]
Figure 12
Figure 12. Figure 12: SNR histogram of all pipeline triggers caused by glitches at each detector over [PITH_FULL_IMAGE:figures/full_fig_p023_12.png]
Figure 13
Figure 13. Figure 13: Bhattacharyya Coefficients comparing trigger parameter distributions across [PITH_FULL_IMAGE:figures/full_fig_p024_13.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

43 extracted references · 8 canonical work pages

  1. [1]

    2015 Classical and Quantum Gravity 32 074001 (Preprint 1411.4547)

    Aasi J et al. 2015 Classical and Quantum Gravity 32 074001 (Preprint 1411.4547)

    arXiv 2015

  2. [2]

    2015 Classical and Quantum Gravity 32 024001 (Preprint 1408.3978)

    Acernese F et al. 2015 Classical and Quantum Gravity 32 024001 (Preprint 1408.3978)

    arXiv 2015

  3. [3]

    2021 Progress of Theoretical and Experimental Physics 2021 05A101 ( Preprint 2005.05574)

    Akutsu T et al. 2021 Progress of Theoretical and Experimental Physics 2021 05A101 ( Preprint 2005.05574)

    arXiv 2021

  4. [4]

    (KAGRA, VIRGO, LIGO Scientific) 2023 Phys

    Abbott R et al. (KAGRA, VIRGO, LIGO Scientific) 2023 Phys. Rev. X 13 041039 ( Preprint 2111.03606)

    arXiv 2023

  5. [5]

    Moe B, Brady P, Stephens B, Katsavounidis E, Williams R and Zhang F 2014 GraceDB: A Gravitational Wave Candidate Event Database URL https://dcc.ligo.org/LIGO-T1400365/ public

    work page 2014

  6. [6]

    (aLIGO) 2020 Phys

    Buikema A et al. (aLIGO) 2020 Phys. Rev. D 102 062003 (Preprint 2008.01301)

    arXiv 2020

  7. [7]

    (LIGO) 2021 Class

    Davis D et al. (LIGO) 2021 Class. Quant. Grav. 38 135014 (Preprint 2101.11673)

    arXiv 2021

  8. [8]

    (LIGO) 2025 Class

    Soni S et al. (LIGO) 2025 Class. Quant. Grav. 42 085016 (Preprint 2409.02831)

    arXiv 2025

Show all 43 references
  1. [9]

    (Virgo) 2023 Class

    Acernese F et al. (Virgo) 2023 Class. Quant. Grav. 40 185006 (Preprint 2210.15633)

    2023 arXiv

  2. [10]

    (KAGRA) 2021 Progress of Theoretical and Experimental Physics 2021 05A102 (Preprint 2009.09305)

    Akutsu T et al. (KAGRA) 2021 Progress of Theoretical and Experimental Physics 2021 05A102 (Preprint 2009.09305)

    2021

  3. [11]

    (LIGO Scientific, Virgo) 2018 Class

    Abbott B P et al. (LIGO Scientific, Virgo) 2018 Class. Quant. Grav. 35 065010 ( Preprint 1710.02185)

    2018 arXiv

  4. [12]

    2019 Class

    Cabero M et al. 2019 Class. Quant. Grav. 36 15 (Preprint 1901.05093)

    2019 arXiv

  5. [13]

    (LIGO Scientific, Virgo) 2016 Class

    Abbott B P et al. (LIGO Scientific, Virgo) 2016 Class. Quant. Grav. 33 134001 ( Preprint 1602.03844)

    2016 arXiv

  6. [14]

    Canton T D, Bhagwat S, Dhurandhar S V and Lundgren A 2014 Class. Quant. Grav. 31 015016 (Preprint 1304.0008)

    2014 arXiv

  7. [15]

    org/10.1088/1361-6382/aacf18

    Powell J 2018 Classical and Quantum Gravity 35 155017 ISSN 1361-6382 URL http://dx.doi. org/10.1088/1361-6382/aacf18

    2018 doi

  8. [16]

    2018 Phys

    Pankow C et al. 2018 Phys. Rev. D 98 084016 (Preprint 1808.03619) 28

    2018 arXiv

  9. [17]

    2017 Physical Review D 95 ISSN 2470-0029 URL http://dx.doi.org/10.1103/ PhysRevD.95.042001

    Messick C et al. 2017 Physical Review D 95 ISSN 2470-0029 URL http://dx.doi.org/10.1103/ PhysRevD.95.042001

    2017

  10. [18]

    2024 Physical Review D 109 042008 (Preprint 2305.05625)

    Ewing B et al. 2024 Physical Review D 109 042008 (Preprint 2305.05625)

    2024 arXiv

  11. [19]

    2020 Phys

    Hanna C et al. 2020 Phys. Rev. D 101 022003 (Preprint 1901.02227)

    2020 arXiv

  12. [20]

    2019 The GstLAL Search Analysis Methods for Compact Binary Mergers in Advanced LIGO’s Second and Advanced Virgo’s First Observing Runs ( Preprint 1901.08580)

    Sachdev S et al. 2019 The GstLAL Search Analysis Methods for Compact Binary Mergers in Advanced LIGO’s Second and Advanced Virgo’s First Observing Runs ( Preprint 1901.08580)

    2019 arXiv

  13. [21]

    Cannon K, Cariou R, Chapman A, Crispin-Ortuzar M, Fotopoulos N, Frei M, Hanna C, Kara E, Keppel D, Liao L, Privitera S, Searle A, Singer L and Weinstein A 2012 The Astrophysical Journal 748 136 ISSN 1538-4357 URL http://dx.doi.org/10.1088/0004-637X/748/2/136

    2012 doi

  14. [22]

    2023 Physical Review D 108 043004 (Preprint 2303.15700)

    Tsukada L et al. 2023 Physical Review D 108 043004 (Preprint 2303.15700)

    2023 arXiv

  15. [23]

    Davis D, White L V and Saulson P R 2020 Class. Quant. Grav. 37 145001 (Preprint 2002.09429)

    2020 arXiv

  16. [24]

    Lopez M, Caneva Santoro G, Martins A, Schmidt S, Schoppink J, van Straalen W, Capano C and Caudill S 2025 Phys. Rev. D 111 103020 (Preprint 2412.17169)

    2025

  17. [25]

    Ashton G, Thiele S, Lecoeuche Y, McIver J and Nuttall L K 2022 Class. Quant. Grav. 39 175004 (Preprint 2110.02689)

    2022 arXiv

  18. [26]

    Heinzel J, Talbot C, Ashton G and Vitale S 2023 Mon. Not. Roy. Astron. Soc. 523 5972–5984 (Preprint 2304.02665)

    2023 arXiv

  19. [27]

    Bondarescu R, Lundgren A and Macas R 2023 Phys. Rev. D 108 122004 (Preprint 2309.06594)

    2023 arXiv

  20. [28]

    Robinet F, Arnaud N, Leroy N, Lundgren A, Macleod D and McIver J 2020 SoftwareX 12 100620 (Preprint 2007.11374)

    2020 arXiv

  21. [29]

    2023 Class

    Glanzer J et al. 2023 Class. Quant. Grav. 40 065004 (Preprint 2208.12849)

    2023 arXiv

  22. [30]

    Bahaadini S, Noroozi V, Rohani N, Coughlin S, Zevin M, Smith J R, Kalogera V and Katsaggelos A 2018 Info. Sci. 444 172–186

    2018

  23. [31]

    2021 Class

    Soni S et al. 2021 Class. Quant. Grav. 38 195016 (Preprint 2103.12104)

    2021 arXiv

  24. [32]

    Mukund N, Abraham S, Kandhasamy S, Mitra S and Philip N S 2017 Phys. Rev. D 95 104059 (Preprint 1609.07259)

    2017 arXiv

  25. [33]

    Wu Y, Zevin M, Berry C P L, Crowston K, Østerlund C, Doctor Z, Banagiri S, Jackson C B, Kalogera V and Katsaggelos A K 2024 Phys. Rev. D (Preprint 2401.12913)

    2024 arXiv

  26. [34]

    Joshi P, Tsukada L and Hanna C 2023 Phys. Rev. D 108 084032 (Preprint 2305.18233)

    2023 arXiv

  27. [35]

    Cannon K, Hanna C and Keppel D 2013 Phys. Rev. D 88 024025 (Preprint 1209.0718)

    2013 arXiv

  28. [36]

    Cannon K, Hanna C and Peoples J 2015 arXiv e-prints arXiv:1504.04632 (Preprint 1504.04632)

    2015 arXiv

  29. [37]

    Bhattacharyya A 1943 Bulletin of the Calcutta Mathematical Society 35 99–109

    1943

  30. [38]

    Duda R O, Hart P E and Stork D G 2000 Pattern Classification 2nd ed (New York: Wiley- Interscience) ISBN 978-0471056690

    2000

  31. [39]

    Bishop C M 2006 Pattern Recognition and Machine Learning (New York: Springer) ISBN 978- 0387310732

    2006

  32. [40]

    2021 Classical and Quantum Gravity 38 025016 URL https: //arxiv.org/abs/2007.14876

    Soni S, Austin C, Effler A, Schofield R M S, Gonzalez G, Frolov V V, Driggers J C, Pele A, Urban A L, Valdes G et al. 2021 Classical and Quantum Gravity 38 025016 URL https: //arxiv.org/abs/2007.14876

    2021 arXiv

  33. [41]

    Magee R, Sharma R, Agrawal A and Udall R 2024 Mitigating the impact of noise transients in gravitational-wave searches using reduced basis timeseries and convolutional neural networks (Preprint 2410.15513) URL https://arxiv.org/abs/2410.15513

    2024 arXiv

  34. [42]

    2017 Class

    Zevin M et al. 2017 Class. Quant. Grav. 34 064003 (Preprint 1611.04596)

    2017 arXiv

  35. [43]

    Nitz A H 2018 Class. Quant. Grav. 35 035016 (Preprint 1709.08974) 29

    2018 arXiv

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.