pith. sign in

arxiv: 2410.19956 · v1 · submitted 2024-10-25 · 🌌 astro-ph.IM · astro-ph.HE· gr-qc

Gravitational-Wave Parameter Estimation in non-Gaussian noise using Score-Based Likelihood Characterization

Ronan Legin , Maximiliano Isi , Kaze W. K. Wong , Yashar Hezaveh , Laurence Perreault-Levasseur This is my paper

Pith reviewed 2026-05-23 19:41 UTC · model grok-4.3

classification 🌌 astro-ph.IM astro-ph.HEgr-qc
keywords gravitational wavesparameter estimationnon-Gaussian noisediffusion modelsLIGOglitcheslikelihood
0
0 comments X

The pith

Score-based diffusion models learn real LIGO noise distributions to enable unbiased gravitational-wave parameter estimation without cleaning glitches or assuming Gaussianity.

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

The paper demonstrates that score-based diffusion models trained directly on detector data can capture the empirical distribution of instrumental noise, including non-Gaussian transients. This learned distribution is then combined with a deterministic waveform model to form a likelihood that requires only noise additivity. Tests on 400 mock signals injected into real Livingston and Hanford data show that source parameters are recovered accurately even when loud glitches are present, and that population-level estimates remain unbiased without any data cleaning. Traditional methods rely on case-by-case glitch removal that can introduce biases in properties like binary precession; the new approach avoids those steps while remaining computationally feasible.

Core claim

By training score-based diffusion models on real LIGO detector data, the method learns an empirical representation of the noise distribution and uses it to construct an unbiased likelihood for gravitational-wave signals, recovering the true parameters from observations containing loud glitches and yielding unbiased inference across a population of signals without applying any cleaning to the data.

What carries the argument

Score-based diffusion models trained on detector data to represent the empirical noise distribution, combined with deterministic waveform models under the assumption of noise additivity.

If this is right

  • True source parameters are recovered from mock observations containing real LIGO noise with loud glitches present.
  • Inference remains unbiased over a population of signals when no cleaning is applied to the data.
  • The approach avoids biases that arise from case-by-case glitch removal procedures.
  • The method provides a scalable route to unbiased source property extraction for future observations.

Where Pith is reading between the lines

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

  • The same noise-characterization step could be applied to data from other detectors or to combined networks without separate cleaning pipelines.
  • Testing the recovered likelihood on actual detected events with well-constrained parameters would provide a direct check on real-world performance.
  • The learned noise model might be inspected to identify previously unrecognized statistical features in the detector output.

Load-bearing premise

The score-based diffusion model trained on detector data accurately represents the true conditional distribution of the noise that will be present during future observations containing signals.

What would settle it

Applying the method to a large set of real gravitational-wave events whose parameters have been independently verified by other means and finding systematic offsets in the recovered values when glitches are present would falsify the central claim.

Figures

Figures reproduced from arXiv: 2410.19956 by Kaze W. K. Wong, Laurence Perreault-Levasseur, Maximiliano Isi, Ronan Legin, Yashar Hezaveh.

Figure 1
Figure 1. Figure 1: Example of a sampled posterior distribution using SLIC (blue) versus using a Gaussian likelihood (orange). The ground truth model parameter is located at the intersection of the black lines. In the top right corner, the true signal is shown in orange superposed on top of the mock observation in blue. The contours represent the 68% and 99% credible intervals. The ground truth parameters of the waveform are … view at source ↗
Figure 2
Figure 2. Figure 2: Example similar to [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Coverage probability test for our sampled pos￾teriors from 200 mock observations with real noise from the Livingston detector and 200 from the Hanford detector. The orange (Livingston) and blue (Hanford) shaded regions rep￾resent the 1σ error estimated by bootstrapping over the set of sampled posteriors. The shaded grey regions show the 1σ and 3σ error of the null hypothesis in which the sampled posteriors… view at source ↗
read the original abstract

Gravitational-wave (GW) parameter estimation typically assumes that instrumental noise is Gaussian and stationary. Obvious departures from this idealization are typically handled on a case-by-case basis, e.g., through bespoke procedures to ``clean'' non-Gaussian noise transients (glitches), as was famously the case for the GW170817 neutron-star binary. Although effective, manipulating the data in this way can introduce biases in the inference of key astrophysical properties, like binary precession, and compound in unpredictable ways when combining multiple observations; alternative procedures free of the same biases, like joint inference of noise and signal properties, have so far proved too computationally expensive to execute at scale. Here we take a different approach: rather than explicitly modeling individual non-Gaussianities to then apply the traditional GW likelihood, we seek to learn the true distribution of instrumental noise without presuming Gaussianity and stationarity in the first place. Assuming only noise additivity, we employ score-based diffusion models to learn an empirical noise distribution directly from detector data and then combine it with a deterministic waveform model to provide an unbiased estimate of the likelihood function. We validate the method by performing inference on a subset of GW parameters from 400 mock observations, containing real LIGO noise from either the Livingston or Hanford detectors. We show that the proposed method can recover the true parameters even in the presence of loud glitches, and that the inference is unbiased over a population of signals without applying any cleaning to the data. This work provides a promising avenue for extracting unbiased source properties in future GW observations over the coming decade.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 2 minor

Summary. The paper proposes learning an empirical model of non-Gaussian, non-stationary LIGO noise via score-based diffusion models trained on detector data segments, then using the resulting score to construct a likelihood p(d|θ) = p_model(d − h(θ)) under the additivity assumption. This is combined with a deterministic waveform model to perform parameter estimation without Gaussian assumptions or glitch cleaning. Validation consists of recovering a subset of parameters from 400 mock injections into real Hanford/Livingston noise, with the claim that true parameters are recovered even in the presence of loud glitches and that the population-level inference is unbiased.

Significance. If the central claim holds after more rigorous validation, the approach would offer a scalable alternative to ad-hoc cleaning or joint noise-signal inference for unbiased GW parameter estimation in realistic noise, which is relevant for the high-event-rate era. The method's reliance on only additivity and its use of diffusion models for noise characterization are strengths. However, the current validation provides limited quantitative support for calibration or generalization.

major comments (2)
  1. [Validation results] Validation section (results on 400 mocks): the central claim that 'the inference is unbiased over a population of signals' and that parameters are recovered 'even in the presence of loud glitches' is presented without reported quantitative metrics such as posterior coverage fractions, calibration plots, bias statistics, or effective sample size. This leaves the strength of the claim difficult to assess from the provided evidence.
  2. [Method and validation] Method and validation: the score-based model is trained on segments independent of the signal model, but no tests are shown for whether the learned noise distribution matches the statistics of the specific test segments used for the 400 injections (or future observations). Any mismatch would directly bias the constructed likelihood p(d|θ) even under additivity; explicit checks for distribution shift or out-of-sample performance on held-out noise segments are needed to support the generalization claim.
minor comments (2)
  1. [Abstract] Abstract: states inference is performed on 'a subset of GW parameters' but does not specify which parameters (e.g., masses, spins, distance) are varied versus fixed.
  2. [Method] Notation: the precise form of the score-based likelihood (how the learned score is converted to a density or used in sampling) should be stated explicitly with an equation, as this is central to reproducibility.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive comments, which highlight opportunities to strengthen the quantitative support for our claims. We address each major point below and will revise the manuscript accordingly.

read point-by-point responses

  1. Referee: [Validation results] Validation section (results on 400 mocks): the central claim that 'the inference is unbiased over a population of signals' and that parameters are recovered 'even in the presence of loud glitches' is presented without reported quantitative metrics such as posterior coverage fractions, calibration plots, bias statistics, or effective sample size. This leaves the strength of the claim difficult to assess from the provided evidence.

    Authors: We agree that the validation would benefit from explicit quantitative metrics beyond the visual posterior recoveries shown for the 400 mocks. The current figures illustrate parameter recovery in the presence of glitches and population-level consistency, but lack formal calibration statistics. In the revised manuscript we will add posterior coverage fractions, bias statistics, and effective sample size values computed across the injection set to provide a more rigorous quantitative assessment of the unbiased inference claim. revision: yes

  2. Referee: [Method and validation] Method and validation: the score-based model is trained on segments independent of the signal model, but no tests are shown for whether the learned noise distribution matches the statistics of the specific test segments used for the 400 injections (or future observations). Any mismatch would directly bias the constructed likelihood p(d|θ) even under additivity; explicit checks for distribution shift or out-of-sample performance on held-out noise segments are needed to support the generalization claim.

    Authors: The referee correctly notes that explicit checks for distribution shift between training and test noise segments are not currently included. Although the training and injection segments are drawn from the same detector observing periods, we did not report direct comparisons of noise statistics or score evaluations on held-out segments. We will revise the manuscript to include such out-of-sample tests (e.g., comparisons of higher-order moments and score-function consistency on held-out noise) to support the generalization of the learned noise model. revision: yes

Circularity Check

0 steps flagged

No significant circularity

full rationale

The paper's derivation is self-contained: it trains a score-based diffusion model on segments of real detector noise to characterize the empirical distribution p_noise, then forms the likelihood as p(d|θ) = p_noise(d - h(θ)) under the explicit additivity assumption, and validates by running inference on 400 independent mock injections of signals into separate real LIGO noise segments. No step equates a claimed prediction or result to its inputs by construction, no fitted parameter is relabeled as a prediction, and no load-bearing premise reduces to a self-citation or imported uniqueness theorem. The reported unbiased recovery is an empirical outcome of the inference procedure rather than a definitional identity.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The method rests on one explicit domain assumption and on the empirical adequacy of a learned generative model whose training details are not supplied in the abstract.

axioms (1)
  • domain assumption Instrumental noise is additive to the gravitational-wave signal.
    Explicitly stated as the sole modeling assumption in the abstract.

pith-pipeline@v0.9.0 · 5844 in / 1122 out tokens · 25137 ms · 2026-05-23T19:41:34.732701+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. Learning What's Real: Disentangling Signal and Measurement Artifacts in Multi-Sensor Data, with Applications to Astrophysics

    astro-ph.IM 2026-04 unverdicted novelty 7.0

    A dual-encoder deep learning method disentangles intrinsic astrophysical signals from measurement artifacts by treating sensor effects as augmentations and using counterfactual generation on overlapping observations.

Reference graph

Works this paper leans on

65 extracted references · 65 canonical work pages · cited by 1 Pith paper · 6 internal anchors

  1. [1]

    , " * write output.state after.block = add.period write newline

    ENTRY address archivePrefix author booktitle chapter doi edition editor eprint howpublished institution journal key month number organization pages publisher school series title misctitle type volume year version url label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block FUNCTION init.state.consts ...

    work page

  2. [2]

    write newline

    " write newline "" before.all 'output.state := FUNCTION format.url url empty "" new.block "" url * "" * if FUNCTION format.eprint eprint empty "" archivePrefix empty "" archivePrefix "arXiv" = new.block " " eprint * " " * new.block " " eprint * " " * if if if FUNCTION format.doi doi empty "" " " doi * " " * if FUNCTION format.pid doi empty eprint empty ur...

    work page

  3. [3]

    Advanced LIGO

    thebibliography [1] 20pt to REFERENCES 6pt =0pt -12pt 10pt plus 3pt =0pt =0pt =1pt plus 1pt =0pt =0pt -12pt =13pt plus 1pt =20pt =13pt plus 1pt \@M =10000 =-1.0em =0pt =0pt 0pt =0pt =1.0em @enumiv\@empty 10000 10000 `\.\@m \@noitemerr \@latex@warning Empty `thebibliography' environment \@ifnextchar \@reference \@latexerr Missing key on reference command E...

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.5281/zenodo.831784 2021

  4. [4]

    2015, Class

    Aasi, J., et al. 2015, Class. Quant. Grav., 32, 074001, 10.1088/0264-9381/32/7/074001

    work page doi:10.1088/0264-9381/32/7/074001 2015

  5. [5]

    2023, Gravitational Wave Open Science Center (1126216262-1126302662)

    Abbot, R., et al. 2023, Gravitational Wave Open Science Center (1126216262-1126302662). https://gwosc.org/archive/links/O1/H1/1126216262/1126302662/simple/

    work page 2023

  6. [6]

    P., et al

    Abbott, B. P., et al. 2016 a , Phys. Rev. Lett., 116, 061102, 10.1103/PhysRevLett.116.061102

    work page doi:10.1103/physrevlett.116.061102 2016

  7. [7]

    2016 b , Class

    ---. 2016 b , Class. Quant. Grav., 33, 134001, 10.1088/0264-9381/33/13/134001

    work page doi:10.1088/0264-9381/33/13/134001 2016

  8. [8]

    2016 c , Phys

    ---. 2016 c , Phys. Rev. D, 93, 112004, 10.1103/PhysRevD.93.112004

    work page doi:10.1103/physrevd.93.112004 2016

  9. [9]

    2016 d , Phys

    ---. 2016 d , Phys. Rev. Lett., 116, 241102, 10.1103/PhysRevLett.116.241102

    work page doi:10.1103/physrevlett.116.241102 2016

  10. [10]

    2017, Phys

    ---. 2017, Phys. Rev. Lett., 119, 161101, 10.1103/PhysRevLett.119.161101

    work page doi:10.1103/physrevlett.119.161101 2017

  11. [11]

    2020, Class

    ---. 2020, Class. Quant. Grav., 37, 055002, 10.1088/1361-6382/ab685e

    work page doi:10.1088/1361-6382/ab685e 2020

  12. [12]

    2021, SoftwareX, 13, 100658, 10.1016/j.softx.2021.100658

    Abbott, R., et al. 2021, SoftwareX, 13, 100658, 10.1016/j.softx.2021.100658

    work page doi:10.1016/j.softx.2021.100658 2021

  13. [13]

    2023, Phys

    ---. 2023, Phys. Rev. X, 13, 041039, 10.1103/PhysRevX.13.041039

    work page doi:10.1103/physrevx.13.041039 2023

  14. [14]

    2024, Phys

    ---. 2024, Phys. Rev. D, 109, 022001, 10.1103/PhysRevD.109.022001

    work page doi:10.1103/physrevd.109.022001 2024

  15. [15]

    2015, Class

    Acernese, F., et al. 2015, Class. Quant. Grav., 32, 024001, 10.1088/0264-9381/32/2/024001

    work page doi:10.1088/0264-9381/32/2/024001 2015

  16. [16]

    2023, Mon

    Ashton, G. 2023, Mon. Not. Roy. Astron. Soc., 520, 2983, 10.1093/mnras/stad341

    work page doi:10.1093/mnras/stad341 2023

  17. [17]

    D., et al

    Ashton, G., et al. 2019, Astrophys. J. Suppl., 241, 27, 10.3847/1538-4365/ab06fc

    work page doi:10.3847/1538-4365/ab06fc 2019

  18. [18]

    2013, Phys

    Aso, Y., Michimura, Y., Somiya, K., et al. 2013, Phys. Rev. D, 88, 043007, 10.1103/PhysRevD.88.043007

    work page doi:10.1103/physrevd.88.043007 2013

  19. [19]

    Berry, C. P. L., et al. 2015, Astrophys. J., 804, 114, 10.1088/0004-637X/804/2/114

    work page doi:10.1088/0004-637x/804/2/114 2015

  20. [20]

    M., Capano, C

    Biwer, C. M., Capano, C. D., De, S., et al. 2019, Publ. Astron. Soc. Pac., 131, 024503, 10.1088/1538-3873/aaef0b

    work page doi:10.1088/1538-3873/aaef0b 2019

  21. [21]

    Large Scale GAN Training for High Fidelity Natural Image Synthesis

    Brock , A., Donahue , J., & Simonyan , K. 2018, arXiv e-prints, arXiv:1809.11096, 10.48550/arXiv.1809.11096

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.1809.11096 2018

  22. [22]

    2020, Phys

    Buikema, A., et al. 2020, Phys. Rev. D, 102, 062003, 10.1103/PhysRevD.102.062003

    work page doi:10.1103/physrevd.102.062003 2020

  23. [23]

    Chatziioannou, K., Cornish, N., Wijngaarden, M., & Littenberg, T. B. 2021, Phys. Rev. D, 103, 044013, 10.1103/PhysRevD.103.044013

    work page doi:10.1103/physrevd.103.044013 2021

  24. [24]

    R., Gelman, A., & Rubin, D

    Cook, S. R., Gelman, A., & Rubin, D. B. 2006, Journal of Computational and Graphical Statistics, 15, 675. http://www.jstor.org/stable/27594203

    work page arXiv 2006

  25. [25]

    J., & Littenberg, T

    Cornish, N. J., & Littenberg, T. B. 2015, Class. Quant. Grav., 32, 135012, 10.1088/0264-9381/32/13/135012

    work page doi:10.1088/0264-9381/32/13/135012 2015

  26. [26]

    J., Littenberg, T

    Cornish, N. J., Littenberg, T. B., B\'ecsy, B., et al. 2021, Phys. Rev. D, 103, 044006, 10.1103/PhysRevD.103.044006

    work page doi:10.1103/physrevd.103.044006 2021

  27. [27]

    B., Romero-Shaw, I

    Davis, D., Littenberg, T. B., Romero-Shaw, I. M., et al. 2022, Class. Quant. Grav., 39, 245013, 10.1088/1361-6382/aca238

    work page doi:10.1088/1361-6382/aca238 2022

  28. [28]

    J., Lundgren, A

    Davis, D., Massinger, T. J., Lundgren, A. P., et al. 2019, Class. Quant. Grav., 36, 055011, 10.1088/1361-6382/ab01c5

    work page doi:10.1088/1361-6382/ab01c5 2019

  29. [29]

    R., Gair , J., et al

    Dax , M., Green , S. R., Gair , J., et al. 2021, , 127, 241103, 10.1103/PhysRevLett.127.241103

    work page doi:10.1103/physrevlett.127.241103 2021

  30. [30]

    R., Gair, J., et al

    Dax, M., Green, S. R., Gair, J., et al. 2024. 2407.09602

    work page arXiv 2024

  31. [31]

    Edwards, T. D. P., Wong, K. W. K., Lam, K. K. H., et al. 2024, Phys. Rev. D, 110, 064028, 10.1103/PhysRevD.110.064028

    work page doi:10.1103/physrevd.110.064028 2024

  32. [32]

    2022, Nature, 610, 652, 10.1038/s41586-022-05212-z

    Hannam, M., et al. 2022, Nature, 610, 652, 10.1038/s41586-022-05212-z

    work page doi:10.1038/s41586-022-05212-z 2022

  33. [33]

    2023, Mon

    Heinzel, J., Talbot, C., Ashton, G., & Vitale, S. 2023, Mon. Not. Roy. Astron. Soc., 523, 5972, 10.1093/mnras/stad1823

    work page doi:10.1093/mnras/stad1823 2023

  34. [34]

    Denoising Diffusion Probabilistic Models

    Ho , J., Jain , A., & Abbeel , P. 2020, arXiv e-prints, arXiv:2006.11239, 10.48550/arXiv.2006.11239

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2006.11239 2020

  35. [35]

    2022, Phys

    Hourihane, S., Chatziioannou, K., Wijngaarden, M., et al. 2022, Phys. Rev. D, 106, 042006, 10.1103/PhysRevD.106.042006

    work page doi:10.1103/physrevd.106.042006 2022

  36. [36]

    Hyv\" a rinen, A. 2005, J. Mach. Learn. Res., 6, 695–709

    work page 2005

  37. [37]

    2016, Phys

    Khan, S., Husa, S., Hannam, M., et al. 2016, Phys. Rev. D, 93, 044007, 10.1103/PhysRevD.93.044007

    work page doi:10.1103/physrevd.93.044007 2016

  38. [38]

    Adam: A Method for Stochastic Optimization

    Kingma, D. P., & Ba, J. 2014, arXiv preprint arXiv:1412.6980

    work page internal anchor Pith review Pith/arXiv arXiv 2014

  39. [39]

    Kobyzev , I., Prince , S. J. D., & Brubaker , M. A. 2019, arXiv e-prints, arXiv:1908.09257, 10.48550/arXiv.1908.09257

    work page doi:10.48550/arxiv.1908.09257 2019

  40. [40]

    2024, scoregen\_jax

    Legin, R. 2024, scoregen\_jax. https://github.com/RonanLegin/scoregen_jax

    work page 2024

  41. [41]

    2023, arXiv e-prints, arXiv:2302.03046, 10.48550/arXiv.2302.03046

    Legin , R., Adam , A., Hezaveh , Y., & Perreault Levasseur , L. 2023, arXiv e-prints, arXiv:2302.03046, 10.48550/arXiv.2302.03046

    work page doi:10.48550/arxiv.2302.03046 2023

  42. [42]

    2023, in Machine Learning for Astrophysics

    Legin, R., Isi, M., Wong, K., et al. 2023, in Machine Learning for Astrophysics. Workshop at the Fortieth International Conference on Machine Learning (ICML 2023), 17. https://ml4astro.github.io/icml2023/assets/70.pdf

    work page 2023

  43. [43]

    2024, slicgw

    Legin, R., Wong, K., & Isi, M. 2024, slicgw. https://github.com/RonanLegin/slicgw

    work page 2024

  44. [44]

    Lemos, A

    Lemos , P., Coogan , A., Hezaveh , Y., & Perreault-Levasseur , L. 2023, 40th International Conference on Machine Learning, 202, 19256, 10.48550/arXiv.2302.03026

    work page doi:10.48550/arxiv.2302.03026 2023

  45. [45]

    R., Toubiana , A., & Gair , J

    Leyde , K., Green , S. R., Toubiana , A., & Gair , J. 2024, , 109, 064056, 10.1103/PhysRevD.109.064056

    work page doi:10.1103/physrevd.109.064056 2024

  46. [46]

    2024, Phys

    Macas, R., Lundgren, A., & Ashton, G. 2024, Phys. Rev. D, 109, 062006, 10.1103/PhysRevD.109.062006

    work page doi:10.1103/physrevd.109.062006 2024

  47. [47]

    2018, Phys

    Pankow, C., et al. 2018, Phys. Rev. D, 98, 084016, 10.1103/PhysRevD.98.084016

    work page doi:10.1103/physrevd.98.084016 2018

  48. [48]

    Papamakarios, E

    Papamakarios , G., Nalisnick , E., Jimenez Rezende , D., Mohamed , S., & Lakshminarayanan , B. 2019, arXiv e-prints, arXiv:1912.02762, 10.48550/arXiv.1912.02762

    work page doi:10.48550/arxiv.1912.02762 2019

  49. [49]

    2022, Phys

    Payne, E., Hourihane, S., Golomb, J., et al. 2022, Phys. Rev. D, 106, 104017, 10.1103/PhysRevD.106.104017

    work page doi:10.1103/physrevd.106.104017 2022

  50. [50]

    Raymond, V., Al-Shammari, S., & G\"ottel, A. 2024. 2406.03935

    work page arXiv 2024

  51. [51]

    2023, , 672, A51, 10.1051/0004-6361/202243054

    Remy , B., Lanusse , F., Jeffrey , N., et al. 2023, , 672, A51, 10.1051/0004-6361/202243054

    work page doi:10.1051/0004-6361/202243054 2023

  52. [52]

    O., & Tweedie, R

    Roberts, G. O., & Tweedie, R. L. 1996, Bernoulli, 2, 341. http://www.jstor.org/stable/3318418

    work page arXiv 1996

  53. [53]

    M., et al

    Romero-Shaw, I. M., et al. 2020, Mon. Not. Roy. Astron. Soc., 499, 3295, 10.1093/mnras/staa2850

    work page doi:10.1093/mnras/staa2850 2020

  54. [54]

    Generative Modeling by Estimating Gradients of the Data Distribution

    Song , Y., & Ermon , S. 2019, arXiv e-prints, arXiv:1907.05600, 10.48550/arXiv.1907.05600

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.1907.05600 2019

  55. [55]

    Score-Based Generative Modeling through Stochastic Differential Equations

    Song , Y., Sohl-Dickstein , J., Kingma , D. P., et al. 2020, arXiv e-prints, arXiv:2011.13456, 10.48550/arXiv.2011.13456

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2011.13456 2020

  56. [56]

    2014, Journal of Machine Learning Research, 15, 1929

    Srivastava, N., Hinton, G., Krizhevsky, A., Sutskever, I., & Salakhutdinov, R. 2014, Journal of Machine Learning Research, 15, 1929. http://jmlr.org/papers/v15/srivastava14a.html

    work page 2014

  57. [57]

    2018, arXiv e-prints, arXiv:1804.06788, 10.48550/arXiv.1804.06788

    Talts , S., Betancourt , M., Simpson , D., Vehtari , A., & Gelman , A. 2018, arXiv e-prints, arXiv:1804.06788, 10.48550/arXiv.1804.06788

    work page doi:10.48550/arxiv.1804.06788 2018

  58. [58]

    2023, Appl

    Udall, R., & Davis, D. 2023, Appl. Phys. Lett., 122, 094103, 10.1063/5.0136896

    work page doi:10.1063/5.0136896 2023

  59. [59]

    Udall, R., Hourihane, S., Miller, S., et al. 2024. 2409.03912

    work page arXiv 2024

  60. [60]

    2015, Phys

    Veitch, J., et al. 2015, Phys. Rev. D, 91, 042003, 10.1103/PhysRevD.91.042003

    work page doi:10.1103/physrevd.91.042003 2015

  61. [61]

    2011, Neural Comput., 23, 1661, 10.1162/NECO\_a\_00142

    Vincent, P. 2011, Neural Comput., 23, 1661, 10.1162/NECO\_a\_00142

    work page doi:10.1162/neco 2011

  62. [62]

    Wong, K. W. k., Gabri\'e, M., & Foreman-Mackey, D. 2023 a , J. Open Source Softw., 8, 5021, 10.21105/joss.05021

    work page doi:10.21105/joss.05021 2023

  63. [63]

    Wong, K. W. K., Isi, M., & Edwards, T. D. P. 2023 b , Astrophys. J., 958, 129, 10.3847/1538-4357/acf5cd

    work page doi:10.3847/1538-4357/acf5cd 2023

  64. [64]

    Xiong, C.-Y., Sun, T.-Y., Zhang, J.-F., & Zhang, X. 2024. 2405.09475

    work page arXiv 2024

  65. [65]

    2022, arXiv e-prints, arXiv:2209.00796, 10.48550/arXiv.2209.00796

    Yang , L., Zhang , Z., Song , Y., et al. 2022, arXiv e-prints, arXiv:2209.00796, 10.48550/arXiv.2209.00796

    work page doi:10.48550/arxiv.2209.00796 2022