Pith. sign in

REVIEW 4 major objections 5 minor 26 references

Diffusion-guided optimization for full waveform inversion

T0 review · 4 major / 5 minor · reviewed 2026-08-01 · deepseek-v4-flash

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

Pith's one-line read A pretrained diffusion model of geological velocity patches can act as a learned regularizer for full waveform inversion, and an alternating split-Gibbs coupling beats standard L2 and total-variation baselines on synthetic benchmarks.

desk verdict A genuinely useful empirical comparison of diffusion-prior couplings for nonlinear FWI, with a believable in-distribution result; the benchmark-scale transfer claims are real but need deployment-mode ablations before you trust them. read the letter →

arxiv 2607.21987 v1 pith:37JSNTVK submitted 2026-07-24 physics.geo-ph

Yiran Shen , Yangkang Chen , Björn Engquist This is my paper · ORCID
classification physics.geo-ph
keywords fullwaveforminversiondiffusiongenerativemodelslearnedregularizationseismicvelocityscore-baseddenoisingalternatingoptimizationgeologicalpriorinverseproblems
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

This paper argues that a pretrained diffusion model of geological velocity patches can serve as a practical learned regularizer inside the full waveform inversion (FWI) loop, without being retrained on seismic data. It compares three ways of coupling diffusion denoising with PDE-based FWI updates and finds that the split Gibbs scheme—alternating a data-misfit FWI step with a diffusion-prior denoising step—is the most consistent. On held-out synthetic models, benchmark-scale Marmousi and Overthrust tests, and noisy data down to 10 dB, this scheme improves reconstruction quality relative to standard L2 and total-variation-regularized FWI. The core claim is that learned diffusion priors can stabilize precisely the parts of FWI that classical regularizers handle poorly, at modest extra computational cost, at least on synthetic benchmarks.

What carries the argument

The load-bearing mechanism is the denoising map of a diffusion model pretrained on 100×100 geological velocity patches, used as a learned prior through Tweedie's estimate of the clean model. In Split Gibbs Diffusion Sampling, each outer cycle solves a penalized FWI subproblem for data consistency, adds controlled noise, then applies the denoiser; the coupling strength is set by the noise level schedule. This alternation is what lets the diffusion prior repair weakly constrained structure without overwhelming the PDE-constrained likelihood update.

What would settle it

Run SGDS on a single benchmark, such as Marmousi, with all four deployment modes—direct patch, downsample-single, downsample-multi, and columnwise denoising—keeping everything else fixed, and report per-mode PSNR with repeated runs. If the advantage over TV disappears or flips sign under any deployment choice, or if the winning mode was selected after seeing results, the benchmark-transfer claim is called into question.

Watch

Extended reading notes

Core claim

The central claim is that separating the likelihood and prior updates—rather than threading FWI gradients into every diffusion step—makes diffusion-guided inversion stable enough to outperform classical regularizers in a nonlinear wave-equation setting. In 18 held-out synthetic inversions, the split Gibbs scheme reaches mean PSNR 21.56 ± 3.51 versus 19.96 ± 3.91 for plain L2 FWI; on Marmousi PSNR rises from 17.44 to 21.01, on Overthrust from 12.73 to 25.36, and it remains best at 20 and 10 dB noise before degrading sharply at 5 dB. The paper interprets the alternating update as a projected-gradient-like step in which the denoiser acts as an approximate projection onto a learned manifold of g

Load-bearing premise

The benchmark-scale improvements rest on transferring a 100×100 patch-trained diffusion prior to larger, structurally distinct models through hand-picked deployment modes; if that transfer, rather than the learned prior itself, produces the reported gains, the central claim weakens.

Editorial extensions

If this is right

  • If correct, learned diffusion priors can be added as practical regularizers alongside classical ones for FWI, improving salt, fault, and deep-layer recovery.
  • SGDS retains its advantage over classical baselines at moderate noise levels (20 and 10 dB) on synthetic Marmousi data, motivating noise-robust likelihood models for extreme noise.
  • Benchmark-scale transfer shows that deployment mode—direct patch, downsample, or columnwise denoising—is a tunable part of the algorithm, not a trivial detail.
  • The alternating-scheme interpretation offers a local stability rationale for why splitting works, suggesting principled ways to schedule noise levels.
  • Diffusion-guided FWI adds computational overhead (about 4.1× wall time in the reported cases) but remains cheaper than tightly coupled guidance strategies.

Reading between the lines

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

  • The benchmark-scale gains may depend substantially on the chosen patch deployment; a reader should treat the Marmousi and Overthrust numbers as workflow-level results, not pure evidence about the learned prior, until deployment modes are ablated with error bars.
  • A natural extension is adaptive noise scheduling pegged to misfit reduction, which the paper leaves for future work; it could remove the manual tuning of the diffusion start time.
  • Because the prior was trained only on one synthetic geological family, the method's value on genuinely out-of-distribution geology is untested; a cost-matched comparison against a diffusion model trained on the target benchmark would separate prior quality from coupling mechanism.
  • The local contraction argument assumes the denoiser is close to an exact projection; measuring that projection defect on real denoisers would give a quantitative stability margin.
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 / 5 minor

Summary. The paper proposes using a pretrained diffusion model as a learned regularizer inside PDE-constrained full waveform inversion (FWI). Three guidance strategies are compared: MPGD (FWI steps injected into diffusion inference), SDEdit (denoising an FWI seed from an intermediate noise level), and SGDS (alternating FWI likelihood updates with diffusion denoising). The core empirical claim is that SGDS improves reconstruction quality over standard L2 FWI and L2+TV FWI in clean and moderately noisy synthetic settings. Evidence includes an 18-case GeoFWI study (Table 6), benchmark-scale Marmousi and Overthrust tests, a noise-robustness study, and a Sigsbee2A salt stress test. The paper also presents a local 'projected-gradient' interpretation of SGDS via Eq. (12). Code and data links are provided.

Significance. If the empirical claims are robust, the paper would provide a practical recipe for using unconditional diffusion priors in strongly nonlinear wave-equation inversion, with useful cost characterization (Table 4) and an honest discussion of deployment and out-of-distribution issues. The in-distribution GeoFWI results, with mean±std over 18 cases, support a modest claim that SGDS outperforms the two classical baselines and the two other diffusion couplings, although the margins are not tested for significance. The larger benchmark-scale claims are less secure, because the deployment strategy is selected after seeing results and is not ablated or repeated. The theoretical result (Eq. 12) is asserted rather than proved. The paper is transparent about several limitations and provides reproducible code and data, which is a strength.

major comments (4)
  1. [§5.4, §5.5, §6.3] The benchmark-scale Marmousi/Overthrust gains are not attributable to the learned prior because the deployment mode (direct patch, downsample-single, downsample-multi, columnwise) is selected per benchmark after observing results, and the reported numbers are single runs. The paper itself states that 'the deployment choice matters' and that 'simple patchwise use can introduce stitching or scale artifacts' (§6.3). Without an ablation that varies the deployment mode while holding the diffusion prior fixed, or error bars over deployment variants, the 21.01 dB Marmousi and 25.36 dB Overthrust results could be artifacts of the chosen patching/smoothing strategy rather than of the diffusion prior. This is load-bearing for the abstract's benchmark-transfer claim.
  2. [§2.1.2, Eq. (12)] The local stability bound in Eq. (12) is asserted as a 'standard projected-gradient expansion' but no proof, explicit constant definitions, or validation of the assumptions are given. The assumptions that the denoiser P_θ is an approximate projection with small defect δ and that ∇J is locally Lipschitz/coercive along tangent directions are not verified for the trained denoiser or the FWI objective. As stated, Eq. (12) is a heuristic. Either provide a rigorous derivation under explicit and checkable conditions, or explicitly label the bound as an interpretive analogy rather than a theoretical guarantee.
  3. [§5.3, Tables 5–6] The diffusion start time σ1=350 is selected from Table 5 using the same family of GeoFWI test cases that later appear in the 18-case statistics of Table 6. No separate validation split is described for hyperparameter selection. Since SGDS's advantage depends on this tuned noise level, the reported mean improvements may be inflated by test-set selection. Please describe the selection procedure, or provide a clear split between cases used for tuning and cases used for final evaluation.
  4. [§3.3, Eqs. (20)–(22)] The conditional likelihood gradient is written as ∇_xt ||y − Ru(x̂0(xt))||², but the adjoint-state expression in Eq. (21) computes the gradient with respect to the model x̂0, not with respect to xt. The RHS of Eq. (22) therefore appears to be ∇_{x̂0} log p_t(y|x̂0), missing the Jacobian ∂x̂0/∂xt. This is technically incorrect as a derivation of the conditional score. Please clarify whether an approximation is intended (e.g., ignoring the Tweedie Jacobian) and state the resulting bias, or correct the chain rule.
minor comments (5)
  1. [§2.1.1, Theorem 1] The existence/convergence theorem for TV regularization is stated without proof or citation. It is standard, but a reference would help readers verify the exact conditions (e.g., the weak* sequential closedness of F).
  2. [§1, References] The text says 'Tikhonov (1963) [3]', but reference [3] is Engl and Ramlau's encyclopedia entry, not Tikhonov's 1963 paper. The citation should be corrected or supplemented.
  3. [§5.1] The phrase 'In additional dissertation experiments...' is vague and not reproducible. Please cite the dissertation or remove the sentence, since it concerns a claimed effect on salt recovery.
  4. [Table 4] MPGD wall time is reported, but its PDE-call count is listed as 'n/a'. This makes the cost comparison incomplete, since the number of PDE solves is the main driver of FWI cost. Please instrument and report it.
  5. [Global] There are minor formatting issues: equation numbers (23) is inline within Algorithm 3, some references to 'L 2' have irregular spacing, and Table 7 reports results from the 'Ap3/downsample-multi' setting without a definition of 'Ap3'.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: SGDS is an empirical composition of existing methods; no reported metric is forced by construction from a fitted parameter or self-citation.

full rationale

The paper's central claims are empirical benchmark comparisons, not analytical derivations whose conclusions are identical to their inputs. The SGDS update (Eq. 11) and the local stability bound (Eq. 12) are explicitly framed as a local 'interpretation' and 'algorithmic interpretation' rather than a convergence theorem; the bound follows from standard projected-gradient estimates under assumptions (approximate projection, Lipschitz/coercivity) that are not themselves the target result. No equation in the paper reduces a fitted parameter to a reported prediction: the diffusion start time and deployment mode are tuned or selected, but the paper does not re-describe that tuning as an independent prediction. The GeoFWI-trained prior is used on held-out GeoFWI cases and on externally defined Marmousi, Overthrust, and Sigsbee2A benchmarks; applying a patchwise prior to a larger model is not circular because the target models are not used to train the prior. The single self-citation [8] (DLM-FWI) appears only as background motivation and is not load-bearing for any of the SGDS claims. The acknowledged 'final benchmark polishing step' and deployment-mode selection are methodological limitations and potential overfitting risks, but the manuscript explicitly discusses deployment choice as part of the algorithm rather than presenting a single pre-specified deployment as having predicted the result. Under the stated rubric, those concerns belong to correctness risk, not circularity. Therefore the derivation chain is self-contained and no circular step can be exhibited.

Assumptions & free parameters 6 free parameters · 5 assumptions · 0 invented entities

No new physical entities are postulated. The 'learned geological manifold' Sigma_theta is an algorithmic abstraction tied to the trained network, with generated samples shown in Fig. 4 rather than independent falsifiable evidence. The main load-bearing free inputs are the SGDS noise schedule, the diffusion start time, the TV weight, the frequency schedule, and the deployment mode, several of which were tuned on the test cases used to report the headline numbers.

free parameters (6)
  • SGDS diffusion start time / initial noise level sigma_1 = t≈350 (sigma_1=350) selected from sweep
    Table 5 and Fig. 9 sweep on representative GeoFWI cases; the best-performing setting is then used in the reported mean and benchmark runs.
  • TV regularization weight lambda_TV = 0.01
    Set uniformly in Table 3; no sensitivity analysis is provided, which matters for the fairness of the classical baseline comparison.
  • Guidance/coupling strength rho and SGDS noise schedule sigma_k, alpha = not reported explicitly
    Eq. (22) uses rho; Algorithm 3 uses sigma_k and alpha*sigma_k/sigma_1 perturbation. No concrete values or schedule are reported, yet these control the balance between FWI likelihood and diffusion prior.
  • Frequency continuation schedule = 10, 15, 20, 25 Hz (35 Hz endpoint for diagnostics)
    Multiscale schedule chosen per benchmark; it strongly affects cycle-skipping and the difficulty of the inversion for all methods.
  • Patch deployment mode = Ap3/downsample-multi, 10% overlap; alternatives: direct patch, downsample-single, columnwise
    Marmousi/Overthrust use a deployment mode selected after seeing results; the paper admits deployment choice matters and can introduce stitching/scale artifacts (§6.3).
  • Number of SGDS outer blocks K = e.g., 4 FWI blocks in the Marmousi diagnostics (Fig. 12)
    Block count chosen empirically; no convergence criterion or sensitivity analysis is given.
assumptions (5)
  • domain assumption The acoustic wave-equation forward model F(m) accurately reproduces the observed data (known wavelet, sources, receivers, fixed density).
    All experiments are synthetic, but the data-consistency step assumes the PDE and acquisition model the data; field-data complications are deferred to §6.5.
  • domain assumption The GeoFWI training distribution is representative enough of the target structures (salt, faults, layers; Marmousi; Overthrust; Sigsbee2A) for the learned prior to be useful.
    This is the transfer premise for the benchmark claims; the paper itself flags the risk in Table 2 ('too strong: hallucination') and §7.
  • ad hoc to paper In a neighborhood of a reference model, the denoiser P_theta acts as an approximate projection onto the learned manifold with small defect delta, and grad J is locally Lipschitz and coercive along tangent directions.
    Eq. (12), the local contractive bound for SGDS, rests on these unverified conditions. The paper calls it a local interpretation rather than a global theorem.
  • standard math Theorem 1 (existence/stability of TV-regularized FWI) is a standard variational result.
    Invoked in §2.1.1 as background; not the paper's contribution.
  • standard math Tweedie's formula and denoising score matching give a valid approximation of the posterior expectation in Eqs. (18)-(19) and (23).
    Standard in the diffusion-model literature; used throughout the guidance derivations.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Diffusion-guided optimization for full waveform inversion." pith.science (2026). https://pith.science/paper/37JSNTVK

@misc{pith2026260721987,
  author       = {Pith},
  title        = {Pith review of: Diffusion-guided optimization for full waveform inversion},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/37JSNTVK}},
  note         = {Machine review of arXiv:2607.21987}
}
read the original abstract

We present a diffusion-guided full waveform inversion (FWI) study in which pretrained diffusion generative models are used as learned regularizers inside a PDE-constrained seismic inversion loop. We compare three training-free guidance strategies: Manifold-Preserving Guided Diffusion (MPGD), SDEdit-based initialization, and Split Gibbs Diffusion Sampling (SGDS), which alternates between FWI likelihood updates and diffusion-prior denoising. The proposed workflow keeps wave-equation modeling in the inversion loop and uses a geological prior to stabilize model components that are weakly constrained by the seismic data. Controlled GeoFWI experiments, benchmark-scale Marmousi and Overthrust tests, a difficult Sigsbee2A salt test, and noise-degradation studies show that SGDS improves reconstruction quality relative to conventional L2 and total-variation regularized FWI in clean and moderately noisy synthetic settings. Overall, these experiments demonstrate that diffusion-guided optimization can serve as a practical learned regularization strategy for synthetic FWI benchmarks while preserving the wave-equation modeling loop.

Figures

Figures reproduced from arXiv: 2607.21987 by the authors.

Figure 1
Figure 1. Illustration of the Manifold-Preserving Guided Diffusion (MPGD) method, where FWI iteratively refines [PITH_FULL_IMAGE:figures/full_fig_p008_1.png] view at source ↗
Figure 2
Figure 2. SDEdit process with FWI result as guidance at time [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗
Figure 3
Figure 3. Workflow of SGDS Method, where FWI and diffusion processes alternate iteratively, refining the velocity [PITH_FULL_IMAGE:figures/full_fig_p011_3.png] view at source ↗
Figures from the paper (13 more)
Figure 4
Figure 4. Figure 4: Representative training velocity models (left) and diffusion-generated samples (right). The generated models [PITH_FULL_IMAGE:figures/full_fig_p012_4.png]
Figure 5
Figure 5. Figure 5: Representative augmented salt training models (left block) and unconditional diffusion-generated samples [PITH_FULL_IMAGE:figures/full_fig_p014_5.png]
Figure 6
Figure 6. Figure 6: Salt body inversion results for four synthetic models. Columns show the ground truth, conventional FWI [PITH_FULL_IMAGE:figures/full_fig_p015_6.png]
Figure 7
Figure 7. Figure 7: Faulted velocity model inversion results for four synthetic examples. Conventional FWI struggles to re [PITH_FULL_IMAGE:figures/full_fig_p016_7.png]
Figure 8
Figure 8. Figure 8: Layered velocity model inversion results for four synthetic examples. Conventional FWI produces oscil [PITH_FULL_IMAGE:figures/full_fig_p017_8.png]
Figure 9
Figure 9. Figure 9: Effect of diffusion initialization level in the SGDS algorithm for four salt body models. The first column [PITH_FULL_IMAGE:figures/full_fig_p018_9.png]
Figure 10
Figure 10. Figure 10: Marmousi inversion results. (a) Initial smooth velocity model used for inversion. (b) Conventional FWI [PITH_FULL_IMAGE:figures/full_fig_p020_10.png]
Figure 11
Figure 11. Figure 11: Step-by-step Marmousi SGDS evolution using the downsample-multi deployment mode. The sequence [PITH_FULL_IMAGE:figures/full_fig_p021_11.png]
Figure 12
Figure 12. Figure 12: Marmousi SGDS Ap3 convergence diagnostics computed from saved outer-loop states. The FWI blocks [PITH_FULL_IMAGE:figures/full_fig_p022_12.png]
Figure 13
Figure 13. Figure 13: Overthrust inversion results. (a) Initial smooth velocity model. (b) Conventional [PITH_FULL_IMAGE:figures/full_fig_p023_13.png]
Figure 14
Figure 14. Figure 14: Noise-degradation curve for the Marmousi experiment. SGDS gives higher PSNR than the classical base [PITH_FULL_IMAGE:figures/full_fig_p025_14.png]
Figure 15
Figure 15. Figure 15: Sigsbee2A salt benchmark. Rows show the smooth initial model, conventional [PITH_FULL_IMAGE:figures/full_fig_p026_15.png]
Figure 16
Figure 16. Figure 16: Six-stage Sigsbee2A SGDS workflow. From top to bottom, the panels show the smooth initial model, the [PITH_FULL_IMAGE:figures/full_fig_p027_16.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

26 extracted references · 1 canonical work pages

  1. [1]

    Diffusion posterior sampling for general noisy inverse problems.arXiv preprint arXiv:2209.14687, 2022

    Hyungjin Chung, Jeongsol Kim, Michael T Mccann, Marc L Klasky, and Jong Chul Ye. Diffusion posterior sampling for general noisy inverse problems.arXiv preprint arXiv:2209.14687, 2022

    arXiv 2022

  2. [2]

    Tweedie’s formula and selection bias.Journal of the American Statistical Association, 106(496): 1602–1614, 2011

    Bradley Efron. Tweedie’s formula and selection bias.Journal of the American Statistical Association, 106(496): 1602–1614, 2011

    2011

  3. [3]

    Regularization of inverse problems

    Heinz W Engl and Ronny Ramlau. Regularization of inverse problems. InEncyclopedia of applied and compu- tational mathematics, pages 1233–1241. Springer, 2015

    2015

  4. [4]

    Manifold preserving guided diffusion.arXiv preprint arXiv:2311.16424, 2023

    Yutong He, Naoki Murata, Chieh-Hsin Lai, Yuhta Takida, Toshimitsu Uesaka, Dongjun Kim, Wei-Hsiang Liao, Yuki Mitsufuji, J Zico Kolter, Ruslan Salakhutdinov, et al. Manifold preserving guided diffusion.arXiv preprint arXiv:2311.16424, 2023

    arXiv 2023

  5. [5]

    Denoising diffusion probabilistic models.Advances in neural information processing systems, 33:6840–6851, 2020

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising diffusion probabilistic models.Advances in neural information processing systems, 33:6840–6851, 2020

    2020

  6. [6]

    Training-image based geostatistical inversion using a spatial generative adversarial neural network.Water Resources Research, 54(1):381–406, 2018

    Eric Laloy, Romain Hérault, Diederik Jacques, and Niklas Linde. Training-image based geostatistical inversion using a spatial generative adversarial neural network.Water Resources Research, 54(1):381–406, 2018

    2018

  7. [7]

    Robust joint adaptive multiparameter waveform inversion with attenuation compensation in viscoacoustic media.Geophysics, 89(3):R231–R246, 2024

    Chao Li, Guochang Liu, Fang Li, and Zhiyong Wang. Robust joint adaptive multiparameter waveform inversion with attenuation compensation in viscoacoustic media.Geophysics, 89(3):R231–R246, 2024

    2024

  8. [8]

    Dlm-fwi: Deep learning matching filtering for full waveform inversion.Geophysical Journal International, page ggag019, 2026

    Chao Li, Sergey Fomel, and Yangkang Chen. Dlm-fwi: Deep learning matching filtering for full waveform inversion.Geophysical Journal International, page ggag019, 2026

    2026

Show all 26 references
  1. [9]

    Deep learning inversion of seismic data.IEEE Transactions on Geoscience and Remote Sensing, 58(3):2135–2149, 2020

    Shucai Li, Bin Liu, Yuxiao Ren, Yangkang Chen, Senlin Yang, Yunhai Wang, and Peng Jiang. Deep learning inversion of seismic data.IEEE Transactions on Geoscience and Remote Sensing, 58(3):2135–2149, 2020. doi: 10.1109/TGRS.2019.2953473

    2020

  2. [10]

    Deep learning seismic full wave- form inversion for realistic structure models.Geophysics, 86(1):R31–R44, 2021

    Bin Liu, Senlin Yang, Yuxiao Ren, Xinji Xu, Peng Jiang, and Yangkang Chen. Deep learning seismic full wave- form inversion for realistic structure models.Geophysics, 86(1):R31–R44, 2021. doi: 10.1190/geo2019-0435.1

    2021 doi

  3. [11]

    Repaint: Inpainting using denoising diffusion probabilistic models

    Andreas Lugmayr, Martin Danelljan, Andres Romero, Fisher Yu, Radu Timofte, and Luc Van Gool. Repaint: Inpainting using denoising diffusion probabilistic models. InProceedings of the IEEE/CVF conference on com- puter vision and pattern recognition, pages 11461–11471, 2022

    2022

  4. [12]

    Sdedit: Guided image synthesis and editing with stochastic differential equations.arXiv preprint arXiv:2108.01073, 2021

    Chenlin Meng, Yutong He, Yang Song, Jiaming Song, Jiajun Wu, Jun-Yan Zhu, and Stefano Ermon. Sdedit: Guided image synthesis and editing with stochastic differential equations.arXiv preprint arXiv:2108.01073, 2021

    2021 arXiv

  5. [13]

    Stochastic seismic waveform inversion using generative adversarial networks as a geological prior.Mathematical Geosciences, 52(1):53–79, 2020

    Lukas Mosser, Olivier Dubrule, and Martin J Blunt. Stochastic seismic waveform inversion using generative adversarial networks as a geological prior.Mathematical Geosciences, 52(1):53–79, 2020

    2020

  6. [14]

    Geophysical inverse problems with measurement-guided diffusion models.arXiv preprint arXiv:2501.04881, 2025

    Matteo Ravasi. Geophysical inverse problems with measurement-guided diffusion models.arXiv preprint arXiv:2501.04881, 2025

    2025 arXiv

  7. [15]

    Building complex seismic velocity models for deep learning inversion.IEEE Access, 9:63767–63778, 2021

    Yuxiao Ren, Lichao Nie, Senlin Yang, Peng Jiang, and Yangkang Chen. Building complex seismic velocity models for deep learning inversion.IEEE Access, 9:63767–63778, 2021

    2021

  8. [16]

    Nonlinear total variation based noise removal algorithms

    Leonid I Rudin, Stanley Osher, and Emad Fatemi. Nonlinear total variation based noise removal algorithms. Physica D: nonlinear phenomena, 60(1-4):259–268, 1992

    1992

  9. [17]

    Generative modeling by estimating gradients of the data distribution.Advances in neural information processing systems, 32, 2019

    Yang Song and Stefano Ermon. Generative modeling by estimating gradients of the data distribution.Advances in neural information processing systems, 32, 2019

    2019

  10. [18]

    Score- based generative modeling through stochastic differential equations.arXiv preprint arXiv:2011.13456, 2020

    Yang Song, Jascha Sohl-Dickstein, Diederik P Kingma, Abhishek Kumar, Stefano Ermon, and Ben Poole. Score- based generative modeling through stochastic differential equations.arXiv preprint arXiv:2011.13456, 2020. Diffusion-guided optimization for full waveform inversion32

    2011 arXiv

  11. [19]

    Inversion of seismic reflection data in the acoustic approximation.Geophysics, 49(8):1259– 1266, 1984

    Albert Tarantola. Inversion of seismic reflection data in the acoustic approximation.Geophysics, 49(8):1259– 1266, 1984

    1984

  12. [20]

    Mohammad H Taufik, Fu Wang, and Tariq Alkhalifah. Learned regularizations for multi-parameter elastic full waveform inversion using diffusion models.Journal of Geophysical Research: Machine Learning and Compu- tation, 1(1):e2024JH000125, 2024

    2024

  13. [21]

    An overview of full-waveform inversion in exploration geophysics

    Jean Virieux and Stéphane Operto. An overview of full-waveform inversion in exploration geophysics. 2010

    2010

  14. [22]

    Split-and-augmented gibbs sampler—application to large-scale inference problems.IEEE Transactions on Signal Processing, 67(6):1648–1661, 2019

    Maxime V ono, Nicolas Dobigeon, and Pierre Chainais. Split-and-augmented gibbs sampler—application to large-scale inference problems.IEEE Transactions on Signal Processing, 67(6):1648–1661, 2019

    2019

  15. [23]

    A prior regularized full waveform inversion using generative diffusion models.IEEE transactions on geoscience and remote sensing, 61:1–11, 2023

    Fu Wang, Xinquan Huang, and Tariq A Alkhalifah. A prior regularized full waveform inversion using generative diffusion models.IEEE transactions on geoscience and remote sensing, 61:1–11, 2023

    2023

  16. [24]

    Controllable seismic velocity synthesis using generative dif- fusion models.Journal of Geophysical Research: Machine Learning and Computation, 1(3):e2024JH000153, 2024

    Fu Wang, Xinquan Huang, and Tariq Alkhalifah. Controllable seismic velocity synthesis using generative dif- fusion models.Journal of Geophysical Research: Machine Learning and Computation, 1(3):e2024JH000153, 2024

    2024

  17. [25]

    Seisfusion: Constrained diffusion model with input guidance for 3d seismic data interpolation and reconstruction.IEEE Transactions on Geoscience and Remote Sensing, 2024

    Shuang Wang, Fei Deng, Peifan Jiang, Zishan Gong, Xiaolin Wei, and Yuqing Wang. Seisfusion: Constrained diffusion model with input guidance for 3d seismic data interpolation and reconstruction.IEEE Transactions on Geoscience and Remote Sensing, 2024

    2024

  18. [26]

    Cle diffusion: Controllable light enhancement diffusion model

    Yuyang Yin, Dejia Xu, Chuangchuang Tan, Ping Liu, Yao Zhao, and Yunchao Wei. Cle diffusion: Controllable light enhancement diffusion model. InProceedings of the 31st ACM International Conference on Multimedia, pages 8145–8156, 2023

    2023

Pith tools

Reviewed August 1, 2026 · model on record in the stance chip above.