Pith. sign in

REVIEW 3 major objections 5 minor 59 references

Foundation Model-Assisted Full Waveform Inversion

T0 review · 3 major / 5 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 The paper claims that comparing modeled and observed seismic traces in the frozen feature space of a pretrained seismic foundation model yields a full-waveform-inversion objective with a broader basin of attraction, mitigating cycle…

desk verdict A genuinely new way to build an FWI misfit from a frozen pretrained seismic encoder, with solid experiments and one missing control that a referee should request. read the letter →

arxiv 2608.05763 v1 pith:WEGQI4IT submitted 2026-08-06 physics.geo-ph

Mustafa Alfarhan , Matteo Ravasi , Fuqiang Chen , George Turkiyyah , David Keyes This is my paper · ORCID
classification physics.geo-ph
keywords fullwaveforminversioncycleskippingseismicfoundationmodelfeature-spacemisfitadjoint-statemethodpretrainedencoderSeisLMvelocitybuilding
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

Full waveform inversion (FWI) recovers subsurface velocity models by matching modeled to observed seismic data, but the conventional $L_2$ waveform misfit is trapped by cycle skipping when the starting model is inaccurate. This paper proposes replacing that comparison, during early-stage inversion, with a discrepancy in the feature space of SeisLM, a seismic foundation model pretrained on unlabeled earthquake waveforms; the frozen encoder maps standardized modeled and observed traces to contextualized feature vectors, and the squared difference of those features serves as the misfit. Time-shift diagnostics show the pretrained feature loss has a broader and smoother basin around the correct alignment than either the $L_2$ objective or the feature loss of a randomly initialized encoder, and the layerwise diagnostics locate the widened basin in the pretrained transformer blocks. In Marmousi and 2D/3D Overthrust tests, this feature-space misfit produces starting velocity models from which conventional $L_2$ refinement succeeds, whereas $L_2$ alone stalls or converges to inaccurate models. The paper's intended result is that pretrained seismic representations are useful as early-stage FWI objectives, not as complete replacements for waveform-domain misfits.

What carries the argument

The central object is the SeisLM encoder, a masked-prediction transformer pretrained on unlabeled three-component earthquake waveforms and kept frozen during inversion. Its input is a per-trace standardized single-component trace replicated across three channels, and its output is a sequence of 240-dimensional contextualized feature vectors, roughly one quarter as many as the input time samples. The misfit is $J_{\mathrm{SeisLM}}(m)=\tfrac12\sum_s\sum_r\|F_{\hat\theta}(\tilde d_{s,r}(m))-F_{\hat\theta}(\tilde d^{\mathrm{obs}}_{s,r})\|_F^2$, and the adjoint source is the chain-rule product of the per-trace standardization Jacobian, the channel-replication transpose, and the encoder Jacobian applied to the feature residual. This construction is what lets a frozen, FWI-agnostic network define a differentiable misfit without training or fine-tuning; the empirical mechanism is that the pretrained transformer features increase monotonically over a much wider range of time shifts than the data-space $L_2$ distance, while a randomly initialized encoder of the same architecture does not.

What would settle it

Run the 2D Overthrust stage-1 workflow to 100 iterations with three random seeds, replacing the frozen pretrained SeisLM encoder with a randomly initialized encoder of identical architecture; if the relative velocity error of the random-encoder workflow is equal to or better than the pretrained one, the claim that pretraining, rather than architecture alone, creates the broadened basin is falsified.

Watch

Extended reading notes

Core claim

The central claim is that an FWI objective defined in the frozen feature space of a pretrained seismic foundation model has a wider basin of attraction than the conventional waveform $L_2$ objective, and that this is enough to make cycle-skipping-prone $L_2$ refinement succeed afterwards. The SeisLM encoder, pretrained by masked prediction on three-component earthquake waveforms, is kept frozen; each single-component acoustic trace is standardized over time and replicated across three channels, and the misfit is the squared Frobenius norm of the difference between modeled and observed feature matrices. Because the encoder is differentiable, the feature residual backpropagates through the standardization, replication, and encoder into an adjoint source that plugs directly into the standard adjoint-state FWI workflow. The paper reports that the pretrained feature loss produces a broader and smoother basin than both $L_2$ and the feature loss of a randomly initialized network, that in Marmousi the feature-loss and hybrid workflows give comparable improvements over $L_2$, that in 2D Overthrust the pure feature loss outperforms both $L_2$ and the hybrid, and that in 3D Overthrust $L_2$ stalls near a linear-gradient start while the feature loss builds a background from which $L_2$ refinement recovers the main structures. The intended role of the feature loss is therefore to steer early-stage inversion into the basin of attraction of the waveform misfit, not to replace the waveform misfit entirely.

Load-bearing premise

The load-bearing premise is that SeisLM's encoder, pretrained on 30-second three-component earthquake waveforms at 100 Hz, still represents shorter, single-component, acoustic, exploration-frequency traces in a way that avoids the cycle-skipping trap after per-trace standardization and channel replication; if that transfer fails, the feature loss has no advantage over a randomly initialized encoder.

Editorial extensions

If this is right

  • Because the feature loss backpropagates into an adjoint source, a frozen pretrained encoder can be dropped into an existing adjoint-state FWI workflow without modifying the wave-equation solver or training a network.
  • The 2D Overthrust hybrid result implies that adding $L_2$ too early can undo the feature loss's protection, so any successful schedule must be conditioned on the current model's accuracy rather than fixed at 100 iterations.
  • The layerwise shift diagnostics imply that the widened basin lives in the deeper pretrained transformer blocks; using early convolutional features alone would likely retain the oscillatory behavior of $L_2$.
  • The 3D experiment indicates that temporally averaged feature vectors retain enough information to guide inversion, although the paper does not directly compare full-sequence versus averaged features.
  • The measured cost table implies that encoder forward and backward passes dominate the added overhead, and bfloat16 roughly halves full-gradient time in 3D without changing the model-error trajectory.

Reading between the lines

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

  • A testable extension the paper leaves implicit is an adaptive hand-off rule: monitor the agreement between feature-space and waveform gradients, or the slope of the feature loss, and switch to $L_2$ only once the two objectives point in similar directions.
  • Because per-trace standardization removes absolute amplitudes, the feature loss probably sacrifices legitimate amplitude information; it should therefore be most reliable for kinematics-dominated early stages and less suitable for recovering amplitude-bearing parameters such as density or attenuation.
  • The 3D experiment's temporal averaging suggests the robust signal is distributed across the feature sequence rather than locked to individual arrivals; directly comparing full-sequence and averaged feature losses on the affordable 2D Overthrust geometry would test this.
  • If the cross-domain transfer from earthquake to exploration waveforms is as robust as these benchmarks suggest, the same frozen encoder could serve as a generic misfit for other seismic time-series inverse problems, though the paper does not test these settings.
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

3 major / 5 minor

Summary. The paper proposes a feature-space misfit for full waveform inversion (FWI) in which modeled and observed seismic traces are compared through the frozen encoder of SeisLM, a pretrained seismic foundation model. The SeisLM feature loss is differentiated with respect to the modeled data to obtain an adjoint source compatible with the standard adjoint-state framework, including the derivatives of per-trace standardization and the channel-replication operator. The authors also test a scheduled hybrid of the feature loss and the conventional L2 objective. Time-shift diagnostics in a 1D toy example and a layerwise sensitivity analysis on Marmousi data indicate that the pretrained encoder produces a broader, smoother misfit basin than both the raw L2 objective and an encoder with the same architecture but randomly initialized weights. Two-stage FWI experiments on Marmousi and 2D/3D SEG/EAGE Overthrust models compare L2, SeisLM-feature, and hybrid workflows, reporting that the SeisLM objective improves early-stage background-model recovery and provides better starting models for subsequent L2 refinement, while the hybrid works well in Marmousi but less well in 2D Overthrust.

Significance. If the central claim holds, the paper makes a useful contribution: it shows that a frozen, self-supervised seismic foundation model pretrained on earthquake waveforms can define an FWI misfit that mitigates cycle skipping without introducing trainable parameters during inversion. The adjoint-source derivation in Section 2.4 is presented carefully and is technically sound, and the paper includes reproducible public code and data availability, a clear computational-cost table, and honest discussion of limitations. The strongest evidence for the pretraining attribution is the 1D time-shift landscape and the layerwise sensitivity profile, both of which include a randomly initialized encoder control. The main weakness is that this control is not carried into the actual FWI benchmarks, so the benchmark improvements are not directly shown to require the pretrained weights rather than the architecture and preprocessing. The 3D experiment also alters the objective by temporal averaging without a direct comparison to the full sequence-level loss.

major comments (3)
  1. [§3.1–3.3] The central attribution of the FWI improvements to the pretrained encoder is not tested in the actual inversion benchmarks. The only pretrained-versus-random comparisons are the 1D time-shift landscape (Figure 3) and the layerwise sensitivity profile (Figure 4) in Section 3.1; the Marmousi and 2D/3D Overthrust inversions in Sections 3.2–3.3 compare L2, SeisLM, and hybrid only. Because the optimization path in model space can interact with a feature landscape differently from a data-space shift sweep, the benchmark gains could in principle be produced by the architecture and preprocessing alone rather than by SeisLM's pretrained weights. Please run the same stage-1 FWI with a randomly initialized encoder of identical architecture, at least for Marmousi and 2D Overthrust and ideally for the 3D test, or explicitly limit the pretraining-attribution claim to the Section 3.1 diagnostics.
  2. [§3.3.2] The 3D Overthrust experiment uses a modified objective: the last-layer feature matrices are averaged over time before comparison, whereas Equation (3) and the 2D experiments use the full feature sequence. No comparison between the full and averaged representations is provided, so the 3D result establishes a property of the temporally averaged feature loss, not of the sequence-level objective used elsewhere. The Discussion concedes this point ('the 2D and 3D results do not provide a direct comparison'). To support the claim that the benefit extends to 3D with the original objective, add at least a 2D experiment comparing full versus averaged feature losses, or a 3D comparison on a subset of shots.
  3. [§3.1, Figs. 3–4] The random-encoder control is based on a single random initialization, and the text repeatedly qualifies it as 'the randomly initialized encoder tested here.' Since this control is load-bearing for the claim that pretraining rather than architecture creates the smooth basin, one initialization is insufficient to establish the general contrast with randomly initialized parameters. Report the range or distribution over several seeds, for example the mean and min/max envelope, for both the time-shift and layerwise diagnostics.
minor comments (5)
  1. [§2.2 and §3.3.2] Equation (3) defines the feature loss for the full feature sequence, but the 3D experiment uses a temporally averaged variant that is never written out; please add an explicit definition of the averaged loss and state where it replaces Equation (3).
  2. [References] The citation 'A. Brougois et al. [1990]' for the Marmousi model appears to contain a misspelled author name; please verify the canonical author list.
  3. [§3.1] The 1D diagnostic is a useful controlled test, but it would be clearer to specify whether the raw L2 misfit is computed on standardized traces or on the raw traces, since the feature loss always includes standardization; if the L2 baseline is unstandardized, state this explicitly to avoid ambiguity in the comparison.
  4. [Figures 7, 10, 14] The loss curves are normalized per-stage and include line-search trial evaluations, while velocity-error curves include only accepted iterates; this is explained in the text, but one sentence in each caption noting this distinction would help readers avoid misreading curve lengths.
  5. [§4] The claim that the feature loss 'changes only the data-comparison term' is accurate, but the 3D temporal averaging is an additional change to the objective; please make clear in this section that the 3D comparison is specifically for the averaged variant.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the frozen SeisLM encoder is pretrained on external earthquake data, no parameters are fit to the FWI benchmarks, and the central claims are evaluated against L2 and random-encoder controls.

full rationale

The paper's derivation chain is self-contained and non-circular. The SeisLM feature loss in Eq. (3) uses a frozen encoder pretrained on SeisBench earthquake waveforms (Liu et al. 2024), with no FWI-specific training or fine-tuning. The adjoint source in Eq. (9) is obtained by backpropagating the feature residual through this fixed encoder, and the only optimized variables are the velocity-model parameters. No parameter in the misfit is fitted to the Marmousi or Overthrust data; the hybrid schedule in Eq. (6) is hand-chosen, and its limitations are reported honestly, including its worse performance in the 2D Overthrust experiment. The time-shift diagnostics directly compare the pretrained encoder against a randomly initialized encoder of identical architecture, providing a control for the claim that pretraining, rather than architecture alone, produces the smoother misfit landscape. The absence of a random-encoder control in the full benchmark inversions is a legitimate correctness concern about the strength of the empirical attribution, but it is not a circularity: the benchmark results are not defined in terms of the diagnostic claims. The paper also explicitly flags domain-shift, noise, source-wavelet uncertainty, and temporal-averaging limitations, further indicating that the results are presented as empirical findings rather than as consequences of the method's construction. Citations to SeisLM, Wav2Vec2, and perceptual-loss literature are external and not self-referential chains that force the conclusion. No equation or fitted quantity is renamed as a prediction, and no uniqueness theorem or self-citation is used to forbid alternatives. The central claim therefore has independent empirical content and does not reduce to its own inputs.

Assumptions & free parameters 4 free parameters · 6 assumptions · 0 invented entities

The central method is an application of an existing frozen encoder, so the ledger is light. The hand-chosen elements are the hybrid schedule, the 3D temporal averaging, the unstated epsilon in trace standardization, and the stage-1 mask. The main axioms are the standard adjoint-state derivation plus the domain-transfer assumption that earthquake-pretrained features remain useful for acoustic FWI traces.

free parameters (4)
  • alpha_i hybrid schedule = min(1, i/100)
    The linear 100-iteration schedule for the L2 weight is chosen by hand, not derived. Its poor result in 2D Overthrust shows that the schedule itself is a tunable component affecting the method's success.
  • temporal averaging in 3D = mean over temporal feature positions
    Introduced to cut memory for caching observed features in 3D. This changes the loss and is never compared against the full feature-matrix loss, so its effect on the objective landscape is unquantified.
  • stabilization constant epsilon in per-trace standardization = not specified
    The value of epsilon in N(d) = (d - mu)/max(sigma, epsilon) is not reported, yet it enters the adjoint source through the derivative of N.
  • stage-1 time-space mask = direct-arrival-time based cutoff
    The mask suppressing reflections is constructed by hand for each acquisition geometry. All objectives share the same mask, so it does not bias cross-objective comparison, but it is part of the workflow.
assumptions (6)
  • domain assumption Acoustic constant-density wave equation is an adequate forward model for the synthetic FWI experiments.
    Section 3 states that Devito acoustic finite-difference modeling is used; all conclusions are restricted to acoustic synthetic data.
  • standard math Adjoint-state method with L(m)^T lambda = P^T dJ/dd gives the correct model gradient.
    Section 2.4 invokes this standard result without proof.
  • domain assumption SeisLM's pretrained representations transfer from 30 s, 100 Hz, three-component earthquake waveforms to shorter, single-component acoustic FWI traces.
    Section 2.1 notes the domain shift and says the numerical examples test whether the representations remain useful; the entire method depends on this transfer.
  • domain assumption Replicating a single-component trace across three channels is a valid input mapping for SeisLM.
    Section 2.2 introduces the operator R and acknowledges that alternative mappings would shift the input distribution; the choice is heuristic.
  • domain assumption The randomly initialized encoder is a sufficient control for architecture effects.
    Section 3.1 uses it to attribute smoothness to pretraining; this assumes the random-init features are a comparable baseline.
  • domain assumption Monotonic increase of feature distance over time shifts implies a useful basin for FWI.
    Section 3.1 interprets the 1D and layerwise diagnostics as evidence of cycle-skipping mitigation; this is an inferential step from a toy diagnostic to full FWI behavior.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Foundation Model-Assisted Full Waveform Inversion." pith.science (2026). https://pith.science/paper/WEGQI4IT

@misc{pith2026260805763,
  author       = {Pith},
  title        = {Pith review of: Foundation Model-Assisted Full Waveform Inversion},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/WEGQI4IT}},
  note         = {Machine review of arXiv:2608.05763}
}
abstract

Full waveform inversion (FWI) can recover high-resolution subsurface velocity models. Conventional waveform-difference objectives, however, are vulnerable to cycle skipping when the starting model is inaccurate. We introduce an FWI objective that compares features produced from modeled and observed seismic traces by SeisLM, a pretrained seismic foundation model. The SeisLM encoder remains frozen during inversion, and the feature discrepancy is differentiated with respect to the modeled traces to construct an adjoint source compatible with the standard adjoint-state framework. We also test a scheduled hybrid loss that combines the SeisLM feature loss with the conventional $L_2$ objective. Time-shift diagnostics show that the loss computed from features produced by the pretrained encoder has a broader and smoother basin around the correct alignment than either the waveform $L_2$ objective or the feature loss obtained from an encoder with the same architecture and randomly initialized parameters. In the Marmousi experiment, the SeisLM and hybrid objectives produce similar improvements during early-stage inversion and provide useful models for subsequent reflection-based $L_2$ refinement. In the 2D Overthrust experiment, which begins from a laterally invariant linear-gradient model, the SeisLM feature-loss workflow outperforms the conventional and hybrid workflows, indicating that introducing the $L_2$ contribution too early can reintroduce cycle-skipping sensitivity. In the 3D Overthrust experiment, conventional $L_2$ inversion stalls near the initial linear gradient, whereas the SeisLM feature loss guides the inversion toward a background model from which $L_2$ refinement recovers the principal structures. These results support using features produced by pretrained seismic networks to define early-stage FWI objectives rather than complete replacements for waveform-domain misfits.

Figures

Figures reproduced from arXiv: 2608.05763 by the authors.

Figure 1
Figure 1. SeisLM-base encoder used as a frozen feature extractor. Two stride-2 convolutional layers reduce the input [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. FWI workflow using SeisLM as a frozen feature-space objective. The feature discrepancy defines [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. 1D example showing sensitivity to two independent time shifts. Top: reference and shifted two-wavelet traces. [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (12 more)
Figure 4
Figure 4. Figure 4: Sensitivity of the SeisLM feature distance to a rigid time shift, resolved by encoder depth, for (a) the pretrained [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]
Figure 5
Figure 5. Figure 5: The Marmousi model: (a) true velocity model and (b) smoothed initial velocity model, with sources and [PITH_FULL_IMAGE:figures/full_fig_p009_5.png]
Figure 6
Figure 6. Figure 6: First-iteration FWI gradients obtained using (a) the conventional [PITH_FULL_IMAGE:figures/full_fig_p010_6.png]
Figure 7
Figure 7. Figure 7: Performance of the Marmousi example. (a) Normalized loss and (b) relative velocity error. Solid curves [PITH_FULL_IMAGE:figures/full_fig_p010_7.png]
Figure 8
Figure 8. Figure 8: Inverted velocity models for the Marmousi example. Left column: stage-1 results obtained using (a) the [PITH_FULL_IMAGE:figures/full_fig_p011_8.png]
Figure 9
Figure 9. Figure 9: The 2D Overthrust model with a 500 m water column: (a) true velocity model and (b) laterally invariant [PITH_FULL_IMAGE:figures/full_fig_p012_9.png]
Figure 10
Figure 10. Figure 10: Performance of the 2D Overthrust experiment. (a) Normalized objective values and (b) relative velocity [PITH_FULL_IMAGE:figures/full_fig_p012_10.png]
Figure 11
Figure 11. Figure 11: Inverted models for the 2D Overthrust experiment. Left column: stage-1 results obtained using (a) the [PITH_FULL_IMAGE:figures/full_fig_p013_11.png]
Figure 12
Figure 12. Figure 12: The selected 3D Overthrust volume before adding the 500 m water layer, together with the surface acquisition [PITH_FULL_IMAGE:figures/full_fig_p014_12.png]
Figure 13
Figure 13. Figure 13: Slices through the 3D Overthrust model: horizontal (top left), inline–depth (bottom left), and crossline–depth [PITH_FULL_IMAGE:figures/full_fig_p014_13.png]
Figure 14
Figure 14. Figure 14: 3D Overthrust convergence: (a) stage-normalized objectives and (b) relative velocity error. The [PITH_FULL_IMAGE:figures/full_fig_p015_14.png]
Figure 15
Figure 15. Figure 15: Final results of the 3D Overthrust workflows: (a) conventional [PITH_FULL_IMAGE:figures/full_fig_p015_15.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

59 extracted references · 47 canonical work pages

  1. [1]

    Geophysics , volume =

    Tarantola, Albert , title =. Geophysics , volume =. 1984 , month =

    work page 1984

  2. [2]

    Geophysics , volume=

    An overview of full-waveform inversion in exploration geophysics , author=. Geophysics , volume=. 2009 , publisher=

    2009

  3. [3]

    Geophysics , volume=

    Wave-equation traveltime inversion , author=. Geophysics , volume=. 1991 , doi=

    work page 1991

  4. [4]

    and Mulder, W

    Van Leeuwen, T. and Mulder, W. A. , title =. Geophysical Journal International , volume =. doi:10.1111/j.1365-246X.2010.04681.x , year =

    arXiv 2010

  5. [5]

    Geophysics , volume=

    Adaptive waveform inversion: Theory , author=. Geophysics , volume=. 2016 , doi=

    work page 2016

  6. [6]

    Journal of Applied Geophysics , volume =

    Full waveform inversion method using envelope objective function without low frequency data , author =. Journal of Applied Geophysics , volume =. 2014 , doi =

    work page 2014

  7. [7]

    Geophysics , volume=

    Seismic envelope inversion and modulation signal model , author=. Geophysics , volume=. 2014 , publisher=

    2014

  8. [8]

    Misfit functions for full waveform inversion based on instantaneous phase and envelope measurements , journal=

    Bozda. Misfit functions for full waveform inversion based on instantaneous phase and envelope measurements , journal=. 2011 , doi=

    work page 2011

Show all 59 references
  1. [9]

    Communications in Mathematical Sciences , volume=

    Optimal transport for seismic full waveform inversion , author=. Communications in Mathematical Sciences , volume=. 2016 , doi=

    2016

  2. [10]

    Geophysical Journal International , volume=

    Measuring the misfit between seismograms using an optimal transport distance: Application to full waveform inversion , author=. Geophysical Journal International , volume=. 2016 , doi=

    2016

  3. [11]

    Application of optimal transport and the quadratic

    Yang, Yunan and Engquist, Bj. Application of optimal transport and the quadratic. Geophysics , volume=. 2018 , doi=

    2018

  4. [12]

    Long-wavelength

    Ramos-Mart. Long-wavelength. The Leading Edge , volume=. 2019 , doi=

    2019

  5. [13]

    Geophysics , volume=

    Wave-equation reflection traveltime inversion with dynamic warping and full-waveform inversion , author=. Geophysics , volume=. 2013 , doi=

    2013

  6. [14]

    Geophysics , volume=

    Cycle-skipping mitigation using misfit measurements based on differentiable dynamic time warping , author=. Geophysics , volume=. 2022 , doi=

    2022

  7. [15]

    Kalita, Mahesh and Purcell, Chris and Casasanta, Lorenzo , title =. 84th. 2023 , doi =

    2023

  8. [16]

    On cycle-skipping and misfit function modification for full-wave inversion: Comparison of five recent approaches , journal =

    Pladys, Arnaud and Brossier, Romain and Li, Yubing and M. On cycle-skipping and misfit function modification for full-wave inversion: Comparison of five recent approaches , journal =. 2021 , doi =

    2021

  9. [17]

    Geophysical Journal International , volume=

    Mitigating local minima in full-waveform inversion by expanding the search space , author=. Geophysical Journal International , volume=. 2013 , doi=

    2013

  10. [18]

    Geophysics , volume=

    Improving full-waveform inversion by wavefield reconstruction with the alternating direction method of multipliers , author=. Geophysics , volume=. 2019 , doi=

    2019

  11. [19]

    Geophysics , volume=

    Efficient waveform inversion and imaging: A strategy for selecting temporal frequencies , author=. Geophysics , volume=. 2004 , publisher=

    2004

  12. [20]

    68th EAGE Conference and Exhibition incorporating SPE EUROPEC 2006 , year=

    The importance of low frequency and large offset in waveform inversion , author=. 68th EAGE Conference and Exhibition incorporating SPE EUROPEC 2006 , year=. doi:10.3997/2214-4609.201402146 , publisher=

    2006

  13. [21]

    The Leading Edge , volume=

    Wang, Ping and Zhang, Zhigang and Mei, Jiawei and Lin, Feng and Huang, Rongxin , title=. The Leading Edge , volume=. 2019 , doi=

    2019

  14. [22]

    Full-waveform inversion for imaging and geologic interpretation: A deepwater

    Cobo, Yannick and Calder. Full-waveform inversion for imaging and geologic interpretation: A deepwater. Interpretation , volume=. 2019 , doi=

    2019

  15. [23]

    Fifth International Meeting for Applied Geoscience and Energy Expanded Abstracts , pages=

    Ji, Shuo and Yang, Zhiping and De Marco Centeno, Ricardo and Zhu, Huifeng and Wray, Brad , title=. Fifth International Meeting for Applied Geoscience and Energy Expanded Abstracts , pages=. 2025 , doi=

    2025

  16. [24]

    IEEE Transactions on Geoscience and Remote Sensing , volume=

    Sun, Hongyu and Demanet, Laurent , title=. IEEE Transactions on Geoscience and Remote Sensing , volume=. 2022 , doi=

    2022

  17. [25]

    2020 , volume=

    Wu, Yue and Lin, Youzuo , journal=. 2020 , volume=

    2020

  18. [26]

    Geophysics , volume=

    Deep-learning inversion: A next-generation seismic velocity model building method , author=. Geophysics , volume=. 2019 , doi=

    2019

  19. [27]

    and Hampson, G

    Kumar, A. and Hampson, G. and Rayment, T. and Burgess, T. , title =. 83rd EAGE Annual Conference and Exhibition , volume =. 2022 , doi =

    2022

  20. [28]

    Geophysics , volume=

    Deep learning-based point-spread function deconvolution for migration image deblurring , author=. Geophysics , volume=. 2022 , doi=

    2022

  21. [29]

    arXiv preprint arXiv:2603.00377 , year=

    Improving full waveform inversion in large model era , author=. arXiv preprint arXiv:2603.00377 , year=

    arXiv

  22. [30]

    Frontiers in Earth Science , volume=

    Sun, Bingbing and Alkhalifah, Tariq , title=. Frontiers in Earth Science , volume=. 2022 , doi=

    2022

  23. [31]

    Journal of Geophysical Research: Solid Earth , volume=

    Yang, Fangshu and Ma, Jianwei , title=. Journal of Geophysical Research: Solid Earth , volume=. 2023 , doi=

    2023

  24. [32]

    Mostafa and Li, Yue , title=

    Wang, Cong and Huang, Xingguo and Mousavi, S. Mostafa and Li, Yue , title=. Geophysical Journal International , volume=. 2026 , doi=

    2026

  25. [33]

    and Harsuko, Randy and Alkhalifah, Tariq , journal=

    Saad, Omar M. and Harsuko, Randy and Alkhalifah, Tariq , journal=. 2024 , doi=

    2024

  26. [34]

    International Journal of Pattern Recognition and Artificial Intelligence , volume=

    Signature verification using a ``Siamese'' time delay neural network , author=. International Journal of Pattern Recognition and Artificial Intelligence , volume=. 1993 , doi=

    1993

  27. [35]

    and Alkhalifah, Tariq , journal=

    Saad, Omar M. and Alkhalifah, Tariq , journal=. 2025 , doi=

    2025

  28. [36]

    Geosciences , volume=

    Li, Chao and Chen, Yangkang , title=. Geosciences , volume=. 2026 , doi=

    2026

  29. [37]

    2019 , address=

    Devlin, Jacob and Chang, Ming-Wei and Lee, Kenton and Toutanova, Kristina , booktitle=. 2019 , address=. doi:10.18653/v1/N19-1423 , url=

    2019 doi

  30. [38]

    OpenAI Blog , volume=

    Language models are unsupervised multitask learners , author=. OpenAI Blog , volume=

  31. [39]

    2020 , publisher=

    Baevski, Alexei and Zhou, Henry and Mohamed, Abdelrahman and Auli, Michael , booktitle=. 2020 , publisher=

    2020

  32. [40]

    2021 , doi=

    Hsu, Wei-Ning and Bolte, Benjamin and Tsai, Yao-Hung Hubert and Lakhotia, Kushal and Salakhutdinov, Ruslan and Mohamed, Abdelrahman , journal=. 2021 , doi=

    2021

  33. [41]

    2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , pages=

    Masked autoencoders are scalable vision learners , author=. 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , pages=. 2022 , doi=

    2022

  34. [42]

    , journal=

    Harsuko, Randy and Alkhalifah, Tariq A. , journal=. 2022 , doi=

    2022

  35. [43]

    Geophysics , volume=

    Seismic foundation model: A next generation deep-learning model in geophysics , author=. Geophysics , volume=. 2025 , doi=

    2025

  36. [44]

    2603.23211 , archivePrefix=

    Ordonez, Alba and Forgaard, Theodor Johannes Line and Wade, David and Bugge, Aina Juell and Nese, Hakon and Waldeland, Anders Ueland , year=. 2603.23211 , archivePrefix=

  37. [45]

    2024 , doi=

    Si, Xu and Wu, Xinming and Sheng, Hanlin and Zhu, Jun and Li, Zefeng , journal=. 2024 , doi=

    2024

  38. [46]

    IEEE Transactions on Geoscience and Remote Sensing , volume=

    Li, Sen and Yang, Xu and Cao, Anye and Wang, Changbin and Liu, Yaoqi and Liu, Yapeng and Niu, Qiang , title=. IEEE Transactions on Geoscience and Remote Sensing , volume=. 2024 , doi=

    2024

  39. [47]

    arXiv preprint arXiv:2410.15765 , year=

    Liu, Tianlin and M. arXiv preprint arXiv:2410.15765 , year=

    arXiv

  40. [48]

    Computer Vision -- ECCV 2016 , pages=

    Perceptual losses for real-time style transfer and super-resolution , author=. Computer Vision -- ECCV 2016 , pages=. 2016 , publisher=

    2016

  41. [49]

    2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) , pages=

    Image style transfer using convolutional neural networks , author=. 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) , pages=. 2016 , doi=

    2016

  42. [50]

    Advances in Neural Information Processing Systems , volume=

    Generating images with perceptual similarity metrics based on deep networks , author=. Advances in Neural Information Processing Systems , volume=. 2016 , publisher=

    2016

  43. [51]

    International Conference on Learning Representations , year=

    Very deep convolutional networks for large-scale image recognition , author=. International Conference on Learning Representations , year=. doi:10.48550/arXiv.1409.1556 , url=

    doi arXiv

  44. [52]

    Seismological Research Letters , volume=

    Woollam, Jack and M. Seismological Research Letters , volume=. 2022 , doi=

    2022

  45. [53]

    and Nocedal, Jorge , journal=

    Liu, Dong C. and Nocedal, Jorge , journal=. On the limited memory. 1989 , doi=

    1989

  46. [54]

    and Bourget, M

    Brougois, A. and Bourget, M. and Lailly, P. and Poulet, M. and Ricarte, P. and Versteeg, R. , title =. Proceedings of the 1990. 1990 , eid =. doi:10.3997/2214-4609.201411190 , publisher =

    1990

  47. [55]

    and Brac, J

    Aminzadeh, F. and Brac, J. and Kunz, T. , title=. 1997 , note=

    1997

  48. [56]

    arXiv preprint arXiv:2108.07258 , year=

    On the opportunities and risks of foundation models , author=. arXiv preprint arXiv:2108.07258 , year=

    arXiv

  49. [57]

    Architecture and performance of

    Luporini, Fabio and Louboutin, Mathias and Lange, Michael and Kukreja, Navjot and Witte, Philipp and H\". Architecture and performance of. ACM Transactions on Mathematical Software , volume=. 2020 , doi=

    2020

  50. [58]

    and Herrmann, Felix J

    Louboutin, Mathias and Lange, Michael and Luporini, Fabio and Kukreja, Navjot and Witte, Philipp A. and Herrmann, Felix J. and Velesko, Paulius and Gorman, Gerard J. , title=. Geoscientific Model Development , volume=. 2019 , doi=

    2019

  51. [59]

    and Haberland, Matt and Reddy, Tyler and Cournapeau, David and Burovski, Evgeni and Peterson, Pearu and Weckesser, Warren and Bright, Jonathan and

    Virtanen, Pauli and Gommers, Ralf and Oliphant, Travis E. and Haberland, Matt and Reddy, Tyler and Cournapeau, David and Burovski, Evgeni and Peterson, Pearu and Weckesser, Warren and Bright, Jonathan and. Nature Methods , year =

Pith tools

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