Pith. sign in

REVIEW 5 major objections 5 minor 62 references

Semi-Supervised Learning for AVO Inversion with Strong Spatial Feature Constraints

T0 review · 5 major / 5 minor · reviewed 2026-08-10 · deepseek-v4-flash

Pith's one-line read The paper proposes a semi-supervised AVO inversion method whose label-annihilation and self-consistency losses create a two-way spatial mapping between well and non-well locations, and reports improved lateral continuity and accuracy over…

desk verdict A credible but incremental masked-reconstruction regularizer for semi-supervised AVO inversion; the comparative claim is plausible, but the low-frequency prior is not controlled and the field blind-well description contradicts itself. read the letter →

arxiv 2501.15473 v2 pith:6YFKUYWX submitted 2025-01-26 physics.geo-ph

classification physics.geo-ph
keywords AVOinversionsemi-supervisedlearningstrongspatialfeatureconstraintslabelannihilationoperatorlateralcontinuitylow-frequencymodelconvolutionalneuralnetworkelasticparameterestimation
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

The paper proposes a semi-supervised deep-learning method for prestack AVO inversion that recovers P-wave velocity, S-wave velocity, and density from multi-angle seismic gathers. Its central claim is that adding a strong spatial feature constraint between well and non-well locations, enforced by a label-annihilation operator and a recovery network, yields laterally continuous and more accurate parameter volumes than one- and two-dimensional deep learning baselines under scarce well-log labels. The method combines an inversion network, a forward-modeling network, and a strong spatial feature constraint (SSFC) network, and it feeds a low-frequency model of the elastic parameters as conditional input. On the synthetic Marmousi2 model and on a South China Sea field dataset, the paper reports the best Pearson correlation, R-squared, and RMSE among the compared methods, with the gains holding when only a fraction of traces carry labels.

What carries the argument

The load-bearing object is the label-annihilation operator R, a matrix that keeps all traces except the well-location (middle) trace, setting that trace's entries to zero. The predicted parameter volume is multiplied by R, and the SSFC network is trained to reconstruct the zeroed trace from its neighbors under a self-consistency loss: the reconstruction must match the true well log where the label was annihilated (L_SSFC) and must reproduce the network's own prediction at the other traces (L_SC). This creates a two-way mapping: well labels constrain adjacent traces through the inversion network, and neighboring traces are used to regenerate the well trace, propagating lateral structure. The forward network provides a semi-supervised reconstruction loss on unlabeled seismic data, and a low-frequency elastic-parameter model is fed into the inversion network alongside the seismic gathers.

What would settle it

On a synthetic model with a strong lateral velocity contrast that the low-frequency interpolation intentionally smooths away, run SSFC-SSL and the 2D-SSL baseline without the low-frequency input; if SSFC-SSL's advantage over 2D-SSL shrinks below the paper's reported margin or disappears, the improvement is attributable to the low-frequency prior rather than the strong spatial feature constraint.

Watch

Extended reading notes

Core claim

The paper claims that AVO inversion can be improved by turning sparse well-log labels into a bidirectional spatial constraint. Instead of using well logs only to supervise the traces where wells exist, the method multiplies the predicted parameter volume by a label-annihilation matrix R that zeros out the well-location trace, then trains a second network to recreate that missing trace from its neighbors. Requiring the recreated trace to match both the well logs and the annihilated prediction makes spatial features flow from non-well to well locations and back, which is the strong two-way mapping that 1D and 2D baselines lack. In the reported experiments, this SSFC-SSL scheme achieves higher PCC and R-squared and lower RMSE for all three elastic parameters than the model-based L-BFGS, 1D supervised, 1D semi-supervised, and 2D semi-supervised comparisons, and it preserves accuracy when only 0.26% or 0.11% of traces carry labels.

Load-bearing premise

The method requires a reliable low-frequency model of the elastic parameters, built by interpolating and smoothing well-log curves; if that model is wrong or missing far from wells, the inversion error will be dominated by the prior's error, and the spatial-constraint gains may vanish.

Editorial extensions

If this is right

  • The method reduces the vertical striping artifacts that plague 1D trace-by-trace inversion, because the SSFC loss forces spatial coherence across adjacent traces.
  • Under scarce well-log labels, SSFC-SSL degrades more gracefully than 1D supervised learning; the paper reports it keeps the highest PCC and R-squared and lowest RMSE with 18, 7, or 3 training traces on the Marmousi2 model.
  • At prediction time only the inversion network is used; the SSFC and forward networks are discarded, so the improved lateral continuity is obtained without extra inference cost.
  • The authors state the framework is intended to extend to post-stack inversion, where the same spatial-constraint and semi-supervised losses would apply.

Reading between the lines

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

  • An ablation that removes the low-frequency model input would isolate how much of the gain comes from the SSFC loss versus the prior; the paper does not report such an ablation.
  • The annihilation-plus-reconstruction recipe is generic: any sparse-label geophysical or image-reconstruction task could adopt it to propagate label information laterally, not just AVO inversion.
  • Because the SSFC network is trained on the inversion network's own predictions, an early inaccurate inversion output could teach the SSFC network a biased spatial prior; starting with a warm-up on well traces only might reduce that risk.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

5 major / 5 minor

Summary. The manuscript proposes SSFC-SSL, a semi-supervised AVO inversion method that combines an inversion network, a forward network, and a strong spatial feature constraint (SSFC) network. The method uses a label-annihilation operator to constrain predictions at well locations from adjacent non-well traces and introduces a self-consistency loss, while also feeding a low-frequency model of elastic parameters as conditional input at every trace. The authors compare SSFC-SSL with 1D-SL, 1D-SSL, 2D-SSL, and model-based (L-BFGS) methods on a Marmousi2 synthetic example and on field data from the South China Sea, reporting improved PCC, R², RMSE, and visually smoother lateral continuity.

Significance. If the empirical claims are supported, the SSFC mechanism would be a useful contribution to deep-learning-based AVO inversion, particularly under scarce well-log labels, and the paper provides a clear architecture, explicit loss functions, and both synthetic and field demonstrations. The proposed two-way spatial constraint idea is interesting and potentially valuable. However, the reported gains currently lack a controlled comparison: the low-frequency prior is not shown to be equally available to the baselines, there is no ablation isolating the SSFC loss, single-run metrics are reported without error bars or significance tests, and the field evaluation protocol contains an internal inconsistency. These issues must be resolved before the comparative claim can be accepted.

major comments (5)
  1. [III-B/III-C, Eq. (17)] The manuscript never states whether the 1D-SL, 1D-SSL, and 2D-SSL baselines received the low-frequency model mlow as input during training and inference. Since the inversion network in Eq. (17) takes (Dl, mlow) and (Du, mlow) as inputs, the proposed method injects low-frequency information that is known to be the most difficult part of band-limited AVO inversion to recover from seismic alone. If the baselines operate on seismic data without mlow, the improvements in Tables II and III (e.g., P-wave RMSE 0.0420 vs. 0.0553 for 2D-SSL) may be an artifact of unequal information rather than the SSFC mechanism. Please either provide mlow to all baselines or add an ablation of SSFC-SSL without mlow, and report both setups.
  2. [III-C] The field-data evaluation is internally inconsistent: the text states that 'Wells G1, G2, and G3 were used for training, while well G3 was used to evaluate the inversion results,' but Fig. 21 and Table III compare the inversion results 'at the location of the blind well G2.' If G2's logs were used in training or in constructing the low-frequency model via well-log interpolation and smoothing, then the reported field metrics are not a blind test. Please clarify which well is the blind test and confirm that its logs were excluded from both training and prior construction.
  3. [III-B/III-C, Eq. (17)] The construction of the low-frequency model is not described reproducibly. For the synthetic case, the text states only that 'a low-pass filter was applied to the model data to obtain the low-frequency model,' without specifying the cutoff frequency, filter order, or design. For the field case, it states that the low-frequency information is obtained by 'layer interpolation, extrapolation, and smoothing of the corresponding well-log curves,' without giving the interpolation or smoothing parameters. These details are essential because mlow is a conditional input at every trace and can dominate the inversion result; without them, the experiments cannot be reproduced or compared fairly.
  4. [III-B, Tables II-III] The central claim that SSFC-SSL 'significantly improves' inversion accuracy is supported only by single-run metrics. There are no repeated runs, error bars, or statistical significance tests, and no ablation that removes the SSFC loss (LSSFC and LSC) or the low-frequency model. Given that the improvement over 2D-SSL is modest for some parameters (e.g., density RMSE 0.0123 vs. 0.0166 on the synthetic data), run-to-run variation could change the conclusion. Please add repeated runs with error bars and an ablation study to isolate the contribution of each loss term and of mlow.
  5. [III-B, Figs. 10-16 and Tables II-III] The paper repeatedly claims improved 'lateral continuity,' but no metric for lateral continuity is reported; the tabulated metrics (PCC, R², RMSE) are pointwise measures. Please quantify lateral continuity directly, for example with SSIM, total variation, or a trace-to-trace smoothness measure, to substantiate the claim.
minor comments (5)
  1. [III-C] The text says Table III reports 'PCC, R2, and SSIM values,' but the table actually reports PCC, R², and RMSE; please correct the text or the table header.
  2. [I] The reference to 'Fig. 2(b)' in the discussion of 2D data-driven inversion should likely be 'Fig. 1(b)', since Fig. 2 shows the training process rather than the workflow schematic.
  3. [II-D, Eq. (18)] The subscript for the self-consistency loss is written as 'Lsc' with a lowercase 's' in Eq. (18), while Eqs. (15)-(17) use 'LSC'; please make the notation consistent.
  4. [III-B] The description of unlabeled data sampling as 'randomly sampled at an interval of 50 traces' is ambiguous: it could mean every 50th trace or a random selection with an average spacing of 50; please clarify.
  5. [Abstract] The word 'significantly' in the abstract is not supported by statistical tests or error bars; please qualify the claim or add the corresponding statistical evidence.

Circularity Check

1 steps flagged · score 4.0 of 10

Low-frequency prior is built from the same well logs used as labels and fed into the inversion network, making part of the reported accuracy gain input-driven rather than due to the SSFC mechanism alone.

  1. fitted input called prediction [Eq. (17); Section II-C; Section III-B; Section III-C]
    "In addition, SSFC-SSL uses the low-frequency information of the elastic parameters as conditional input to improve the inversion accuracy. ... The low-frequency information of the different elastic parameters is obtained by layer interpolation, extrapolation, and smoothing of the corresponding well-log curves."

    mlow is created from the same well-log parameters used as labels: on synthetic data the true model is low-pass filtered, and on field data the well-log curves are interpolated/extrapolated/smoothed. Eq. (17) feeds mlow as conditional input to the inversion network at every trace, so the low-frequency component of the predicted volume is supplied by the labels by construction. Because the paper never states that the 1D/2D baselines also received mlow, the accuracy gains in Tables II and III are not a controlled test of the SSFC mechanism. The field 'blind-well' claim is further undermined by the G2/G3 evaluation-well inconsistency.

full rationale

The SSFC-SSL derivation is not formally self-definitional: the network learns a high-frequency residual and a spatial-consistency mapping beyond the low-frequency input, and the self-consistency loss is a standard auxiliary objective, not a hidden reuse of the target. However, the evaluation of the central claim is partially circular. The low-frequency model is explicitly constructed from the well-log labels (true-model low-pass on synthetic data; well-log interpolation/smoothing on field data) and is inserted as an input to the inversion network. The reported improvements over 1D/2D baselines therefore conflate the effect of injecting label-derived low-frequency information with the effect of the strong spatial feature constraints, unless the baselines also received the same prior, which is not stated. The self-citations [44] and [61] are not load-bearing: self-consistency is independently supported by [59] and [60]. In addition, the field-data protocol is internally inconsistent: Section III-C says 'Wells G1, G2, and G3 were used for training, while well G3 was used to evaluate,' while Fig. 21 and Table III evaluate 'the blind well G2'; this undermines the blind-test evidence but is a protocol flaw rather than a circular derivation. Overall, there is partial circularity in the evaluation setup, but the core SSFC mechanism retains independent content, giving a score of 4.

Assumptions & free parameters 1 free parameters · 3 assumptions · 0 invented entities

The central claim depends on a small set of domain assumptions. The low-frequency prior is the most consequential: it is constructed from the label wells, so the method is not purely label-free at non-well locations. The Zoeppritz forward model and the expressivity of the networks are standard assumptions.

free parameters (1)
  • Low-frequency model construction = unspecified
    The low-frequency model is obtained by low-pass filtering (synthetic) or layer interpolation, extrapolation, and smoothing (field) of well logs (Sections III-B, III-C). The cutoff and interpolation details are not reported; accuracy of the method depends on this prior.
assumptions (3)
  • domain assumption The Zoeppritz equations (Eq. 2) and convolution model (Eq. 1) accurately describe seismic amplitudes for this data.
    Used to generate synthetic reflection coefficients and to motivate the forward-modeling loss.
  • domain assumption The low-frequency model is a reliable prior at all locations.
    Fed to the inversion network as conditional input (Eq. 17); an inaccurate prior propagates into the output.
  • domain assumption The three networks are expressive enough and train stably to represent the required mappings.
    No capacity or convergence analysis is provided; standard deep learning assumption.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Semi-Supervised Learning for AVO Inversion with Strong Spatial Feature Constraints." pith.science (2026). https://pith.science/paper/6YFKUYWX

@misc{pith2026250115473,
  author       = {Pith},
  title        = {Pith review of: Semi-Supervised Learning for AVO Inversion with Strong Spatial Feature Constraints},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/6YFKUYWX}},
  note         = {Machine review of arXiv:2501.15473}
}
read the original abstract

One-dimensional convolution is a widely used deep learning technique in prestack amplitude variation with offset (AVO) inversion; however, it lacks lateral continuity. Although two-dimensional convolution improves lateral continuity, due to the sparsity of well-log data, the model only learns weak spatial features and fails to explore the spatial correlations in seismic data fully. To overcome these challenges, we propose a novel AVO inversion method based on semi-supervised learning with strong spatial feature constraints (SSFC-SSL). First, two-dimensional predicted values are obtained through the inversion network, and the predicted values at well locations are sparsely represented using well-log labels. Subsequently, a label-annihilation operator is introduced, enabling the predicted values at non-well locations to learn the spatial features of well locations through the neural network. Ultimately, a two-way strong spatial feature mapping between non-well locations and well locations is achieved. Additionally, to reduce the dependence on well-log labels, we combine the semi-supervised learning strategy with a low-frequency model, further enhancing the robustness of the method. Experimental results on both synthetic example and field data demonstrate that the proposed method significantly improves lateral continuity and inversion accuracy compared to one- and two-dimensional deep learning techniques.

Figures

Figures reproduced from arXiv: 2501.15473 by the authors.

Figure 1
Figure 1. Training and testing workflows for seismic inversion methods. (a) 1D data-driven seismic inversion, (b) 2D data-driven seismic inversion, and (c) the [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 3
Figure 3. The structure of dilated convolution. TCN block 1×1 CNN + + CNN block 16 32 32 16 32 16 3 Dilated convolution Convolution BN+ReLU Dropout Input [M, N] Output [M, N] K+3 [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figure 4
Figure 4. Architecture of the inversion network. CNN block 16 32 16 Convolution BN+ReLU Input [M, N] Output [M, N] 3 K [PITH_FULL_IMAGE:figures/full_fig_p003_4.png] view at source ↗
Figures from the paper (16 more)
Figure 5
Figure 5. Figure 5: Architecture of the forward network. CNN block 16 32 16 3 Convolution BN+ReLU Input [M, N] Output [M, N] 3 [PITH_FULL_IMAGE:figures/full_fig_p003_5.png]
Figure 6
Figure 6. Figure 6: Architecture of the SSFC network. model. In the absence of random noise, the convolution model [PITH_FULL_IMAGE:figures/full_fig_p003_6.png]
Figure 7
Figure 7. Figure 7: Synthetic angle gather profiles with different incident angles of (a) [PITH_FULL_IMAGE:figures/full_fig_p004_7.png]
Figure 8
Figure 8. Figure 8: Training traces of (a) P-wave velocity, (b) S-wave velocity, and (c) density. [PITH_FULL_IMAGE:figures/full_fig_p004_8.png]
Figure 9
Figure 9. Figure 9: Loss function curves of (a) LI , (b) LF , (c) LSSF C , (d) LRecon, and (e) LSC . TABLE I HYPERPARAMETERS IN SSFC-SSL. The type of the hyperparameter Value Basic parameter Epoch 500 Batch size 100 Learning rate 0.002 Weight decay 0.0001 Dropout 0.2 Initial kernel size 3…
Figure 10
Figure 10. Figure 10: Inversion results for (a)-(e) P-wave velocity, (f)-(j) S-wave velocity, and (k)-(o) density using different methods. The first column shows the model [PITH_FULL_IMAGE:figures/full_fig_p005_10.png]
Figure 11
Figure 11. Figure 11: Absolute difference in (a-e) P-wave velocity, (f-j) S-wave velocity, and (k-o) density inversion results compared to the true data. The figure layout [PITH_FULL_IMAGE:figures/full_fig_p005_11.png]
Figure 12
Figure 12. Figure 12: Inversion results of (a) P-wave velocity, (b) S-wave velocity, and (c) density at trace 600 on the synthetic data. The black lines indicate true data, the [PITH_FULL_IMAGE:figures/full_fig_p006_12.png]
Figure 13
Figure 13. Figure 13: (a) Seven training traces and (b) three training traces for P-wave [PITH_FULL_IMAGE:figures/full_fig_p006_13.png]
Figure 14
Figure 14. Figure 14: Inversion results for (a)-(e) P-wave velocity, (f)-(j) S-wave velocity, and (k)-(o) density using different methods with seven training traces. The first [PITH_FULL_IMAGE:figures/full_fig_p007_14.png]
Figure 15
Figure 15. Figure 15: Inversion results for (a)-(e) P-wave velocity, (f)-(j) S-wave velocity, and (k)-(o) density using different methods with three training traces. The figure [PITH_FULL_IMAGE:figures/full_fig_p007_15.png]
Figure 16
Figure 16. Figure 16: Comparison of the (a)-(c) PCC, (d)-(f) R2 , and (g)-(i) RMSE of the different methods. and reflection coefficients. By substituting (4) into (1), the general form of the forward equation in AVO theory is obtained, expressed as D = W ∗ f(m) = G(m) (5) where G represent…
Figure 17
Figure 17. Figure 17: Field seismic profiles of (a) 0◦, (b) 6◦, (c) 12◦, (d) 18◦, and (e) 24◦. Low-frequency profile of (f) P-wave velocity, (g) S-wave velocity, and (h) density. locations) through the inversion and forward networks. The objective function is expressed as J(Iw, Fw) = min I…
Figure 18
Figure 18. Figure 18: Inversion results of P-wave velocity using (a) model-based, (b) 1D [PITH_FULL_IMAGE:figures/full_fig_p011_18.png]
Figure 20
Figure 20. Figure 20: Inversion results of density using (a) model-based, (b) 1D-SL, (c) [PITH_FULL_IMAGE:figures/full_fig_p012_20.png]
Figure 21
Figure 21. Figure 21: Inversion results of (a) P-wave velocity, (b) S-wave velocity, and (c) density at well G2. [PITH_FULL_IMAGE:figures/full_fig_p013_21.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

62 extracted references · 59 canonical work pages

  1. [1]

    Waveform-based avo inversion and avo prediction-error,

    J. L. Simmons Jr and M. M. Backus, “Waveform-based avo inversion and avo prediction-error,” Geophysics, vol. 61, no. 6, pp. 1575–1588, 1996

  2. [2]

    High-resolution three-term avo inversion by means of a trivariate cauchy probability distribution,

    W. Alemie and M. D. Sacchi, “High-resolution three-term avo inversion by means of a trivariate cauchy probability distribution,” Geophysics, vol. 76, no. 3, pp. R43–R55, 2011

  3. [3]

    Avo inversion based on transfer learning and low-frequency model,

    J. Meng, S. Wang, W. Cheng, Z. Wang, and L. Yang, “Avo inversion based on transfer learning and low-frequency model,” IEEE Geoscience and Remote Sensing Letters , vol. 19, pp. 1–5, 2021

  4. [4]

    On the reflection and propagation of seismic waves,

    K. Zoeppritz, “On the reflection and propagation of seismic waves,” Gottinger Nachrichten, vol. 1, no. 5, pp. 66–84, 1919

  5. [5]

    Approximations to the reflection and transmission co- efficients of plane longitudinal and transverse waves,

    R. Bortfeld, “Approximations to the reflection and transmission co- efficients of plane longitudinal and transverse waves,” Geophysical Prospecting, vol. 9, no. 4, pp. 485–502, 1961

  6. [6]

    Quantitative seismology: Theory and methods,

    K. Aid and P. Richards, “Quantitative seismology: Theory and methods,” San Francisco, 1980

  7. [7]

    A simplification of the zoeppritz equations,

    R. Shuey, “A simplification of the zoeppritz equations,” Geophysics, vol. 50, no. 4, pp. 609–614, 1985

  8. [8]

    Weighted stacking for rock property estimation and detection of gas,

    G. Smith and P. Gidlow, “Weighted stacking for rock property estimation and detection of gas,” Geophysical prospecting, vol. 35, no. 9, pp. 993– 1014, 1987

Show all 62 references
  1. [9]

    Improved avo fluid detection and lithology discrimination using lam ´e petrophysical parameters;“λρ

    B. Goodway, T. Chen, and J. Downton, “Improved avo fluid detection and lithology discrimination using lam ´e petrophysical parameters;“λρ”,“µρ”, & “ λ/µ fluid stack”, from p and s inversions,” in SEG technical program expanded abstracts 1997 . Society of Exploration Geophysici...

  2. [10]

    The application of avo and inversion to the estimation of rock properties,

    D. Gray and E. Andersen, “The application of avo and inversion to the estimation of rock properties,” in SEG International Exposition and Annual Meeting. SEG, 2000, pp. SEG–2000

  3. [11]

    Linearized inversion of 3-d multi-offset data: Background reconstruction and avo inversion,

    L. Ikelle, “Linearized inversion of 3-d multi-offset data: Background reconstruction and avo inversion,” Geophysical Journal International , vol. 123, no. 2, pp. 507–528, 1995

  4. [12]

    Linearized amplitude variation with offset (avo) inversion with supercritical angles,

    J. E. Downton and C. Ursenbach, “Linearized amplitude variation with offset (avo) inversion with supercritical angles,” Geophysics, vol. 71, no. 5, pp. E49–E55, 2006

  5. [13]

    Avo inversion based on inverse operator estimation in trust region,

    X.-Y . Yin, W. Deng, and Z.-Y . Zong, “Avo inversion based on inverse operator estimation in trust region,” Journal of Geophysics and Engi- neering, vol. 13, no. 2, pp. 194–206, 2016

  6. [14]

    Avo inversion with high-order total variation regularization,

    B. She, Y . Wang, J. Zhang, J. Wang, and G. Hu, “Avo inversion with high-order total variation regularization,”Journal of Applied Geophysics, vol. 161, pp. 167–181, 2019

  7. [15]

    Joint avo inversion in the time and frequency domain with bayesian interference,

    Z.-Y . Zong, X.-Y . Yin, and K. Li, “Joint avo inversion in the time and frequency domain with bayesian interference,” Applied Geophysics, vol. 13, no. 4, pp. 631–640, 2016

  8. [16]

    Hierarchical bayesian probabilistic seismic avo inversion using gibbs sampling with ia2rms algorithm,

    K. Li, K. Li, X. Yin, and Z. Zong, “Hierarchical bayesian probabilistic seismic avo inversion using gibbs sampling with ia2rms algorithm,” IEEE Geoscience and Remote Sensing Letters , vol. 19, pp. 1–5, 2022

  9. [17]

    Quantitative interpretation using inverse rock-physics modeling on avo data,

    E. H. Jensen, T. A. Johansen, P. Avseth, and K. Bredesen, “Quantitative interpretation using inverse rock-physics modeling on avo data,” The Leading Edge, vol. 35, no. 8, pp. 677–683, 2016

  10. [18]

    Global optimization with model-space preconditioning: Application to avo inversion,

    S. Misra and M. D. Sacchi, “Global optimization with model-space preconditioning: Application to avo inversion,” Geophysics, vol. 73, no. 5, pp. R71–R82, 2008

  11. [19]

    Constrained non-linear avo inversion based on the adjoint-state optimization,

    N. Ahmed, W. W. Weibull, and D. Grana, “Constrained non-linear avo inversion based on the adjoint-state optimization,” Computers & Geosciences, vol. 168, p. 105214, 2022

  12. [20]

    Frequency-dependent avo inversion applied to physically based models for seismic attenuation,

    N. Ahmed, W. W. Weibull, B. Quintal, D. Grana, and T. Bhakta, “Frequency-dependent avo inversion applied to physically based models for seismic attenuation,” Geophysical Journal International , vol. 233, no. 1, pp. 234–252, 2023

  13. [21]

    Acoustic impedance pre- diction using an attention-based dual-branch double-inversion network,

    W. Feng, Y . Liu, Y . Li, H. Li, and X. Wang, “Acoustic impedance pre- diction using an attention-based dual-branch double-inversion network,” Earth Science Informatics , vol. 18, no. 1, pp. 1–20, 2025

  14. [22]

    Avo inversion in exploration—key learnings from a norwegian sea prospect,

    P. Avseth, A. Janke, and F. Horn, “Avo inversion in exploration—key learnings from a norwegian sea prospect,” The Leading Edge , vol. 35, no. 5, pp. 405–414, 2016

  15. [23]

    Pp and ps joint avo inversion and fluid prediction,

    Q. Du and H. Yan, “Pp and ps joint avo inversion and fluid prediction,” Journal of Applied Geophysics , vol. 90, pp. 110–118, 2013

  16. [24]

    Pertinent multigate mixture-of- experts-based prestack three-parameter seismic inversion,

    Z. Li, X. Chen, J. Li, and J. Zhang, “Pertinent multigate mixture-of- experts-based prestack three-parameter seismic inversion,” IEEE Trans- actions on Geoscience and Remote Sensing , vol. 60, pp. 1–15, 2022

  17. [25]

    Data-driven method for an improved linearised avo inversion,

    L. Niu, J. Geng, X. Wu, L. Zhao, and H. Zhang, “Data-driven method for an improved linearised avo inversion,” Journal of Geophysics and Engineering, vol. 18, no. 1, pp. 1–22, 2021

  18. [26]

    Deep learning seismic inversion based on prestack waveform datasets,

    J. Zhang, H. Sun, G. Zhang, and X. Zhao, “Deep learning seismic inversion based on prestack waveform datasets,” IEEE Transactions on Geoscience and Remote Sensing , vol. 60, pp. 1–11, 2022

  19. [27]

    Deep-learning inversion: A next-generation seismic velocity model building method,

    F. Yang and J. Ma, “Deep-learning inversion: A next-generation seismic velocity model building method,” Geophysics, vol. 84, no. 4, pp. R583– R599, 2019

  20. [28]

    Current status and application prospect of deep learning in geophysics,

    H. W ANG, J.-y. Y AN, G.-m. FU, and X. W ANG, “Current status and application prospect of deep learning in geophysics,” Progress in Geophysics, vol. 35, no. 2, pp. 642–655, 2020

  21. [29]

    Deep learning reservoir porosity prediction based on multilayer long short-term mem- ory network,

    W. Chen, L. Yang, B. Zha, M. Zhang, and Y . Chen, “Deep learning reservoir porosity prediction based on multilayer long short-term mem- ory network,” Geophysics, vol. 85, no. 4, pp. W A213–W A225, 2020. 15

  22. [30]

    A multiple-input deep residual convolutional neural network for reservoir permeability prediction,

    M. Masroor, M. E. Niri, and M. H. Sharifinasab, “A multiple-input deep residual convolutional neural network for reservoir permeability prediction,” Geoenergy Science and Engineering , vol. 222, p. 211420, 2023

  23. [31]

    Seismic fault detection with convolutional neural network,

    W. Xiong, X. Ji, Y . Ma, Y . Wang, N. M. AlBinHassan, M. N. Ali, and Y . Luo, “Seismic fault detection with convolutional neural network,” Geophysics, vol. 83, no. 5, pp. O97–O103, 2018

  24. [32]

    Fast global self-attention for seismic image fault identification,

    S. Wang, X. Si, Z. Cai, L. Sun, W. Wang, and Z. Jiang, “Fast global self-attention for seismic image fault identification,” IEEE Transactions on Geoscience and Remote Sensing , 2024

  25. [33]

    A twice denoising autoencoder framework for random seismic noise attenuation,

    Z. Liao, Y . Li, E. Xia, Y . Liu, and R. Hu, “A twice denoising autoencoder framework for random seismic noise attenuation,” IEEE Transactions on Geoscience and Remote Sensing , vol. 61, pp. 1–15, 2023

  26. [34]

    Low-frequency noise suppres- sion method based on improved dncnn in desert seismic data,

    Y . Zhao, Y . Li, X. Dong, and B. Yang, “Low-frequency noise suppres- sion method based on improved dncnn in desert seismic data,” IEEE Geoscience and Remote Sensing Letters , vol. 16, no. 5, pp. 811–815, 2018

  27. [35]

    Seismic stratigraphy interpretation by deep convolutional neural networks: A semisupervised workflow,

    H. Di, Z. Li, H. Maniar, and A. Abubakar, “Seismic stratigraphy interpretation by deep convolutional neural networks: A semisupervised workflow,” Geophysics, vol. 85, no. 4, pp. W A77–W A86, 2020

  28. [36]

    Semi-supervised seismic strati- graphic interpretation constrained by spatial structure,

    X. Gu, W. Lu, Y . Li, and Y . Wang, “Semi-supervised seismic strati- graphic interpretation constrained by spatial structure,” IEEE Transac- tions on Geoscience and Remote Sensing , vol. 61, pp. 1–10, 2023

  29. [37]

    Convolutional neural network for seismic impedance inversion,

    V . Das, A. Pollack, U. Wollner, and T. Mukerji, “Convolutional neural network for seismic impedance inversion,” Geophysics, vol. 84, no. 6, pp. R869–R880, 2019

  30. [38]

    Estimation of acoustic impedance from seismic data using temporal convolutional network,

    A. Mustafa, M. Alfarraj, and G. AlRegib, “Estimation of acoustic impedance from seismic data using temporal convolutional network,” in SEG technical program expanded abstracts 2019 . Society of Exploration Geophysicists, 2019, pp. 2554–2558

  31. [39]

    Elastic-parameters inversion from ei based on the deep-learning method,

    D. Cao, P. An, and S. Liu, “Elastic-parameters inversion from ei based on the deep-learning method,” in SEG Technical Program Expanded Abstracts 2018. Society of Exploration Geophysicists, 2018, pp. 640– 644

  32. [40]

    Robust deep learning seismic inversion with a priori initial model constraint,

    J. Zhang, J. Li, X. Chen, Y . Li, G. Huang, and Y . Chen, “Robust deep learning seismic inversion with a priori initial model constraint,” Geophysical Journal International, vol. 225, no. 3, pp. 2001–2019, 2021

  33. [41]

    Deep- learning inversion of seismic data,

    S. Li, B. Liu, Y . Ren, Y . Chen, S. Yang, Y . Wang, and P. Jiang, “Deep- learning inversion of seismic data,” arXiv preprint arXiv:1901.07733 , 2019

  34. [42]

    Applications of supervised deep learning for seismic interpretation and inversion,

    Y . Zheng, Q. Zhang, A. Yusifov, and Y . Shi, “Applications of supervised deep learning for seismic interpretation and inversion,” The Leading Edge, vol. 38, no. 7, pp. 526–533, 2019

  35. [43]

    Semisupervised sequence modeling for elastic impedance inversion,

    M. Alfarraj and G. AlRegib, “Semisupervised sequence modeling for elastic impedance inversion,” Interpretation, vol. 7, no. 3, pp. SE237– SE249, 2019

  36. [44]

    The nash-mtl-stcn for prestack three-parameter inversion,

    Y . Liu, Y . Li, H. Li, J. Peng, Z. Liao, and W. Feng, “The nash-mtl-stcn for prestack three-parameter inversion,” arXiv preprint arXiv:2407.00684, 2024

  37. [45]

    Multi-task learning as a bargaining game,

    A. Navon, A. Shamsian, I. Achituve, H. Maron, K. Kawaguchi, G. Chechik, and E. Fetaya, “Multi-task learning as a bargaining game,” arXiv preprint arXiv:2202.01017 , 2022

  38. [46]

    Physics-guided deep learning for seismic inversion with hybrid training and uncertainty analysis,

    J. Sun, K. A. Innanen, and C. Huang, “Physics-guided deep learning for seismic inversion with hybrid training and uncertainty analysis,” Geophysics, vol. 86, no. 3, pp. R303–R317, 2021

  39. [47]

    Prestack and post- stack inversion using a physics-guided convolutional neural network,

    R. Biswas, M. K. Sen, V . Das, and T. Mukerji, “Prestack and post- stack inversion using a physics-guided convolutional neural network,” Interpretation, vol. 7, no. 3, pp. SE161–SE174, 2019

  40. [48]

    Deep learning- based pre-stack seismic inversion constrained by avo attributes,

    Q. Ge, H. Cao, Z. Yang, S. Yuan, and C. Song, “Deep learning- based pre-stack seismic inversion constrained by avo attributes,” IEEE Geoscience and Remote Sensing Letters , 2024

  41. [49]

    Deep recurrent architectures for seismic tomography,

    A. Adler, M. Araya-Polo, and T. Poggio, “Deep recurrent architectures for seismic tomography,” in 81st EAGE conference and exhibition 2019, vol. 2019, no. 1. European Association of Geoscientists & Engineers, 2019, pp. 1–5

  42. [50]

    Application of bi- directional long short-term memory recurrent neural network for seismic impedance inversion,

    R. Guo, J. Zhang, D. Liu, Y . Zhang, and D. Zhang, “Application of bi- directional long short-term memory recurrent neural network for seismic impedance inversion,” in 81st EAGE Conference and Exhibition 2019 , vol. 2019, no. 1. European Association of Geoscientists & Engineer...

  43. [51]

    Seismic velocity estimation: A deep recurrent neural-network approach,

    G. Fabien-Ouellet and R. Sarkar, “Seismic velocity estimation: A deep recurrent neural-network approach,” Geophysics, vol. 85, no. 1, pp. U21– U29, 2020

  44. [52]

    Double-scale supervised inversion with a data-driven forward model for low-frequency impedance recovery,

    S. Yuan, X. Jiao, Y . Luo, W. Sang, and S. Wang, “Double-scale supervised inversion with a data-driven forward model for low-frequency impedance recovery,” Geophysics, vol. 87, no. 2, pp. R165–R181, 2022

  45. [53]

    Seimic impedance inversion based on semi-supervised learning,

    S. Shi, M. Li, J. Wang, W. Chang, L. Li, and D. Xie, “Seimic impedance inversion based on semi-supervised learning,” Computers & Geosciences, vol. 182, p. 105497, 2024

  46. [54]

    Avo inversion based on closed-loop multitask conditional wasserstein generative adversarial net- work,

    Z. Wang, S. Wang, C. Zhou, and W. Cheng, “Avo inversion based on closed-loop multitask conditional wasserstein generative adversarial net- work,” IEEE Transactions on Geoscience and Remote Sensing , vol. 61, pp. 1–13, 2023

  47. [55]

    Deep learning for multidimensional seismic impedance inversion,

    X. Wu, S. Yan, Z. Bi, S. Zhang, and H. Si, “Deep learning for multidimensional seismic impedance inversion,” Geophysics, vol. 86, no. 5, pp. R735–R745, 2021

  48. [56]

    Seismic inversion based on 2d-cnns and domain adaption,

    Q. Wang, Y . Wang, Y . Ao, and W. Lu, “Seismic inversion based on 2d-cnns and domain adaption,” IEEE Transactions on Geoscience and Remote Sensing, vol. 60, pp. 1–12, 2022

  49. [57]

    Joint learning for spatial context-based seismic inversion of multiple data sets for improved generalizability and robustness,

    A. Mustafa, M. Alfarraj, and G. AlRegib, “Joint learning for spatial context-based seismic inversion of multiple data sets for improved generalizability and robustness,” Geophysics, vol. 86, no. 4, pp. O37– O48, 2021

  50. [58]

    Transinver: 3d data- driven seismic inversion based on self-attention,

    K. Li, Y . Dou, Y . Xiao, R. Jing, J. Zhu, and C. Ma, “Transinver: 3d data- driven seismic inversion based on self-attention,” Geophysics, vol. 89, no. 1, pp. W A127–W A141, 2024

  51. [59]

    A review of generalized zero-shot learning meth- ods,

    F. Pourpanah, M. Abdar, Y . Luo, X. Zhou, R. Wang, C. P. Lim, X.-Z. Wang, and Q. J. Wu, “A review of generalized zero-shot learning meth- ods,” IEEE transactions on pattern analysis and machine intelligence , vol. 45, no. 4, pp. 4051–4070, 2022

  52. [60]

    Self-consistency improves chain of thought reasoning in language models,

    X. Wang, J. Wei, D. Schuurmans, Q. Le, E. Chi, S. Narang, A. Chowdh- ery, and D. Zhou, “Self-consistency improves chain of thought reasoning in language models,” arXiv preprint arXiv:2203.11171 , 2022

  53. [61]

    Zero-shot self-consistency learning for seismic irregular spatial sampling reconstruction,

    J. Peng, Y . Liu, M. Wang, Y . Li, and H. Li, “Zero-shot self-consistency learning for seismic irregular spatial sampling reconstruction,” arXiv preprint arXiv:2411.00911, 2024

  54. [62]

    Pytorch: An imperative style, high-performance deep learning library,

    A. Paszke, S. Gross, F. Massa, A. Lerer, J. Bradbury, G. Chanan, T. Killeen, Z. Lin, N. Gimelshein, L. Antiga et al. , “Pytorch: An imperative style, high-performance deep learning library,” Advances in neural information processing systems , vol. 32, 2019

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.