pith. sign in

arxiv: 2606.22867 · v1 · pith:5K54DFW7new · submitted 2026-06-22 · ⚛️ physics.geo-ph · cs.LG

Tensor Train Decomposition-based 3D Implicit Full Waveform Inversion with Multi-scale Structural Similarity

Liangsheng He , Chao Song , Tiansheng Chen , Tao Liu , Cai Liu This is my paper

Pith reviewed 2026-06-26 06:13 UTC · model grok-4.3

classification ⚛️ physics.geo-ph cs.LG
keywords tensor train decompositionfull waveform inversionimplicit neural networksstructural similarityvelocity model reconstructioncycle skippingseismic inversion3D subsurface imaging
0
0 comments X

The pith

Tensor train decomposition of the velocity model into core tensors predicted by axis-specific implicit networks enables memory-efficient 3D full waveform inversion.

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

The paper introduces a framework that represents the 3D subsurface velocity model as a tensor train decomposition, with each core tensor predicted by a dedicated implicit neural network using one-dimensional coordinates. This approach cuts the memory required for training the networks compared to directly modeling the full velocity field. The low-rank property of the decomposition promotes structural consistency in the inverted model. An objective function based on multi-scale structural similarity helps avoid cycle skipping by incorporating ultra-low frequency information. Tests on synthetic and real land data show the method produces continuous, accurate velocity models even when starting from poor initial guesses or lacking low-frequency content.

Core claim

The 3D velocity model is expressed as the product of low-rank core tensors in tensor train format, each core predicted by an axis-specific implicit neural network; optimizing these networks with a multi-scale structural similarity loss yields accurate and continuous velocity reconstructions from 3D full waveform data, even under challenging conditions such as poor starting models or absent low frequencies.

What carries the argument

Tensor train (TT) decomposition of the velocity model into low-rank core tensors, each generated by an axis-specific implicit neural network (INR) from 1D coordinate inputs.

If this is right

  • The inversion maintains high resolution and accuracy with significantly lower memory consumption than direct INR methods.
  • Structural consistency from the low-rank TT structure improves continuity of the reconstructed velocity field.
  • The M-SSIM loss mitigates cycle skipping by leveraging multi-scale features including ultra-low frequencies.
  • Accurate results are obtained on both synthetic examples and challenging land datasets.
  • Performance holds even when initial models are poor or low-frequency data is missing.

Where Pith is reading between the lines

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

  • Similar TT-based reparameterization could be applied to other inverse problems in geophysics that involve high-dimensional model spaces.
  • Testing the method on time-lapse or 4D data could show whether the low-rank structure captures temporal consistency.
  • The axis-specific INRs might be swapped for other approximators if the TT cores exhibit different smoothness properties.

Load-bearing premise

The subsurface velocity model possesses a low-rank structure in tensor train format that captures the essential features needed for accurate wave propagation modeling.

What would settle it

Running the method on a velocity model known to require high TT rank for accurate representation and comparing the inversion error to standard FWI or direct INR approaches on the same data.

Figures

Figures reproduced from arXiv: 2606.22867 by Cai Liu, Chao Song, Liangsheng He, Tao Liu, Tiansheng Chen.

Figure 1
Figure 1. Figure 1: Ricker wavelet misfit function test. (a) Ricker wavelet. (b) Nominalization misfit. [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Workflow of the proposed method. Gaussian window with standard deviation σw(i) , and C2 = 9 × 10−4 is a small constant introduced to ensure numerical stability. When the local energy distributions of the predicted and observed data are similar, the value of C approaches 1, indicating a high degree of similarity. Conversely, large differences in local energies lead to smaller values of C, reflecting reduced… view at source ↗
Figure 3
Figure 3. Figure 3: The source wavelet and corresponding frequency spectrum. (a) Source wavelet. (b) Frequency spectrum. [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: The synthetic data acquisition geometry. [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Overthrust velocity model [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: M-SSIM extracts multi-scale structural features from predicted and observed data. [PITH_FULL_IMAGE:figures/full_fig_p008_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Normalized frequency spectrum of multi-scale structural features. (a) Local mean amplitude. (b) Local [PITH_FULL_IMAGE:figures/full_fig_p009_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Constant gradient initial model and inversion results. (a) Initial modeld. (b) 3DFWI. (c) 3DIFWI. (d) 3DIFWI [PITH_FULL_IMAGE:figures/full_fig_p009_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: The absolute error between the inversion results and the true model. (a) Initial modeld. (b) 3DFWI. (c) [PITH_FULL_IMAGE:figures/full_fig_p009_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: The slices results obtained from constant gradient initialization model at Inline = 0.6 km, crossline = 0.3 km, [PITH_FULL_IMAGE:figures/full_fig_p010_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: The slices absolute error obtained from constant gradient initialization model at Inline = 0.6 km, crossline = [PITH_FULL_IMAGE:figures/full_fig_p010_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Homogeneous gradient initial model and inversion results. (a) Initial modeld. (b) 3DFWI. (c) 3DIFWI. (d) [PITH_FULL_IMAGE:figures/full_fig_p011_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: The slices results obtained from homogeneous gradient initialization model at Inline = 0.6 km, crossline = [PITH_FULL_IMAGE:figures/full_fig_p011_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Comparison of velocity curves. multi-scale structural features. As the σw increases, the low-frequency components of the local average amplitude and local energy data increase. This helps reduce the risk of cycle skipping during the inversion process. Figure 5a shows the true 3D Overthrust model, and Figure 5b is the constant gradient initial model. The inversion results obtained using 3DFWI, 3DIFWI, 3DIF… view at source ↗
Figure 15
Figure 15. Figure 15: The field data acquisition geometry [PITH_FULL_IMAGE:figures/full_fig_p014_15.png] view at source ↗
Figure 16
Figure 16. Figure 16: Field data from a single shot at different Crosslines. [PITH_FULL_IMAGE:figures/full_fig_p014_16.png] view at source ↗
Figure 17
Figure 17. Figure 17: Field data frequency spectrum [PITH_FULL_IMAGE:figures/full_fig_p015_17.png] view at source ↗
Figure 18
Figure 18. Figure 18: Field data velocity model. (a) Initial model. (b) TT-3DIFWI with M-SSIM inversion result. [PITH_FULL_IMAGE:figures/full_fig_p015_18.png] view at source ↗
Figure 19
Figure 19. Figure 19: TT-3DIFWI with M-SSIM slices inversion results from field data. (a) Inline direction. (b) Crossline direction. [PITH_FULL_IMAGE:figures/full_fig_p015_19.png] view at source ↗
Figure 20
Figure 20. Figure 20: Comparison of observed and predicted data. [PITH_FULL_IMAGE:figures/full_fig_p016_20.png] view at source ↗
Figure 21
Figure 21. Figure 21: Comparison of the well-log. reflections and refractions at long offsets. This high level of data consistency indicates that the inverted model effectively honors the wavefield physics recorded in the field. Finally, a quantitative validation is performed by comparing the inversion result with the available well-log data, as shown in [PITH_FULL_IMAGE:figures/full_fig_p017_21.png] view at source ↗
read the original abstract

Three-dimensional full waveform inversion (3DFWI) is a powerful technique for reconstructing high-resolution subsurface velocity models. However, its application is often limited by high memory requirements, computational costs, and sensitivity to cycle skipping. To overcome these challenges, we propose a novel tensor train (TT) decomposition-based 3D implicit full waveform inversion framework (TT-3DIFWI) combined with a multi-scale structural similarity (M-SSIM) objective function. In this framework, the 3D velocity model is represented by TT decomposition as a product of a series of low-rank core tensors. Then, three axis-specific implicit neural network representations (INR) based on one-dimensional vector coordinates as input are constructed to predict these core tensors, rather than directly predicting the velocity model. This INR reparameterization method based on TT decomposition can significantly reduce the memory consumption of INR training while maintaining the accuracy and resolution of the 3D velocity model reconstruction. Meanwhile, the low-rank structure of TT decomposition also ensures the structural consistency of the reconstruction velocity, thereby improving the accuracy and continuity of the inversion result. Furthermore, the M-SSIM objective function can compare the multi-scale structural differences between predicted and observed data, and utilize the ultra-low frequency features to reduce cycle skipping. Numerical experiments on synthetic and challenging land datasets demonstrate that TT-3DIFWI with M-SSIM achieves accurate and continuous velocity reconstruction, even with poor initial models or missing low-frequency data.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 1 minor

Summary. The manuscript proposes TT-3DIFWI, a 3D full-waveform inversion framework that represents the velocity model via tensor-train (TT) decomposition whose low-rank cores are predicted by three axis-specific implicit neural representations (INRs) rather than a single 3D INR. An M-SSIM loss is introduced to compare multi-scale structural features between predicted and observed data and thereby mitigate cycle skipping. The authors claim that the TT low-rank structure additionally enforces structural consistency, yielding accurate and continuous velocity models on both synthetic and land datasets even from poor initial models or band-limited data.

Significance. If the experimental claims are substantiated, the TT+axis-wise INR parameterization offers a memory-efficient route to high-resolution 3D FWI while the M-SSIM term addresses a long-standing robustness issue. The combination is technically novel within the geophysics literature and could be of practical interest for land datasets where low-frequency content is limited.

major comments (2)
  1. [Abstract, §3] Abstract and §3 (methods): the central assertion that "the low-rank structure of TT decomposition also ensures the structural consistency of the reconstruction velocity" is load-bearing for the claimed improvement in continuity, yet the reported experiments compare the full TT-3DIFWI+M-SSIM pipeline against baselines that differ simultaneously in objective function, parameterization, and decomposition. No ablation that varies TT rank while holding M-SSIM and INR architecture fixed is described, nor are quantitative continuity metrics (total variation, edge coherence, or structural similarity on the velocity model itself) supplied to isolate the TT contribution.
  2. [Results] Results section: the abstract states that "numerical experiments … demonstrate that TT-3DIFWI with M-SSIM achieves accurate and continuous velocity reconstruction," but supplies no quantitative metrics (RMSE, SSIM, or misfit values), baseline tables, error analysis, or details on the synthetic/land data sets, initial models, or frequency content. Without these, the soundness of the central empirical claim cannot be evaluated.
minor comments (1)
  1. [§3] Notation for the TT cores and the three axis-specific INRs should be introduced with explicit equations and dimension indices to avoid ambiguity when the reader compares the memory scaling claims.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback, which helps clarify the contributions and strengthen the empirical support in our manuscript. We address each major comment below and will incorporate revisions as noted.

read point-by-point responses

  1. Referee: [Abstract, §3] Abstract and §3 (methods): the central assertion that "the low-rank structure of TT decomposition also ensures the structural consistency of the reconstruction velocity" is load-bearing for the claimed improvement in continuity, yet the reported experiments compare the full TT-3DIFWI+M-SSIM pipeline against baselines that differ simultaneously in objective function, parameterization, and decomposition. No ablation that varies TT rank while holding M-SSIM and INR architecture fixed is described, nor are quantitative continuity metrics (total variation, edge coherence, or structural similarity on the velocity model itself) supplied to isolate the TT contribution.

    Authors: We agree that isolating the TT decomposition's contribution requires an ablation that varies only the TT rank while holding the M-SSIM loss and INR architecture fixed, and that quantitative continuity metrics on the velocity model (e.g., total variation or edge coherence) would directly support the structural-consistency claim. The current comparisons demonstrate overall pipeline improvement but do not separate the low-rank effect. We will add the requested ablation study and continuity metrics to the revised manuscript. revision: yes

  2. Referee: [Results] Results section: the abstract states that "numerical experiments … demonstrate that TT-3DIFWI with M-SSIM achieves accurate and continuous velocity reconstruction," but supplies no quantitative metrics (RMSE, SSIM, or misfit values), baseline tables, error analysis, or details on the synthetic/land data sets, initial models, or frequency content. Without these, the soundness of the central empirical claim cannot be evaluated.

    Authors: We acknowledge that the results section would benefit from explicit quantitative tables (RMSE, SSIM, data misfit) and expanded details on datasets, initial models, and frequency bands to allow direct evaluation of the claims. While the manuscript contains visual comparisons and some supporting numbers, a consolidated baseline table and error analysis are absent. We will add these elements in the revision. revision: yes

Circularity Check

0 steps flagged

No significant circularity in derivation chain

full rationale

The paper proposes TT-3DIFWI, a parameterization of the velocity model via tensor-train cores predicted by axis-specific INRs, optimized under an M-SSIM loss. The statement that low-rank TT structure ensures structural consistency is an appeal to a known algebraic property of TT decomposition rather than a result derived inside the paper. No equation reduces a claimed prediction to a fitted quantity by construction, no self-citation chain is load-bearing, and no ansatz is smuggled via prior work. The numerical experiments compare full pipelines but do not constitute a circular derivation; the method remains self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Based solely on the abstract; no explicit free parameters, axioms, or invented entities are identifiable. The approach relies on standard assumptions from neural network optimization and tensor decomposition literature, but details such as TT ranks or network architectures are absent.

pith-pipeline@v0.9.1-grok · 5801 in / 1059 out tokens · 41614 ms · 2026-06-26T06:13:04.900451+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

50 extracted references · 2 canonical work pages · 1 internal anchor

  1. [1]

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

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

    1984

  2. [2]

    Virieux and S

    J. Virieux and S. Operto. An overview of full-waveform inversion in exploration geophysics.GEOPHYSICS, 74(6):WCC1–WCC26, November 2009

    2009

  3. [3]

    Full waveform tomography for lithospheric imaging: results from a blind test in a realistic crustal model.Geophysical Journal International, 168(1):133–151, 2007

    AJ Brenders and RG Pratt. Full waveform tomography for lithospheric imaging: results from a blind test in a realistic crustal model.Geophysical Journal International, 168(1):133–151, 2007

    2007

  4. [4]

    Full-waveform inversion: The next leap forward in subsalt imaging.The Leading Edge, 37(1):67b1–67b6, 2018

    Xukai Shen, Imtiaz Ahmed, Andrew Brenders, Joe Dellinger, John Etgen, and Scott Michell. Full-waveform inversion: The next leap forward in subsalt imaging.The Leading Edge, 37(1):67b1–67b6, 2018

    2018

  5. [5]

    Reflection seismic waveform tomography.Journal of Geophysical Research: Solid Earth, 114(B3), 2009

    Yanghua Wang and Ying Rao. Reflection seismic waveform tomography.Journal of Geophysical Research: Solid Earth, 114(B3), 2009

    2009

  6. [6]

    Anisotropic 3d full-waveform inversion.Geophysics, 78(2):R59– R80, 2013

    Michael Warner, Andrew Ratcliffe, Tenice Nangoo, Joanna Morgan, Adrian Umpleby, Nikhil Shah, Vetle Vinje, Ivan Štekl, Lluís Guasch, Caroline Win, et al. Anisotropic 3d full-waveform inversion.Geophysics, 78(2):R59– R80, 2013

    2013

  7. [7]

    Correlation-based reflection full-waveform inversion.Geophysics, 80(4):R189–R202, 2015

    Benxin Chi, Liangguo Dong, and Yuzhu Liu. Correlation-based reflection full-waveform inversion.Geophysics, 80(4):R189–R202, 2015

    2015

  8. [8]

    Full-waveform inversion for salt: A coming of age.The Leading Edge, 38(3):204–213, 2019

    Ping Wang, Zhigang Zhang, Jiawei Mei, Feng Lin, and Rongxin Huang. Full-waveform inversion for salt: A coming of age.The Leading Edge, 38(3):204–213, 2019

    2019

  9. [9]

    Wavefield reconstruction inversion via physics-informed neural networks

    Chao Song and Tariq A Alkhalifah. Wavefield reconstruction inversion via physics-informed neural networks. IEEE Transactions on Geoscience and Remote Sensing, 60:1–12, 2021

    2021

  10. [10]

    Wave-equation traveltime inversion.Geophysics, 56(5):645–653, 1991

    Yi Luo and Gerard T Schuster. Wave-equation traveltime inversion.Geophysics, 56(5):645–653, 1991

    1991

  11. [11]

    Frequency-domain wave-equation traveltime inversion with a monofrequency component.GEOPHYSICS, 86(6):R913–R926, November 2021

    Jianhua Wang, Jizhong Yang, Liangguo Dong, and Yuzhu Liu. Frequency-domain wave-equation traveltime inversion with a monofrequency component.GEOPHYSICS, 86(6):R913–R926, November 2021

    2021

  12. [12]

    Wave-equation migration velocity analysis

    Paul Sava and Biondo Biondi. Wave-equation migration velocity analysis. i. theory.Geophysical Prospecting, 52(6):593–606, 2004

    2004

  13. [13]

    Migration velocity analysis and waveform inversion.Geophysical prospecting, 56(6):765–790, 2008

    William W Symes. Migration velocity analysis and waveform inversion.Geophysical prospecting, 56(6):765–790, 2008

    2008

  14. [14]

    Abdullah Alali, Bingbing Sun, and Tariq Alkhalifah. The effectiveness of a pseudo-inverse extended Born operator to handle lateral heterogeneity for imaging and velocity analysis applications.Geophysical Prospecting, 68(4):1154–1166, May 2020

    2020

  15. [15]

    Full waveform inversion for reflected seismic data

    Sheng Xu, D Wang, F Chen, Yu Zhang, and G Lambare. Full waveform inversion for reflected seismic data. In 74th EAGE Conference and Exhibition incorporating EUROPEC 2012, pages cp–293. European Association of Geoscientists & Engineers, 2012

    2012

  16. [16]

    Simultaneous inversion of the background velocity and the perturbation in full-waveform inversion.Geophysics, 80(6):R317–R329, 2015

    Zedong Wu and Tariq Alkhalifah. Simultaneous inversion of the background velocity and the perturbation in full-waveform inversion.Geophysics, 80(6):R317–R329, 2015

    2015

  17. [17]

    A review on reflection-waveform inversion.Petroleum Science, 17(2):334–351, April 2020

    Gang Yao, Di Wu, and Shang-Xu Wang. A review on reflection-waveform inversion.Petroleum Science, 17(2):334–351, April 2020. 18 APREPRINT- JUNE23, 2026

    2020

  18. [18]

    Physics- informed machine learning.Nature Reviews Physics, 3(6):422–440, 2021

    George Em Karniadakis, Ioannis G Kevrekidis, Lu Lu, Paris Perdikaris, Sifan Wang, and Liu Yang. Physics- informed machine learning.Nature Reviews Physics, 3(6):422–440, 2021

    2021

  19. [19]

    Sensing prior constraints in deep neural networks for solving exploration geophysical problems.Proceedings of the National Academy of Sciences, 120(23):e2219573120, 2023

    Xinming Wu, Jianwei Ma, Xu Si, Zhengfa Bi, Jiarun Yang, Hui Gao, Dongzi Xie, Zhixiang Guo, and Jie Zhang. Sensing prior constraints in deep neural networks for solving exploration geophysical problems.Proceedings of the National Academy of Sciences, 120(23):e2219573120, 2023

    2023

  20. [20]

    Review of physics-informed machine-learning inversion of geophysical data.Geophysics, 89(6):T337–T356, 2024

    Gerard T Schuster, Yuqing Chen, and Shihang Feng. Review of physics-informed machine-learning inversion of geophysical data.Geophysics, 89(6):T337–T356, 2024

    2024

  21. [21]

    Chao Song, Tariq Alkhalifah, Umair bin Waheed, Silin Wang, and Cai Liu. Physics-constrained deep-learning- based full-waveform inversion: A practical source-independent strategy with two real data applications.IEEE Transactions on Geoscience and Remote Sensing, 2026

    2026

  22. [22]

    Velocity model building with a modified fully convolutional network

    Wenlong Wang, Fangshu Yang, and Jianwei Ma. Velocity model building with a modified fully convolutional network. InSEG International Exposition and Annual Meeting, pages SEG–2018. SEG, 2018

    2018

  23. [23]

    Deep-learning inversion: A next-generation seismic velocity model building method.Geophysics, 84(4):R583–R599, 2019

    Fangshu Yang and Jianwei Ma. Deep-learning inversion: A next-generation seismic velocity model building method.Geophysics, 84(4):R583–R599, 2019

    2019

  24. [24]

    Inversionnet: An efficient and accurate data-driven full waveform inversion.IEEE Transactions on Computational Imaging, 6:419–433, 2019

    Yue Wu and Youzuo Lin. Inversionnet: An efficient and accurate data-driven full waveform inversion.IEEE Transactions on Computational Imaging, 6:419–433, 2019

    2019

  25. [25]

    Multiscale data-driven seismic full-waveform inversion with field data study.IEEE transactions on geoscience and remote sensing, 60:1–14, 2021

    Shihang Feng, Youzuo Lin, and Brendt Wohlberg. Multiscale data-driven seismic full-waveform inversion with field data study.IEEE transactions on geoscience and remote sensing, 60:1–14, 2021

    2021

  26. [26]

    Seismic Full-Waveform Inversion Using Deep Learning Tools and Techniques

    Alan Richardson. Seismic full-waveform inversion using deep learning tools and techniques.arXiv preprint arXiv:1801.07232, 2018

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  27. [27]

    Deep recurrent architectures for seismic tomography

    Amir Adler, Mauricio Araya-Polo, and Tomaso Poggio. Deep recurrent architectures for seismic tomography. In 81st EAGE conference and exhibition 2019, volume 2019, pages 1–5. European Association of Geoscientists & Engineers, 2019

    2019

  28. [28]

    Seismic velocity estimation: A deep recurrent neural-network approach

    Gabriel Fabien-Ouellet and Rahul Sarkar. Seismic velocity estimation: A deep recurrent neural-network approach. Geophysics, 85(1):U21–U29, 2020

    2020

  29. [29]

    Data-driven seismic waveform inversion: A study on the robustness and generalization.IEEE Transactions on Geoscience and Remote sensing, 58(10):6900–6913, 2020

    Zhongping Zhang and Youzuo Lin. Data-driven seismic waveform inversion: A study on the robustness and generalization.IEEE Transactions on Geoscience and Remote sensing, 58(10):6900–6913, 2020

    2020

  30. [30]

    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

  31. [31]

    Seismic impedance inversion based on cycle-consistent generative adversarial network.Petroleum Science, 19(1):147–161, 2022

    Yu-Qing Wang, Qi Wang, Wen-Kai Lu, Qiang Ge, and Xin-Fei Yan. Seismic impedance inversion based on cycle-consistent generative adversarial network.Petroleum Science, 19(1):147–161, 2022

    2022

  32. [32]

    Implicit seismic full waveform inversion with deep neural representation.Journal of Geophysical Research: Solid Earth, 128(3):e2022JB025964, 2023

    Jian Sun, Kristopher Innanen, Tianze Zhang, and Daniel Trad. Implicit seismic full waveform inversion with deep neural representation.Journal of Geophysical Research: Solid Earth, 128(3):e2022JB025964, 2023

    2023

  33. [33]

    Multilayer perceptron and bayesian neural network- based elastic implicit full waveform inversion.IEEE Transactions on Geoscience and Remote Sensing, 61:1–16, 2023

    Tianze Zhang, Jian Sun, Daniel Trad, and Kristopher Innanen. Multilayer perceptron and bayesian neural network- based elastic implicit full waveform inversion.IEEE Transactions on Geoscience and Remote Sensing, 61:1–16, 2023

    2023

  34. [34]

    Fwigan: Full-waveform inversion via a physics-informed generative adversarial network.Journal of Geophysical Research: Solid Earth, 128(4):e2022JB025493, 2023

    Fangshu Yang and Jianwei Ma. Fwigan: Full-waveform inversion via a physics-informed generative adversarial network.Journal of Geophysical Research: Solid Earth, 128(4):e2022JB025493, 2023

    2023

  35. [35]

    Gabor wavelet-activation implicit neural learning for full-waveform inversion

    Fangshu Yang and Jianwei Ma. Gabor wavelet-activation implicit neural learning for full-waveform inversion. Geophysics, 90(3):R71–R87, 2025

    2025

  36. [36]

    Velocity model building by 3d frequency-domain, full-waveform inversion of wide-aperture seismic data.Geophysics, 73(5):VE101–VE117, 2008

    Hafedh Ben-Hadj-Ali, Stéphane Operto, and Jean Virieux. Velocity model building by 3d frequency-domain, full-waveform inversion of wide-aperture seismic data.Geophysics, 73(5):VE101–VE117, 2008

    2008

  37. [37]

    Three-dimensional elastic full waveform inversion using seismic data from the sleipner area.Geophysical Journal International, 202(3):1877–1894, 2015

    Espen Birger Raknes, Børge Arntsen, and Wiktor Weibull. Three-dimensional elastic full waveform inversion using seismic data from the sleipner area.Geophysical Journal International, 202(3):1877–1894, 2015

    2015

  38. [38]

    Patrick Amestoy, Romain Brossier, Alfredo Buttari, Jean-Yves L’Excellent, Theo Mary, Ludovic Métivier, Alain Miniussi, and Stephane Operto. Fast 3d frequency-domain full-waveform inversion with a parallel block low-rank multifrontal direct solver: Application to obc data from the north sea.Geophysics, 81(6):R363–R383, 2016

    2016

  39. [39]

    Anisotropic 3d elastic full-wavefield inversion to directly estimate elastic properties and its role in interpretation

    Haiyang Wang, Olivier Burtz, Partha Routh, Don Wang, Jake Violet, Rongrong Lu, and Spyros Lazaratos. Anisotropic 3d elastic full-wavefield inversion to directly estimate elastic properties and its role in interpretation. The Leading Edge, 40(4):277–286, 2021

    2021

  40. [40]

    3-d bayesian variational full waveform inversion.Geophysical Journal International, 234(1):546–561, 2023

    Xin Zhang, Angus Lomas, Muhong Zhou, York Zheng, and Andrew Curtis. 3-d bayesian variational full waveform inversion.Geophysical Journal International, 234(1):546–561, 2023. 19 APREPRINT- JUNE23, 2026

    2023

  41. [41]

    M-ssim-3difwi: Three- dimensional implicit full waveform inversion based on the multi-scale structural similarity.IEEE Transactions on Geoscience and Remote Sensing, (99):1–1, 2026

    Liangsheng He, Chao Song, Tariq Alkhalifah, Tiansheng Chen, Tao Liu, and Cai Liu. M-ssim-3difwi: Three- dimensional implicit full waveform inversion based on the multi-scale structural similarity.IEEE Transactions on Geoscience and Remote Sensing, (99):1–1, 2026

    2026

  42. [42]

    Approximation capabilities of multilayer feedforward networks.Neural networks, 4(2):251–257, 1991

    Kurt Hornik. Approximation capabilities of multilayer feedforward networks.Neural networks, 4(2):251–257, 1991

    1991

  43. [43]

    Multilayer feedforward networks with a nonpolynomial activation function can approximate any function.Neural networks, 6(6):861–867, 1993

    Moshe Leshno, Vladimir Ya Lin, Allan Pinkus, and Shimon Schocken. Multilayer feedforward networks with a nonpolynomial activation function can approximate any function.Neural networks, 6(6):861–867, 1993

    1993

  44. [44]

    Integrating deep neural networks with full-waveform inversion: Reparameterization, regularization, and uncertainty quantification.Geophysics, 87(1):R93–R109, 2022

    Weiqiang Zhu, Kailai Xu, Eric Darve, Biondo Biondi, and Gregory C Beroza. Integrating deep neural networks with full-waveform inversion: Reparameterization, regularization, and uncertainty quantification.Geophysics, 87(1):R93–R109, 2022

    2022

  45. [45]

    Implicit neural representations with periodic activation functions.Advances in neural information processing systems, 33:7462– 7473, 2020

    Vincent Sitzmann, Julien Martel, Alexander Bergman, David Lindell, and Gordon Wetzstein. Implicit neural representations with periodic activation functions.Advances in neural information processing systems, 33:7462– 7473, 2020

    2020

  46. [46]

    MS_ATpV-FWI: Full waveform inversion based on multi-scale structural similarity index measure and anisotropic total p-variation regularization.arXiv preprint arXiv:2504.01695, 2025

    Liangsheng He, Chao Song, and Cai Liu. MS_ATpV-FWI: Full waveform inversion based on multi-scale structural similarity index measure and anisotropic total p-variation regularization.arXiv preprint arXiv:2504.01695, 2025

    work page arXiv 2025

  47. [47]

    Local-crosscorrelation elastic full-waveform inversion.Geophysics, 84(6):R897–R908, 2019

    Zhendong Zhang and Tariq Alkhalifah. Local-crosscorrelation elastic full-waveform inversion.Geophysics, 84(6):R897–R908, 2019

    2019

  48. [48]

    Toward high-fidelity imaging: Dynamic matching fwi and its applications.The Leading Edge, 42(2):124–132, 2023

    Yi Huang, Jian Mao, James Sheng, Mike Perz, Yang He, Feng Hao, Faqi Liu, Bin Wang, Seet Li Yong, Daniel Chaikin, et al. Toward high-fidelity imaging: Dynamic matching fwi and its applications.The Leading Edge, 42(2):124–132, 2023

    2023

  49. [49]

    Full-intensity waveform inversion

    Yike Liu, Bin He, Huiyi Lu, Zhendong Zhang, Xiao-Bi Xie, and Yingcai Zheng. Full-intensity waveform inversion. Geophysics, 83(6):R649–R658, 2018

    2018

  50. [50]

    Addressing cycle-skipping in full-waveform inversion using acoustic wave energy

    Zhonglei Li and Gang Yao. Addressing cycle-skipping in full-waveform inversion using acoustic wave energy. Geophysical Prospecting, 73(3):960–976, 2025. 20

    2025