Pith. sign in

REVIEW 5 major objections 4 minor 39 references

Operator learning regularization for macroscopic permeability prediction in dual-scale flow problem

T0 review · 5 major / 4 minor · reviewed 2026-08-12 · deepseek-v4-flash

Pith's one-line read A neural operator trained with H2 regularization maps the Stokes-Brinkman coefficient to a velocity field and recovers macroscopic permeability with a relative log-error of about 0.09.

desk verdict Useful loss-function benchmark for neural-operator permeability prediction, but the 1e-12 GRF standard deviation makes the test fields effectively binary, so the 'heterogeneous local permeability' claim is overstated. read the letter →

arxiv 2412.00579 v1 pith:HNG6TZQU submitted 2024-11-30 physics.flu-dyn cs.LGcs.NAmath.NAphysics.comp-ph

classification physics.flu-dyncs.LGcs.NAmath.NAphysics.comp-ph PACS 47.56.+r
keywords Dual-scaleflowPorousmediaNeuraloperatorCompositemanufacturingPermeabilityStokes-BrinkmanequationFourierLossregularization
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 replaces the expensive Stokes-Brinkman solver with a neural operator that maps the heterogeneous coefficient $\beta$ to the velocity field, then recovers the macroscopic permeability tensor by upscaling. It claims the choice of loss is decisive: penalizing the second derivatives of the predicted velocity field during training, through an $H^2$ loss or an $H^1$ loss with additional $\beta u$ and Laplacian terms, markedly improves the computed permeability. The best Tensorized Fourier Neural Operator achieves a relative log-permeability error of $0.0913$ on the test set and degrades only slightly under zero-shot super-resolution from $64\times64$ to $128\times128$. A fast, accurate surrogate for permeability matters because direct numerical simulation of dual-scale fibrous media is computationally prohibitive for industrial process design.

What carries the argument

The central objects are the Stokes-Brinkman equation $\varphi\Delta u - \beta u - \nabla p = 0$ with periodic conditions, and a two-step surrogate: a Fourier neural operator (a network whose integral kernel acts by multiplication in Fourier space, here the Tensorized Fourier Neural Operator with Tucker factorization) that approximates the map $\beta \mapsto u$, followed by Darcy upscaling $U = -K G/\mu$ with $G = \langle \varphi\Delta u \rangle_{\Omega} - \langle \beta u \rangle_{\Omega}$. The loss functions carry the argument: relative $L^2$, $H^1$, $H^2$, and variants regularized by $\beta u$ and $\Delta u$ terms, with $H^2$ and $\beta u\Delta u$-regularized $H^1$ performing best for permeability. The model's resolution independence is what allows zero-shot super-resolution tests from a $64\times64$ to a $128\times128$ grid.

What would settle it

Run the same TFNO-$H^2$ training on a dataset whose local permeability maps have standard deviation varying over several orders of magnitude rather than fixed at $10^{-12}$, and compare the relative $\log K$ error on the test set. If the error rises substantially above $0.0913$, the claim that the method handles heterogeneous dual-scale fabrics is not supported.

Watch

Extended reading notes

Core claim

The central claim is that regularizing the second derivatives of the velocity field is what makes operator learning viable for the dual-scale permeability problem. Directly learning the map $\beta \mapsto K$ proved unstable because it requires inverting the macroscopic pressure gradient, and an unregularized $L^2$ fit to the velocity field yields poor permeability values. The paper establishes that the $H^2$ loss, which explicitly penalizes differences in the second partial derivatives of predicted and ground-truth velocity, gives the best macroscopic permeability, with a relative supremum-norm error $J = \|\log K_{\text{pred}} - \log K_{\text{GT}}\|/\|\log K_{\text{GT}}\|$ of $0.0913$ for the TFNO model. The authors conclude that this highlights the importance of second-derivative regularization because computing $K$ involves applying the Laplacian to the velocity field.

Load-bearing premise

The claim assumes the Gaussian random field samples, with the standard deviation of $\log(k_s)$ fixed at $10^{-12}$, are representative of the heterogeneous dual-scale porous media the model is meant to predict.

Editorial extensions

If this is right

  • The TFNO-$H^2$ surrogate computes macroscopic permeability at near-zero marginal cost, making it usable in design loops where thousands of Stokes-Brinkman solves would be too slow.
  • Because resolution independence persists, a model trained on coarse grids can be evaluated on finer microstructural images, easing transfer from simulation grids to high-resolution fabric scans.
  • The result identifies second-derivative fidelity, not just pointwise velocity accuracy, as the quantity neural operators should preserve when the observable of interest involves the Laplacian of the field.
  • The same two-step strategy (predict the field with a neural operator, then apply the physical upscaling operator) is directly applicable to other upscaling problems governed by PDEs with heterogeneous coefficients.

Reading between the lines

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

  • Inference: the 800-sample training set, with the standard deviation of $\log(k_s)$ fixed at $10^{-12}$, effectively makes each local permeability field nearly constant, so the reported error is a benchmark on near-binary $\beta$ fields rather than on fields with local permeability varying over many orders of magnitude.
  • Inference: because the best losses explicitly target $\Delta u$, the same regularization principle could transfer to operator learning for other elliptic problems where the quantity of interest is a functional of the solution's second derivatives, such as stress or dissipation.
  • Inference: a natural extension is to train with the standard deviation of the Gaussian random field sampled over several orders of magnitude, which would test whether the $H^2$ advantage persists when local permeability genuinely varies across the domain.
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

5 major / 4 minor

Summary. The manuscript proposes a two-stage surrogate for the Stokes–Brinkman dual-scale flow problem: a Fourier neural operator (FNO/TFNO) maps the heterogeneous coefficient β to a velocity field, from which the macroscopic permeability K is recovered by volume averaging and Darcy upscaling. Using 800 generated samples, the authors compare seven loss formulations (L2, H1, βu-regularized H1, βu-plus-Laplacian H1, H2, βu-regularized H2, and macroscopic pressure-gradient). On the test set, the TFNO trained with the H2 loss achieves the lowest reported log-permeability error (J = 0.0913) and retains accuracy in zero-shot super-resolution from 64×64 to 128×128 grids. The central claim is that regularizing second derivatives of the velocity during training significantly improves permeability prediction.

Significance. If the result holds, the paper would be a useful step toward cheap permeability surrogates for liquid composite moulding, and the H2-regularization insight is physically plausible because computing K involves the Laplacian of the velocity field. The zero-shot super-resolution result is practically attractive, and the empirical comparison in Tables 1 and 2 is internally consistent. However, the significance is currently limited by three issues: the input fields are effectively binary due to the 1e-12 GRF standard deviation, no error bars or seed statistics are given, and several loss-function hyperparameters are unreported. These gaps make it difficult to assess whether the method transfers to genuinely heterogeneous dual-scale fabrics and whether the reported rankings are robust.

major comments (5)
  1. [§3 (Data generation)] The choice of Gaussian random field standard deviation fixed to 10^{-12} makes log(k_s) essentially constant within each generated realization, so the thresholded input β is a binary mask with a single solid-phase value. The manuscript repeatedly emphasizes 'extreme heterogeneity spanning several orders of magnitude', but that heterogeneity is present only across the 800 samples (through the mean parameter), not inside any field, and the operator learning task reduces to geometry segmentation plus a scalar magnitude. The central claim that the method addresses the challenging dual-scale flow problem with large local permeability variation is therefore not supported for the stated problem class. Please either justify the 1e-12 choice, generate fields with internal log-permeability variation spanning several orders of magnitude within a single realization, and report errors on such fields, or explicitly reframe the contribution as a two-phase geometry problem.
  2. [§5, Tables 1 and 2] All reported errors come from a single training run for each loss/architecture combination, with no seed averaging, error bars, or significance testing. Several differences that drive the central conclusion are small (e.g., TFNO H2 J=0.0913 versus H1βuΔu J=0.0979 versus L2 J=0.1018 in Table 2), so the ranking may be within training noise. The claim that regularising second derivatives 'significantly enhances' the results needs repeated-seed statistics (mean ± standard deviation over at least three seeds) or a paired test across the fixed test set.
  3. [§5.2, Table 2] The metric J := ∥ log(K_pred) − log(K_GT) ∥ / ∥ log(K_GT) ∥ with ∥K∥ = max_{i,j} |k_ij| is not well defined if any component of the permeability tensor is negative, and a logarithm of a matrix/tensor is not standard. For rotated anisotropic media the off-diagonal components of K can be negative, so the paper must specify the exact elementwise transformation (e.g., log of absolute values, log of eigenvalues, or another positive quantity) and confirm that it is defined on all 200 test samples.
  4. [§4, Eqs. (15), (16), (19)] The regularization weights λ, λ1, and λ2 are introduced in the loss definitions but their values (or tuning procedure) are never reported. Since the central comparison is between different regularizations, the absence of these hyperparameters makes the experiments irreproducible and prevents the reader from separating the effect of the loss form from the effect of the weight choice.
  5. [§3–§4] The neural operator input is described as log(β) or log10(β), while β = 0 in the clear-fluid regions. The paper does not state how the zero values are encoded (e.g., a large negative sentinel, a mask channel, beta+epsilon), which is essential for reproducing the model and for interpreting Figures 3–4.
minor comments (4)
  1. [§5.2, §5.3, Table 2] There are several typos: 'Since we our primary motivation' should be 'Since our primary motivation', 'corse grained' should be 'coarse grained', and 'surpremum' should be 'supremum'.
  2. [§3] The sentence 'We generate one sample for each set of parameters' implies that the GRF randomness is effectively unused; please clarify how the 80% quantile thresholding interacts with the near-constant field and how the mask geometry is controlled.
  3. [§5.1, Figures 3 and 4] The visual comparison is based on a single test/training sample per model; please state whether the displayed examples are representative and/or provide quantitative error maps for the full test set.
  4. [§4, Eq. (18)] The mixed-derivative term in the definition of the H2 norm appears to be missing the ∂² symbol in the typeset formula; please correct the notation so that all second partial derivatives are explicitly shown.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the permeability prediction is a supervised operator-learning pipeline evaluated on held-out solver data, and the H2 regularization comparison is not forced by the loss construction.

full rationale

The paper's derivation chain is: generate heterogeneous coefficient beta from Gaussian random fields, solve the Stokes-Brinkman equation with an FFT solver to obtain the velocity field u, and compute the macroscopic permeability K via the upscaled Darcy law (Eqs. 5-6). The neural operator is trained on input-output pairs (beta, u) from this solver and then K is computed from the predicted u by the same post-processing formula. This is ordinary supervised learning: the network never fitted K directly in the main H2-loss experiment, and the reported best model uses an H2 loss that penalizes second derivatives of the velocity field, not the permeability target. The H2 loss is physically motivated because K depends on the Laplacian of u, but this is a causal link, not a circular reduction. The macroscopic-pressure-gradient loss (l_G) does include a K-related quantity in the training objective, but the paper's central claim (that H2 regularization performs best) is established independently of that loss, and the G-loss model is reported separately as a comparison. The FFT solver is cited from prior work [8] that includes one of the authors, but it is an open-source numerical method (AMITEX) and is used as a data generator, not as an unverified premise that presupposes the conclusions. No equation is defined in terms of the quantity it is supposed to predict, and no fitted parameter is renamed as a prediction. The data-generation caveat that the Gaussian random field standard deviation is fixed to 1e-12, making beta nearly binary, is a legitimate concern about generalizability to truly heterogeneous fabrics, but it is a limitation of the dataset, not a circularity in the derivation. Therefore the paper is self-contained as an empirical operator-learning study and merits a circularity score of 0.

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

The central claim rests on a simulation model, a hand-chosen threshold, and unreported regularization weights; no new physical entities are introduced.

free parameters (4)
  • regularization weight lambda (beta u term) = not reported
    Appears in losses (15) and (19); the reported results depend on this weight but its value is not given.
  • regularization weights lambda1 and lambda2 (beta u and Laplacian terms) = not reported
    Appear in loss (16); values are not stated, so the H1 beta u Delta u results cannot be reproduced.
  • clear-fluid threshold quantile = 80%
    Section 3: the 80% quantile of mu divided by ks determines which pixels are macropores; this hand-chosen value shapes the dataset and is not varied.
  • Gaussian random field standard deviation = 1e-12
    Section 3: the stated value would make log(ks) fields nearly constant, and the paper does not explain how this still yields heterogeneous fields.
assumptions (4)
  • domain assumption The Stokes-Brinkman equation (1) is a valid unified model of dual-scale flow in this setting.
    Sections 1 and 2 invoke prior literature; the paper does not validate the model against experiments or a different solver.
  • domain assumption The AMITEX FFT solver produces accurate ground-truth velocity fields for training and testing.
    Section 3 relies on the solver without a convergence study or independent reference in this paper.
  • domain assumption Macroscopic permeability K computed from equations (5) and (6) via volume averaging and upscaled Darcy's law is the correct target.
    Section 2 assumes this homogenization; it is the quantity the whole benchmark optimizes.
  • standard math With periodic boundary conditions and constant effective viscosity, the volume average of Delta u is zero, so the first term in equation (5) does not contribute.
    This follows from the divergence theorem; it matters because the paper motivates H2 losses by the need to compute the Laplacian term in G.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Operator learning regularization for macroscopic permeability prediction in dual-scale flow problem." pith.science (2026). https://pith.science/paper/HNG6TZQU

@misc{pith2026241200579,
  author       = {Pith},
  title        = {Pith review of: Operator learning regularization for macroscopic permeability prediction in dual-scale flow problem},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/HNG6TZQU}},
  note         = {Machine review of arXiv:2412.00579}
}
abstract

Liquid composites moulding is an important manufacturing technology for fibre reinforced composites, due to its cost-effectiveness. Challenges lie in the optimisation of the process due to the lack of understanding of key characteristic of textile fabrics - permeability. The problem of computing the permeability coefficient can be modelled as the well-known Stokes-Brinkman equation, which introduces a heterogeneous parameter $\beta$ distinguishing macropore regions and fibre-bundle regions. In the present work, we train a Fourier neural operator to learn the nonlinear map from the heterogeneous coefficient $\beta$ to the velocity field $u$, and recover the corresponding macroscopic permeability $K$. This is a challenging inverse problem since both the input and output fields span several order of magnitudes, we introduce different regularization techniques for the loss function and perform a quantitative comparison between them.

Figures

Figures reproduced from arXiv: 2412.00579 by the authors.

Figure 1
Figure 1. Distribution of heterogeneous coefficient [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Examples of generated input: left – map of local permeability [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Comparison of ground truth (GT) velocity field u1 and u2 vs. the output of the FNO model trained on the seven different loss functions (top on and bottom plots) depicted in section 4 for an example input log10 β of the test dataset. The best results for this example can be achieved for training with a H1βu∆u loss function or when optimizing with a H1βu loss. The neural operators that have been trained on the H1 loss… view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Comparison of ground truth (GT) velocity field u1 and u2 vs. the output of the TFNO model trained on the seven different loss functions depicted in Subsection 4 for an example input log10 β of the test dataset. The best results for this example can be achieved for trai…
Figure 5
Figure 5. Figure 5: Ground truth macroscopic permeability KGT (x-axis) vs. predicted Kpred (y￾axis) for the best performing model. Training the TFNO model with an H2 loss function yields the best results out of all the models and regularizations tested. Except for a few outliers, the comp…
Figure 6
Figure 6. Figure 6: Ground truth macroscopic permeability KGT (x-axis) vs. predicted Kpred (y-axis) for the models performing second to fourth best. Training the TFNO model with the H1 loss with βu∆u-regularization term yields the second best results. Using the L 2 loss function to train …
Figure 7
Figure 7. Figure 7: Ground truth macroscopic permeability KGT (x-axis) vs. predicted Kpred (y￾axis) for TFNO model and zero-shot super-resolution. We train the model with an H2 loss on 64 × 64 grid, while testing on 128 × 128 grid. The values for the macroscopic permeability coefficient e…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

39 extracted references · 34 canonical work pages

  1. [1]

    H. C. Brinkman, A calculation of the viscous force exerted by a flowing fluid on a dense swarm of particles, Flow, Turbulence and Combustion 1 (1) (1949) 27–34. 19

  2. [2]

    H. C. Brinkman, On the permeability of media consisting of closely packed porous particles, Flow, Turbulence and Combustion 1 (1) (1949) 81–86

  3. [3]

    W. R. Hwang, S. G. Advani, Numerical simulations of stokes–brinkman equations for permeability prediction of dual scale fibrous porous media, Physics of Fluids 22 (11) (2010)

  4. [4]

    Mezhoud, V

    S. Mezhoud, V. Monchiet, M. Bornert, D. Grande, Computation of macroscopic permeability of doubly porous media with fft based numer- ical homogenization method, European Journal of Mechanics-B/Fluids 83 (2020) 141–155

  5. [5]

    X. Xie, J. Xu, G. Xue, Uniformly-stable finite element methods for darcy-stokes-brinkman models, Journal of Computational Mathematics (2008) 437–455

  6. [6]

    J. A. Evans, T. J. Hughes, Isogeometric divergence-conforming b-splines for the darcy–stokes–brinkman equations, Mathematical Models and Methods in Applied Sciences 23 (04) (2013) 671–741

  7. [7]

    N. S. Martys, Improved approximation of the brinkman equation using a lattice boltzmann method, Physics of fluids 13 (6) (2001) 1807–1810

  8. [8]

    Y. Chen, High-performance computational homogenization of stokes– brinkman flow with an anderson-accelerated fft method, International Journal for Numerical Methods in Fluids 95 (9) (2023) 1441–1467

Show all 39 references
  1. [9]

    Boull´ e, A

    N. Boull´ e, A. Townsend, A mathematical guide to operator learning, in: Handbook of Numerical Analysis, Vol. 25, Elsevier, 2024, pp. 83–125

  2. [10]

    N. B. Kovachki, S. Lanthaler, A. M. Stuart, Operator learning: Al- gorithms and analysis, in: Handbook of Numerical Analysis, Vol. 25, Elsevier, 2024, pp. 419–467

  3. [11]

    Kovachki, Z

    N. Kovachki, Z. Li, B. Liu, K. Azizzadenesheli, K. Bhattacharya, A. Stu- art, A. Anandkumar, Neural operator: Learning maps between function spaces with applications to PDEs, J. Mach. Learn. Res. 24 (89) (2023) 1–97. 20

  4. [12]

    D. Z. Huang, N. H. Nelsen, M. Trautner, An operator learning perspec- tive on parameter-to-observable maps, Found. Data Sci. (2024)

  5. [13]

    H. F. Walker, P. Ni, Anderson acceleration for fixed-point iterations, SIAM Journal on Numerical Analysis 49 (4) (2011) 1715–1735

  6. [14]

    Y. Chen, L. G´ el´ ebart, C. Chateau, M. Bornert, C. Sauder, A. King, Analysis of the damage initiation in a sic/sic composite tube from a direct comparison between large-scale numerical simulation and syn- chrotron x-ray micro-computed tomography, International Journal of Sol...

  7. [15]

    M. L. Ravalec, B. Noetinger, L. Y. Hu, The fft moving average (fft-ma) generator: An efficient numerical method for generating and condition- ing gaussian simulations, Mathematical Geology 32 (2000) 701–723

  8. [16]

    C. K. Williams, C. E. Rasmussen, Gaussian Processes for Machine Learning, MIT Press, 2006

  9. [17]

    Amitex home page, https://amitexfftp.github.io/AMITEX/index. html

  10. [18]

    L. Lu, P. Jin, G. Pang, Z. Zhang, G. E. Karniadakis, Learning nonlinear operators via deeponet based on the universal approximation theorem of operators, Nat. Mach. Intell. 3 (3) (2021) 218–229

  11. [19]

    Kontolati, S

    K. Kontolati, S. Goswami, M. D. Shields, G. E. Karniadakis, On the influence of over-parameterization in manifold based surrogates and deep neural operators, J. Comput. Phys. 479 (2023) 112008

  12. [20]

    L. Lu, X. Meng, S. Cai, Z. Mao, S. Goswami, Z. Zhang, G. E. Karni- adakis, A comprehensive and fair comparison of two neural operators (with practical extensions) based on fair data, Comput. Methods Appl. Mech. Eng. 393 (2022) 114778

  13. [21]

    Kissas, J

    G. Kissas, J. H. Seidman, L. F. Guilhoto, V. M. Preciado, G. J. Pap- pas, P. Perdikaris, Learning operators with coupled attention, J. Mach. Learn. Res. 23 (215) (2022) 1–63

  14. [22]

    Oommen, K

    V. Oommen, K. Shukla, S. Goswami, R. Dingreville, G. E. Karniadakis, Learning two-phase microstructure evolution using neural operators and autoencoder architectures, Npj Comput. Mater. 8 (1) (2022) 190. 21

  15. [23]

    M. V. de Hoop, D. Z. Huang, E. Qian, A. M. Stuart, The cost-accuracy trade-off in operator learning with neural networks, J. Mach. Learn. 1 (3) (2022) 299–341

  16. [24]

    P. Jin, S. Meng, L. Lu, Mionet: Learning multiple-input operators via tensor product, SIAM J. Sci. Comput. 44 (6) (2022) A3490–A3514

  17. [25]

    Goswami, D

    S. Goswami, D. S. Li, B. V. Rego, M. Latorre, J. D. Humphrey, G. E. Karniadakis, Neural operator learning of heterogeneous mechanobiologi- cal insults contributing to aortic aneurysms, J. R. Soc. Interface 19 (193) (2022) 20220410

  18. [26]

    Y. Zhu, N. Zabaras, P.-S. Koutsourelakis, P. Perdikaris, Physics- constrained deep learning for high-dimensional surrogate modeling and uncertainty quantification without labeled data, J. Comput. Phys. 394 (2019) 56–81

  19. [27]

    S. Wang, H. Wang, P. Perdikaris, Learning the solution operator of para- metric partial differential equations with physics-informed DeepONets, Sci. Adv. 7 (40) (2021) eabi8605

  20. [28]

    Goswami, M

    S. Goswami, M. Yin, Y. Yu, G. E. Karniadakis, A physics-informed variational deeponet for predicting crack path in quasi-brittle materials, Comput. Methods Appl. Mech. Eng. 391 (2022) 114587

  21. [29]

    Z. Li, N. Kovachki, K. Azizzadenesheli, B. Liu, K. Bhattacharya, A. Stu- art, A. Anandkumar, Neural operator: Graph kernel network for partial differential equations, arXiv preprint arXiv:2003.03485 (2020)

  22. [30]

    H. You, Y. Yu, M. D’Elia, T. Gao, S. Silling, Nonlocal kernel network (nkn): A stable and resolution-independent deep neural network, J. Comput. Phys. 469 (2022) 111536

  23. [31]

    Z. Li, N. Kovachki, K. Azizzadenesheli, B. Liu, K. Bhattacharya, A. Stu- art, A. Anandkumar, Fourier neural operator for parametric partial dif- ferential equations, in: International Conference on Learning Represen- tations, 2021

  24. [32]

    Tripura, S

    T. Tripura, S. Chakraborty, Wavelet neural operator: a neural operator for parametric partial differential equations, Comput. Methods Appl. Mech. Eng. 404 (2023) 115783. 22

  25. [33]

    H. You, Q. Zhang, C. J. Ross, C.-H. Lee, Y. Yu, Learning deep im- plicit Fourier neural operators (IFNOs) with applications to hetero- geneous material modeling, Comput. Methods Appl. Mech. Eng. 398 (2022) 115296

  26. [34]

    Z. Li, H. Zheng, N. Kovachki, D. Jin, H. Chen, B. Liu, K. Azizzade- nesheli, A. Anandkumar, Physics-informed neural operator for learning partial differential equations, ACM/JMS J. Data Sci. 1 (3) (2024) 1–27

  27. [35]

    Konuk, J

    T. Konuk, J. Shragge, Physics-guided deep learning using fourier neural operators for solving the acoustic vti wave equation, in: 82nd EAGE Annual Conference & Exhibition, European Association of Geoscientists & Engineers, 2021, pp. 1–5

  28. [36]

    Hendrycks, K

    D. Hendrycks, K. Gimpel, Gaussian error linear units (gelus), arXiv preprint arXiv:1606.08415 (2016)

  29. [37]

    Kossaifi, N

    J. Kossaifi, N. Kovachki, K. Azizzadenesheli, A. Anandkumar, Multi- grid tensorized fourier neural operator for high-resolution pdes, arXiv preprint arXiv:2310.00120 (2023)

  30. [38]

    F. L. Hitchcock, The expression of a tensor or a polyadic as a sum of products, J. Math. Phys. 6 (1-4) (1927) 164–189

  31. [39]

    D. P. Kingma, Adam: A method for stochastic optimization, arXiv preprint arXiv:1412.6980 (2014). 23

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.