pith. sign in

arxiv: 2605.24658 · v1 · pith:UYQ4YSAFnew · submitted 2026-05-23 · 💻 cs.LG

WLNO: Wavelet-Laplace Neural Operator for Solving Partial Differential Equations

Muhammad Abid , Arth Sojitra , Omer San This is my paper

Pith reviewed 2026-06-30 14:54 UTC · model grok-4.3

classification 💻 cs.LG
keywords neural operatorswaveletspartial differential equationsmachine learningLaplace transformmulti-scale decompositionHaar waveletPDE solving
0
0 comments X

The pith

Fusing a Haar wavelet branch with the Laplace Neural Operator improves performance on PDE problems with multi-scale features.

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

The paper introduces WLNO as an extension to LNO that incorporates a parallel Haar wavelet transform branch to capture spatially localized multi-scale features in PDE solutions. This branch decomposes the feature map into four subbands and processes them with separate convolutions before inverse transform and fusion via a learnable gate. Evaluation on five standard PDE benchmarks shows consistent gains over LNO, particularly on problems like Burgers equation and Navier-Stokes with sharp or vortical structures. A reader would care if they work with PDEs where scale separation matters for accurate prediction of dynamics.

Core claim

The central discovery is that adding a single-level Haar DWT branch with independent 1x1 convolutions on subbands, fused by a sigmoid-gated weight initialized small, to the LNO's pole-residue formulation leads to better operator learning for transient and steady-state PDE dynamics, especially where spatial multi-scale structure is prominent.

What carries the argument

The parallel wavelet branch using Haar DWT and inverse DWT with learned convolutions per subband, adaptively weighted against the Laplace branch.

If this is right

  • WLNO achieves higher accuracy than LNO on the same training data and hyperparameters for diffusion, Burgers, reaction-diffusion, Darcy flow, and 2D Navier-Stokes equations.
  • The gains are largest for problems with sharp shock fronts and coherent vortical structures.
  • The learnable gate allows the model to balance the contribution of the wavelet branch during training.
  • The method provides an explicit way to extract multi-scale spatial features missing in pure Laplace-domain approaches.

Where Pith is reading between the lines

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

  • Similar wavelet augmentations might enhance other neural operator architectures beyond LNO.
  • Applying this to time-dependent problems with evolving multi-scale features could be a natural next test.
  • The approach hints at benefits from combining different frequency decompositions in operator learning.

Load-bearing premise

That the observed improvements result from the wavelet branch and not from implementation variations or training differences despite claims of identical protocols.

What would settle it

Running the exact same LNO and WLNO models multiple times with reported random seeds and checking if the performance difference remains significant and consistent across runs.

Figures

Figures reproduced from arXiv: 2605.24658 by Arth Sojitra, Muhammad Abid, Omer San.

Figure 1
Figure 1. Figure 1: Architecture of WLNO. (a) Full pipeline: the input function f(x, t) is concatenated with spatial grid coordinates (ξ1, ξ2) and lifted to latent space v ∈ R dz by P. The WLNO layer processes v through two parallel branches, PR2d Laplace and Haar DWT wavelet, whose outputs are fused via the learnable weight αwav and added to the local bypass Wloc. The result u is projected to the output field y(x, t) by Q. (… view at source ↗
Figure 2
Figure 2. Figure 2: Prediction comparison for one representative diffusion equation test sample, showing the input [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Test error (mean ± std) over 130 diffusion equation test samples. WLNO reduces mean error by 32.3% and standard deviation by 8.4% relative to LNO. 4.3 Burgers Equation The Burgers equation controls nonlinear convection-diffusion phenomena and functions as a standard test for methods that need to manage sudden gradients and shock-like characteristics: ∂u ∂t + u ∂u ∂x = ν ∂ 2u ∂x2 + f(x, t), (21) where ν is … view at source ↗
Figure 4
Figure 4. Figure 4: Training and validation history for the Burgers equation over 1000 epochs. WLNO maintains lower loss [PITH_FULL_IMAGE:figures/full_fig_p011_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Prediction comparison for two representative Burgers equation test samples, showing ground truth, LNO [PITH_FULL_IMAGE:figures/full_fig_p011_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Test error (mean ± std) over 100 Burgers equation test samples. WLNO reduces mean error by 22.8% and standard deviation by 39.3% relative to LNO, confirming that wavelet detail subbands are most effective when the solution contains sharp shock fronts. 4.4 Reaction-Diffusion Equation The reaction-diffusion equation describes the spatiotemporal evolution of a concentration field under the combined influence … view at source ↗
Figure 7
Figure 7. Figure 7: Prediction comparison for one representative reaction-diffusion test sample, showing the ground truth, [PITH_FULL_IMAGE:figures/full_fig_p013_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: shows the test error comparison aggregated over all 130 test samples. WLNO achieves 1.1896 × 10−1 ± 8.6830 × 10−2 compared to LNO’s 1.3848 × 10−1 ± 9.8000 × 10−2 , a +14.1% improvement in mean error with an 11.4% reduction in standard deviation. While this improvement is smaller than those observed for the diffusion and Burgers problems, it remains consistent and meaningful, confirming that the wavelet bra… view at source ↗
Figure 9
Figure 9. Figure 9: Prediction comparison for a representative Darcy flow test sample, showing the input permeability field [PITH_FULL_IMAGE:figures/full_fig_p014_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Test error (mean ± std) over the Darcy flow test samples. WLNO reduces mean error by 12.9% and standard deviation by 20.3% relative to LNO. 4.6 Two-Dimensional Navier-Stokes Equation To further evaluate the capability of WLNO on nonlinear turbulent flow dynamics, we additionally consider the two-dimensional incompressible Navier-Stokes equation in vorticity-streamfunction formulation on the periodic domai… view at source ↗
Figure 11
Figure 11. Figure 11: Prediction comparison for one representative Navier-Stokes test sample, showing the input forcing field [PITH_FULL_IMAGE:figures/full_fig_p016_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Test error (mean ± std) over the Navier-Stokes test samples. WLNO reduces mean error by 20.7% and standard deviation by 34.5% relative to LNO, demonstrating the benefit of wavelet spatial decomposition for turbulent flow operator learning. Across all five benchmark problems, WLNO consistently outperforms LNO with mean relative L2 error improvements of +32.3%, +22.8%, +14.1%, +12.9%, and +20.7% on the Diff… view at source ↗
read the original abstract

This work introduces the Wavelet-Laplace Neural Operator (WLNO), a novel neural operator that fuses Haar wavelet multi-scale spatial decomposition with the Laplace-domain pole-residue formulation of the Laplace Neural Operator (LNO). While LNO captures transient and steady-state dynamics through learnable system poles and residues, it lacks an explicit mechanism for extracting spatially localized multi-scale features inherent in complex PDE solutions. WLNO addresses this by augmenting the LNO core with a parallel single-level Haar discrete wavelet transform (DWT) branch that decomposes the lifted feature map into four frequency subbands: approximation (LL), horizontal detail (LH), vertical detail (HL), and diagonal detail (HH) and applies independent learned $1\times1$ convolutions to each subband before reconstruction via the inverse DWT. The two branches are fused through a learnable sigmoid-gated weight $\alpha_\mathrm{wav}$, initialized to give a small initial contribution to the wavelet branch, allowing the model to adaptively balance Laplace-domain dynamics against spatial multi-scale features throughout training. WLNO is evaluated against LNO on five benchmark PDE problems using identical hyperparameters, training data, and evaluation protocols: the diffusion equation, the Burgers equation, the reaction-diffusion system, Darcy flow, and the two-dimensional Navier-Stokes equation. WLNO consistently outperforms LNO on all five problems, with the most pronounced improvement on problems with strong spatial multi-scale structure, such as the Burgers equation with sharp shock fronts and the Navier-Stokes equation with coherent vortical structures, while remaining consistent across smoother and elliptic problems. These results demonstrate that wavelet-based multi-scale spatial decomposition is a principled and effective complement to Laplace-domain operator learning.

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 / 2 minor

Summary. The manuscript introduces the Wavelet-Laplace Neural Operator (WLNO), which augments the Laplace Neural Operator (LNO) with a parallel single-level Haar DWT branch. The branch decomposes lifted features into LL/LH/HL/HH subbands, applies independent learned 1x1 convolutions per subband, reconstructs via inverse DWT, and fuses the result with the LNO core through a learnable sigmoid-gated scalar alpha_wav (initialized for small initial wavelet contribution). WLNO is evaluated on five PDE benchmarks (diffusion, Burgers, reaction-diffusion, Darcy flow, 2D Navier-Stokes) using identical hyperparameters, data, and protocols to LNO, with the claim of consistent outperformance that is most pronounced on problems with strong spatial multi-scale structure.

Significance. If the reported gains are robustly attributable to the wavelet branch, the work would show that explicit multi-scale spatial decomposition can usefully complement Laplace-domain pole-residue modeling for neural operators. The adaptive gating mechanism and per-subband convolutions constitute a clean architectural extension that could be adopted in other frequency-domain operator frameworks.

major comments (2)
  1. [Experiments section] Experiments section (and abstract claim of outperformance): the manuscript states that identical hyperparameters, training data, and evaluation protocols are used, yet supplies neither ablation results (e.g., alpha_wav fixed at 0 or wavelet branch removed) nor multi-seed statistics, error bars, or variance estimates. Without these controls the observed deltas cannot be confidently attributed to the wavelet-Laplace fusion rather than optimizer stochasticity, initialization, or minor implementation differences.
  2. [Method section] Method section (description of fusion): the claim that the model 'adaptively balances' the two branches throughout training rests on the learned alpha_wav, but no analysis is provided of its converged values, sensitivity to initialization, or correlation with performance gains on the multi-scale problems (Burgers, Navier-Stokes). This leaves the mechanistic explanation for the reported improvements under-specified.
minor comments (2)
  1. [Abstract] Abstract: asserts 'consistent outperformance' and 'most pronounced improvement' without any numerical values, table references, or error metrics, which reduces the abstract's standalone informativeness.
  2. [Method section] Notation: the per-subband 1x1 convolution weights are described as 'independent learned' but their exact tensor shapes and how they interact with the lifted feature dimension are not stated explicitly, complicating re-implementation.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments on our manuscript. The points raised regarding experimental controls and analysis of the gating mechanism are valid and will be addressed through revisions to strengthen attribution of results and mechanistic understanding.

read point-by-point responses

  1. Referee: [Experiments section] Experiments section (and abstract claim of outperformance): the manuscript states that identical hyperparameters, training data, and evaluation protocols are used, yet supplies neither ablation results (e.g., alpha_wav fixed at 0 or wavelet branch removed) nor multi-seed statistics, error bars, or variance estimates. Without these controls the observed deltas cannot be confidently attributed to the wavelet-Laplace fusion rather than optimizer stochasticity, initialization, or minor implementation differences.

    Authors: We agree that the absence of ablations and statistical reporting limits confident attribution of the gains. In the revised manuscript we will add ablation experiments with the wavelet branch disabled (alpha_wav fixed at zero) and report results over multiple random seeds including means and standard deviations. revision: yes

  2. Referee: [Method section] Method section (description of fusion): the claim that the model 'adaptively balances' the two branches throughout training rests on the learned alpha_wav, but no analysis is provided of its converged values, sensitivity to initialization, or correlation with performance gains on the multi-scale problems (Burgers, Navier-Stokes). This leaves the mechanistic explanation for the reported improvements under-specified.

    Authors: We acknowledge that further analysis of alpha_wav is needed to support the adaptive balancing claim. The revision will report converged alpha_wav values per benchmark, include sensitivity experiments to different initializations, and examine correlations between alpha_wav and performance gains on multi-scale problems such as Burgers and Navier-Stokes. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical architecture comparison with no derivation chain

full rationale

The paper proposes WLNO by augmenting LNO with a parallel Haar DWT branch and a learnable sigmoid-gated fusion weight alpha_wav. All load-bearing claims are empirical head-to-head results on five PDE benchmarks under stated identical hyperparameters and protocols. No first-principles derivation, uniqueness theorem, or prediction is offered that reduces by construction to fitted inputs or self-citations. The learnable gate and 1x1 convolutions are standard trainable components whose values are determined by optimization, not by definitional equivalence to the reported performance deltas. Self-citation to LNO is to prior independent work and does not substitute for the current empirical evidence.

Axiom & Free-Parameter Ledger

2 free parameters · 1 axioms · 0 invented entities

The model introduces a small number of learned parameters for the wavelet branch and gate on top of the base LNO; it assumes the Haar DWT decomposition is a useful inductive bias for PDE feature maps without providing independent justification beyond the empirical results.

free parameters (2)
  • alpha_wav
    Learnable sigmoid-gated scalar that balances the wavelet branch contribution, initialized to favor the LNO branch.
  • per-subband 1x1 convolution weights
    Four independent learned 1x1 convolutions applied to the LL, LH, HL, and HH wavelet subbands.
axioms (1)
  • domain assumption Single-level Haar DWT decomposition into approximation and detail subbands extracts spatially localized multi-scale features relevant to PDE solutions.
    Invoked when the parallel wavelet branch is added to capture features the LNO core lacks.

pith-pipeline@v0.9.1-grok · 5839 in / 1319 out tokens · 39064 ms · 2026-06-30T14:54:31.613508+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

68 extracted references · 17 canonical work pages · 6 internal anchors

  1. [1]

    L. C. Evans, Partial Differential Equations, 2nd Edition, American Mathematical Society, 2010

    2010

  2. [2]

    R. J. LeVeque, Finite V olume Methods for Hyperbolic Problems, Cambridge University Press, 2002

    2002

  3. [3]

    Quarteroni, R

    A. Quarteroni, R. Sacco, F. Saleri, Numerical Mathematics, 2nd Edition, Springer, 2007

    2007

  4. [4]

    J. S. Hesthaven, G. Rozza, B. Stamm, Certified Reduced Basis Methods for Parametrized Partial Differential Equations, Springer, 2016

    2016

  5. [5]

    Rozza, D

    G. Rozza, D. B. P. Huynh, A. T. Patera, Reduced Basis Approximation and A Posteriori Error Estimation for Affinely Parametrized Elliptic Coercive Partial Differential Equations, Archives of Computational Methods in Engineering 15 (3) (2008) 229–275

    2008

  6. [6]

    Benner, S

    P. Benner, S. Gugercin, K. Willcox, A Survey of Projection-Based Model Reduction Methods for Parametric Dynamical Systems, SIAM Review 57 (4) (2015) 483–531

    2015

  7. [7]

    J. S. Hesthaven, S. Ubbiali, Non-Intrusive Reduced Order Modeling of Nonlinear Problems Using Neural Networks, Journal of Computational Physics 363 (2018) 55–78

    2018

  8. [8]

    Hasani, R

    E. Hasani, R. A. Ward, Generating synthetic data for neural operators, arXiv preprint arXiv:2401.02398 (2024)

    work page arXiv 2024

  9. [9]

    Bischof, M

    R. Bischof, M. Piovarci, M. Kraus, S. Mishra, B. Bickel, Hypino: Multi-physics neural operators via hyperpinns and the method of manufactured solutions, Advances in Neural Information Processing Systems 38 (2026) 144798–144831

    2026

  10. [10]

    Sojitra, O

    A. Sojitra, O. San, Method of Manufactured Learning for Solver-free Training of Neural Operators, arXiv preprint arXiv:2511.12890 (2025)

    work page arXiv 2025

  11. [11]

    LeCun, Y

    Y . LeCun, Y . Bengio, G. Hinton, Deep learning, Nature 521 (7553) (2015) 436–444

    2015

  12. [12]

    Goodfellow, Y

    I. Goodfellow, Y . Bengio, A. Courville, Deep Learning, MIT Press, 2016

    2016

  13. [13]

    Dissanayake, N

    M. Dissanayake, N. Phan-Thien, Neural-Network-Based Approximations for Solving Partial Differential Equa- tions, Communications in Numerical Methods in Engineering 10 (3) (1994) 195–201

    1994

  14. [14]

    I. E. Lagaris, A. Likas, D. I. Fotiadis, Artificial Neural Networks for Solving Ordinary and Partial Differential Equations, IEEE Transactions on Neural Networks 9 (5) (1998) 987–1000

    1998

  15. [15]

    Kovachki, Z

    N. Kovachki, Z. Li, B. Liu, K. Azizzadenesheli, K. Bhattacharya, A. Stuart, A. Anandkumar, Neural Operator: Learning Maps Between Function Spaces with Applications to PDEs, Journal of Machine Learning Research 24 (89) (2023) 1–97

    2023

  16. [16]

    G. E. Karniadakis, I. G. Kevrekidis, L. Lu, P. Perdikaris, S. Wang, L. Yang, Physics-Informed Machine Learning, Nature Reviews Physics 3 (6) (2021) 422–440

    2021

  17. [17]

    Cuomo, V

    S. Cuomo, V . S. Di Cola, F. Giampaolo, G. Rozza, M. Raissi, F. Piccialli, Scientific Machine Learning through Physics-Informed Neural Networks: Where We Are and What’s Next, Journal of Scientific Computing 92 (3) (2022) 88

    2022

  18. [18]

    G. C. Peng, M. Alber, A. Buganza Tepole, W. R. Cannon, S. De, S. Dura-Bernal, K. Garikipati, G. Karniadakis, W. W. Lytton, P. Perdikaris, et al., Multiscale Modeling Meets Machine Learning: What Can We Learn?, Archives of Computational Methods in Engineering 28 (2021) 1017–1037

    2021

  19. [19]

    L. Lu, P. Jin, G. Pang, Z. Zhang, G. E. Karniadakis, Learning Nonlinear Operators via DeepONet Based on the Universal Approximation Theorem of Operators, Nature Machine Intelligence 3 (3) (2021) 218–229

    2021

  20. [20]

    Z. Li, N. Kovachki, K. Azizzadenesheli, B. Liu, K. Bhattacharya, A. Stuart, A. Anandkumar, Fourier Neural Operator for Parametric Partial Differential Equations, arXiv preprint arXiv:2010.08895 (2020)

    work page internal anchor Pith review Pith/arXiv arXiv 2010

  21. [21]

    Z. Li, N. Kovachki, K. Azizzadenesheli, B. Liu, K. Bhattacharya, A. Stuart, A. Anandkumar, Multipole Graph Neural Operator for Parametric Partial Differential Equations, Advances in Neural Information Processing Systems 33 (2020) 6755–6766

    2020

  22. [22]

    G. Wen, Z. Li, K. Azizzadenesheli, A. Anandkumar, S. M. Benson, U-FNO-An Enhanced Fourier Neural Operator-Based Deep-Learning Model for Multiphase Flow, Advances in Water Resources 163 (2022) 104180

    2022

  23. [23]

    Z. Li, D. Z. Huang, B. Liu, A. Anandkumar, Fourier Neural Operator with Learned Deformations for PDEs on General Geometries, Journal of Machine Learning Research 24 (388) (2023) 1–26

    2023

  24. [24]

    A. Tran, A. Mathews, L. Xie, C. S. Ong, Factorized Fourier Neural Operators, arXiv preprint arXiv:2111.13802 (2021). 19 APREPRINT

    work page arXiv 2021

  25. [25]

    FourCastNet: A Global Data-driven High-resolution Weather Model using Adaptive Fourier Neural Operators

    J. Pathak, S. Subramanian, P. Harrington, S. Raja, A. Chattopadhyay, M. Mardani, T. Kurth, D. Hall, Z. Li, K. Azizzadenesheli, et al., FourCastNet: A Global Data-Driven High-Resolution Weather Model using Adaptive Fourier Neural Operators, arXiv preprint arXiv:2202.11214 (2022)

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  26. [26]

    Y . Yang, A. F. Gao, J. C. Castellanos, Z. E. Ross, K. Azizzadenesheli, K. Bhattacharya, Seismic Wave Propagation and Inversion with Neural Operators, The Seismic Record 1 (3) (2021) 126–135

    2021

  27. [27]

    Q. Cao, S. Goswami, G. E. Karniadakis, LNO: Laplace Neural Operator for Solving Differential Equations, arXiv preprint arXiv:2303.10528 (2023)

    work page arXiv 2023

  28. [28]

    Daubechies, Ten Lectures on Wavelets, SIAM, 1992

    I. Daubechies, Ten Lectures on Wavelets, SIAM, 1992

    1992

  29. [29]

    Mallat, A Wavelet Tour of Signal Processing, 2nd Edition, Academic Press, 1999

    S. Mallat, A Wavelet Tour of Signal Processing, 2nd Edition, Academic Press, 1999

    1999

  30. [30]

    Tripura, S

    T. Tripura, S. Chakraborty, Wavelet Neural Operator for Solving Parametric Partial Differential Equations in Computational Mechanics Problems, Computer Methods in Applied Mechanics and Engineering 404 (2023) 115783

    2023

  31. [31]

    Bhattacharya, B

    K. Bhattacharya, B. Hosseini, N. B. Kovachki, A. M. Stuart, Model Reduction and Neural Networks for Parametric PDEs, SMAI Journal of Computational Mathematics 7 (2021) 121–157

    2021

  32. [32]

    T. Chen, H. Chen, Universal Approximation to Nonlinear Operators by Neural Networks with Arbitrary Activation Functions and its Application to Dynamical Systems, IEEE Transactions on Neural Networks 6 (4) (1995) 911–917

    1995

  33. [33]

    S. Wang, H. Wang, P. Perdikaris, Improved Architectures and Training Algorithms for Deep Operator Networks with Applications to Geomechanics, Engineering with Computers 38 (6) (2022) 4651–4678

    2022

  34. [34]

    de Wolff, G

    T. de Wolff, G. Gola, D. Morrison, C. Kuthe, S. Chen, V . Oommen, J. Hao, G. E. Karniadakis, Multifidelity Deep Operator Networks, arXiv preprint arXiv:2204.09234 (2022)

    work page arXiv 2022

  35. [35]

    FEDONet : Fourier-Embedded DeepONet for Spectrally Accurate Operator Learning

    A. Sojitra, M. Dhingra, O. San, FEDONet: Fourier-embedded DeepONet for spectrally accurate operator learning, arXiv preprint arXiv:2509.12344 (2025)

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  36. [36]

    M. Abid, O. San, Spectral Embedding via Chebyshev Bases for Robust DeepONet Approximation, arXiv preprint arXiv:2512.09165 (2025)

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  37. [37]

    C. Lin, M. Maxey, Z. Li, G. E. Karniadakis, Operator Learning for Predicting Multiscale Bubble Growth Dynamics, The Journal of Chemical Physics 154 (10) (2021)

    2021

  38. [38]

    Z. Li, N. Kovachki, C. Choy, B. Li, J. Kossaifi, S. P. Otta, M. A. Nabian, M. Stadler, C. Hundt, K. Azizzadenesheli, A. Anandkumar, Geometry-Informed Neural Operator for Large-Scale 3D PDEs, arXiv preprint arXiv:2309.00583 (2023)

    work page arXiv 2023

  39. [39]

    Bonev, T

    B. Bonev, T. Kurth, C. Hundt, J. Pathak, M. Baust, K. Kashinath, A. Anandkumar, Spherical Fourier Neural Operators: Learning Stable Dynamics on the Sphere, arXiv preprint arXiv:2306.03838 (2023)

    work page arXiv 2023

  40. [40]

    M. Abid, O. San, SIMR-NO: A Spectrally-Informed Multi-Resolution Neural Operator for Turbulent Flow Super-Resolution, arXiv preprint arXiv:2603.28073 (2026)

    work page arXiv 2026

  41. [41]

    Vaswani, N

    A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, L. Kaiser, I. Polosukhin, Attention Is All You Need, in: Advances in Neural Information Processing Systems, V ol. 30, 2017

    2017

  42. [42]

    Ronneberger, P

    O. Ronneberger, P. Fischer, T. Brox, U-Net: Convolutional Networks for Biomedical Image Segmentation, in: Medical Image Computing and Computer-Assisted Intervention, 2015, pp. 234–241

    2015

  43. [43]

    K. He, X. Zhang, S. Ren, J. Sun, Deep Residual Learning for Image Recognition, in: IEEE Conference on Computer Vision and Pattern Recognition, 2016, pp. 770–778

    2016

  44. [44]

    Kissas, J

    G. Kissas, J. H. Seidman, L. F. Guilhoto, V . M. Celli, L. Pigneur, P. Perdikaris, Learning Operators with Coupled Attention, Journal of Machine Learning Research 23 (215) (2022) 1–63

    2022

  45. [45]

    Raonic, R

    B. Raonic, R. Molinaro, T. De Ryck, T. Rohner, F. Bartolucci, R. Alaifari, S. Mishra, E. de Bézenac, Convolutional Neural Operators for Robust and Accurate Learning of PDEs, Advances in Neural Information Processing Systems 36 (2024)

    2024

  46. [46]

    Poseidon: Efficient foundation models for PDEs

    M. Herde, B. Raonic, T. Rohner, R. Käppeli, R. Molinaro, E. de Bézenac, S. Mishra, Poseidon: Efficient Foundation Models for PDEs, arXiv preprint arXiv:2405.19101 (2024)

    work page arXiv 2024

  47. [47]

    Cao, Choose a transformer: Fourier or Galerkin, in: Advances in Neural Information Processing Systems, V ol

    S. Cao, Choose a transformer: Fourier or Galerkin, in: Advances in Neural Information Processing Systems, V ol. 34, 2021, pp. 24924–24940

    2021

  48. [48]

    Gaussian Error Linear Units (GELUs)

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

    work page internal anchor Pith review Pith/arXiv arXiv 2016

  49. [49]

    C. K. Chui, An Introduction to Wavelets, Academic Press, 1992. 20 APREPRINT

    1992

  50. [50]

    Mallat, Group Invariant Scattering, Communications on Pure and Applied Mathematics 65 (10) (2012) 1331– 1398

    S. Mallat, Group Invariant Scattering, Communications on Pure and Applied Mathematics 65 (10) (2012) 1331– 1398

    2012

  51. [51]

    Bruna, S

    J. Bruna, S. Mallat, Invariant Scattering Convolution Networks, IEEE Transactions on Pattern Analysis and Machine Intelligence 35 (8) (2013) 1872–1886

    2013

  52. [52]

    P. Liu, H. Zhang, W. Lian, W. Zuo, Multi-Level Wavelet Convolutional Neural Networks, IEEE Access 7 (2019) 74973–74985

    2019

  53. [53]

    Williams, R

    T. Williams, R. Li, Wavelet Pooling for Convolutional Neural Networks, in: International Conference on Learning Representations, 2018

    2018

  54. [54]

    Gupta, X

    G. Gupta, X. Xiao, P. Bogdan, Multiwavelet-based Operator Learning for Differential Equations, Advances in Neural Information Processing Systems 34 (2021) 24048–24062

    2021

  55. [55]

    Raissi, P

    M. Raissi, P. Perdikaris, G. E. Karniadakis, Physics-Informed Neural Networks: A Deep Learning Frame- work for Solving Forward and Inverse Problems Involving Nonlinear Partial Differential Equations, Journal of Computational Physics 378 (2019) 686–707

    2019

  56. [56]

    L. Lu, X. Meng, Z. Mao, G. E. Karniadakis, DeepXDE: A Deep Learning Library for Solving Differential Equations, SIAM Review 63 (1) (2021) 208–228

    2021

  57. [57]

    A. D. Jagtap, G. E. Karniadakis, Extended Physics-Informed Neural Networks (XPINNs): A Generalized Space- Time Domain Decomposition Based Deep Learning Framework for Nonlinear Partial Differential Equations, Communications in Computational Physics 28 (5) (2020) 2002–2041

    2020

  58. [58]

    Kharazmi, Z

    E. Kharazmi, Z. Zhang, G. E. Karniadakis, hp-VPINNs: Variational Physics-Informed Neural Networks with Domain Decomposition, Computer Methods in Applied Mechanics and Engineering 374 (2021) 113547

    2021

  59. [59]

    Goswami, M

    S. Goswami, M. Yin, Y . Yu, G. E. Karniadakis, A Physics-Informed Variational DeepONet for Nonlinear Fracture Mechanics, Computer Methods in Applied Mechanics and Engineering 391 (2022) 114587

    2022

  60. [60]

    Z. Li, H. Zheng, N. Kovachki, D. Jin, H. Chen, B. Liu, K. Azizzadenesheli, A. Anandkumar, Physics-Informed Neural Operator for Learning Partial Differential Equations, ACM/JMS Journal of Data Science 1 (3) (2024) 1–27

    2024

  61. [61]

    P. R. Vlachas, J. Pathak, B. R. Hunt, T. P. Sapsis, M. Girvan, E. Ott, P. Koumoutsakos, Forecasting of Spatio- Temporal Chaotic Dynamics with Recurrent Neural Networks: A Comparative Study of Reservoir Computing and Backpropagation Algorithms, arXiv preprint arXiv:1910.05266 (2019)

    work page arXiv 1910

  62. [62]

    Kreyszig, Advanced Engineering Mathematics, 10th Edition, John Wiley & Sons, 2010

    E. Kreyszig, Advanced Engineering Mathematics, 10th Edition, John Wiley & Sons, 2010

    2010

  63. [63]

    S.-L. J. Hu, F. Liu, B. Gao, H. Li, Pole-Residue Method for Numerical Dynamic Analysis, Journal of Engineering Mechanics 142 (8) (2016) 04016045

    2016

  64. [64]

    Cao, S.-L

    Q. Cao, S.-L. J. Hu, H. Li, Laplace- and Frequency-Domain Methods on Computing Transient Responses of Oscillators with Hysteretic Dampings to Deterministic Loading, Journal of Engineering Mechanics 149 (3) (2023) 04023005

    2023

  65. [65]

    Instance Normalization: The Missing Ingredient for Fast Stylization

    D. Ulyanov, A. Vedaldi, V . Lempitsky, Instance Normalization: The Missing Ingredient for Fast Stylization, arXiv preprint arXiv:1607.08022 (2016)

    work page internal anchor Pith review Pith/arXiv arXiv 2016

  66. [66]

    D. P. Kingma, J. Ba, Adam: A Method for Stochastic Optimization, in: International Conference on Learning Representations, 2015

    2015

  67. [67]

    Paszke, S

    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, in: Advances in Neural Information Processing Systems, V ol. 32, 2019

    2019

  68. [68]

    M. V . de Hoop, D. Z. Huang, E. Qian, A. M. Stuart, The cost-accuracy trade-off in operator learning with neural networks, arXiv preprint arXiv:2203.13181 (2022). 21

    work page arXiv 2022