Pith. sign in

REVIEW 6 major objections 6 minor 43 references

This paper claims that a trainable, sigmoid-bounded weight between a Fourier-KAN branch and an MLP branch, combined with overlapping domain decomposition, lets a physics-informed neural network solve high-frequency and multiscale PDEs more

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · deepseek-v4-flash

2026-08-03 22:14 UTC pith:ACKYLOZR

load-bearing objection Trainable sigmoid-bounded MLP/KAN weight is a plausible idea, but the paper never compares it to fixed-α tuning, and its cost claims contradict its own conclusion. the 6 major comments →

arxiv 2511.11228 v3 pith:ACKYLOZR submitted 2025-11-14 math.NA cs.NA

The modified Physics-Informed Hybrid Parallel Kolmogorov--Arnold and Multilayer Perceptron Architecture with domain decomposition

Qiumei Huang , Xu Wang , Yu Zhao This is my paper
classification math.NA cs.NA MSC 65N9968T07
keywords Physics-informed neural networksKolmogorov-Arnold networksHigh-frequency multiscale problemsDomain decompositionTrainable weighting parameterSpectral biasFourier KANHybrid architecture
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper proposes a modified hybrid network for physics-informed PDE solving in which a Fourier-based Kolmogorov–Arnold branch and a multilayer perceptron branch are combined by a trainable weight, bounded to [0,1] by an S-shaped function, and the computational domain is split into overlapping subdomains. The central claim is that this automatically balances the KAN's high-frequency resolution with the MLP's low-frequency global structure, removing manual mixing-weight tuning and reducing training cost. The paper reports lower L2 errors than MLP-only or KAN-only PINNs across Helmholtz, two- and five-dimensional Poisson, reaction-diffusion, and Allen–Cahn benchmarks, while using 16 rather than 32 neurons per layer. A sympathetic reader would care because standard PINNs are known to underresolve high-frequency and multiscale solutions, and the paper offers a parameter-light architectural fix.

Core claim

On the paper's own terms, the discovery is that replacing the fixed hyperparameter alpha in the hybrid parallel KAN–MLP architecture with a trainable parameter constrained by S(alpha) in [0,1] yields a self-balancing network: as training proceeds, S(alpha) moves toward the MLP for low-frequency problems and toward the KAN for high-frequency problems. Combined with overlapping domain decomposition and hard-constrained boundary and initial conditions, this modified HPKM-PINN reproduces oscillatory solutions such as sin(32πx)sin(32πy), and in Table 1 it reports the best L2 error among the three architectures on every benchmark considered — for example, Helmholtz with ω=48 at 8.698e-4 versus 1.1

What carries the argument

The central object is the modified HPKM output u = S(α) u_KAN + (1 − S(α)) u_MLP, where u_KAN is a Fourier-KAN (a Kolmogorov–Arnold network whose univariate activations are truncated Fourier series) and u_MLP is a standard multilayer perceptron. S is an S-shaped map (sigmoid in the final experiments) that confines α to [0,1]; in the domain-decomposed version each subdomain has its own α^(i), and a partition-of-unity window function ω_i blends the local modified-HPKM outputs into a global solution. The mechanism doing the work is the adaptive convex gate: it is trained jointly with the network weights by gradient descent, so the model can tilt toward whichever branch fits the local frequency

Load-bearing premise

The load-bearing premise is that one scalar weight, squeezed into [0,1], can keep the KAN and MLP branches cooperating so the KAN supplies high frequencies and the MLP supplies low frequencies; if the two branches interfere instead of complementing, the reported advantages could depend on the specific benchmark settings.

What would settle it

On the high-frequency Helmholtz case (ω=32), record the trained S(α) over many seeds: the paper's mechanism predicts it drifts toward 1, favoring KAN, and that this drift is what produces the accuracy gain. A concrete falsifier is to freeze S(α) at 0.5 and rerun the same benchmarks: if accuracy does not degrade materially, the trainable gate is not the active ingredient. A second check is to replace the Fourier basis in the KAN branch with a static nonlinearity while retraining; if the final L2 errors are statistically unchanged, then the high-frequency claim is not being driven by the Fourier

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • The mixing ratio between KAN and MLP no longer needs manual grid search; training sets it, and the S-shaped bound keeps it a valid convex combination.
  • On the reported benchmarks, the hybrid delivers smaller L2 error than either single-branch architecture under matched domain decomposition, so it can be read as a more accurate default for oscillatory PDEs.
  • Because the hybrid matches or beats 32-neuron baselines with 16 neurons per branch, the parameter count and memory footprint are smaller than a pure KAN of comparable width.
  • Input normalization inside each overlapping subdomain converts global high-frequency features into local lower-frequency ones, which the paper identifies as the reason spectral bias is mitigated.
  • The evolution of S(α) gives a diagnostic: the learned gate tilts to MLP on low-frequency problems and to KAN on high-frequency problems, supporting the frequency-separation narrative.

Where Pith is reading between the lines

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

  • Editorial inference: If the learned S(α) is the mechanism, then the same gate idea could be applied at the level of loss terms or frequency bands rather than only subdomains; the paper provides no test of whether per-subdomain weights are essential or merely convenient.
  • Editorial inference: The paper's benchmark list is separable and low-dimensional; whether the 16-neuron hybrid retains its advantage on non-separable, three-dimensional, or time-critical problems is an untested extrapolation.
  • Editorial inference: The reaction-diffusion result, where MLP and KAN both fail at about 0.92 while the hybrid reaches 1e-4, suggests the two branches' errors are not just averaged but cancel or complement; a controlled study varying branch widths could show whether the benefit is from the gate or from having two different inductive biases at all.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

6 major / 6 minor

Summary. The paper proposes a modified Hybrid Parallel Kolmogorov–Arnold Network and Multilayer Perceptron PINN (modified HPKM-PINN) with overlapping domain decomposition. The key modification is replacing the fixed weighting parameter α in the HPKM combination with a trainable parameter transformed by an S-shaped function, so that the convex combination of KAN and MLP outputs is adaptively balanced. The method is tested on Helmholtz (ω=16,32,48), 2D and 5D Poisson, reaction–diffusion, and Allen–Cahn benchmarks, reporting lower L2 errors than MLP-only and KAN-only PINNs. The abstract claims that the method reduces training costs and improves computational efficiency compared with manual hyperparameter tuning.

Significance. If substantiated, the contribution is a simple and inexpensive architectural modification with a plausible intuitive appeal: combining a KAN branch and an MLP branch with a bounded trainable weight within an overlapping domain decomposition. The benchmark suite is reasonably broad, including high-frequency and multiscale problems, and the method is easy to implement. However, the central efficiency claim is unsupported and is in direct tension with the paper's own conclusion that the model requires higher training costs than MLP architectures. The experiments do not isolate the effect of the adaptive weighting mechanism, report no statistical uncertainty, and provide no code. The frequency-separation narrative is also not established. The paper therefore presents a promising empirical idea that requires substantially stronger validation, rather than a demonstrated method.

major comments (6)
  1. [§3.1.1, Eq. (2.2)] The central novelty is the trainable weighting parameter S(α). The experiments in Figs. 4–7 compare only different S-shaped functions and the unweighted 'No α' direct sum; they do not include any fixed-α baseline (e.g., α=0, 0.25, 0.5, 0.75, 1) or the original HPKM with manually tuned α. Since the abstract claims reduced cost compared with manual hyperparameter tuning, this missing control is load-bearing. The reported gains could come from the hybrid architecture itself or the domain decomposition rather than from the adaptive mechanism. Add fixed-α baselines and report the cost of tuning α.
  2. [§4 vs. Abstract] The conclusion states that the model 'requires higher training costs than traditional MLP architectures due to its larger parameter size,' while the abstract claims the method 'reduces training costs and improves computational efficiency compared with manual hyperparameter tuning.' No runtime, FLOPs, or parameter counts are reported anywhere. The efficiency claim is therefore unsupported and internally inconsistent. The authors should either substantiate the claim with measurements or remove it from the abstract and introduction.
  3. [§3.1, §3.2.1, §3.3, §3.4] Overlap ratios, subdomain counts, sampling points, Fourier frequency count K, hard-constraint steepness σ, and the choice of S-shaped function are selected per benchmark with statements such as 'Based on our tests' and 'determined through testing.' The sigmoid function is selected after comparing candidates on the same Helmholtz problems that are later used to claim superiority. This is selection on the test problems and risks circularity. The authors should provide sensitivity analyses and make it clear that all baselines were tuned with the same effort, or conduct validation on separate problems.
  4. [§2.2.2, Algorithm 1] The hard-constraint operator C is not defined in general. An explicit tanh-based construction is given only for the Helmholtz equation; for the Poisson, reaction–diffusion, and Allen–Cahn examples no hard-constraint construction is provided, even though all experiments rely on it. This omission prevents reproduction. Provide the concrete form of C for each benchmark or switch to soft constraints with reported loss weights.
  5. [§2.1.2, Figs. 5 and 7] The paper claims that a single scalar S(α) automatically balances low-frequency (MLP) and high-frequency (KAN) components. However, both branches receive identical inputs and S(α) is a global scalar per subdomain, not a frequency-dependent filter. The observed improvements may instead reflect an ensemble effect or the higher capacity of having two branches. The S(α) trajectory plots do not demonstrate frequency separation. Please provide a concrete analysis of when the two branches interfere or soften the interpretation.
  6. [Table 1] All L2 errors in Table 1 are single-run numbers with no seeds, no error bars, and no code. Neural network training is stochastic, so differences such as 2.362×10⁻⁴ vs. 5.115×10⁻⁴ may lie within run-to-run variability. The manuscript should report means and standard deviations over multiple seeds and, ideally, release code to make the comparisons reproducible.
minor comments (6)
  1. [§3.1 title] Typo: 'Hemholtz' should be 'Helmholtz'.
  2. [§3.1.2] Typo: 'KAN-baed model' should be 'KAN-based model'.
  3. [Fig. 9 caption] Typo: 'Possion' should be 'Poisson'.
  4. [References [16]–[17]] The KAN citation is inconsistent: the introduction refers to 'Li et al. [17]' but reference [17] is Kodama (2011); the KAN paper appears as reference [16] with different authors. Reference [6] is cited for a lack of theoretical analysis but actually points to a Fourier features paper.
  5. [Eq. (3.7)] The definition of f(x) has unbalanced parentheses, which should be corrected.
  6. [Algorithm 1] Line 14 says 'Hard constraints: û^(i)=C u^(i)' but the operator C is not specified there; please clarify that it refers to the construction in §2.2.2 and give the explicit form.

Circularity Check

0 steps flagged

No circularity found: architecture is defined by Eq. (2.2) and compared on benchmarks; noted weaknesses are experimental-support issues, not derivation-by-construction circularity.

full rationale

The paper's claimed contribution is a trainable weighting S(α) inserted in Eq. (2.2); S(α) is a trainable parameter, not fitted to the reported benchmark outputs. No equation in Section 2 reduces to another by construction: Eq. (2.3) is the definition of the network output, and Eq. (2.4) is the standard PINN loss. The HPKM architecture is attributed to external prior work [37], Fourier KAN to [36], and FBPINN domain decomposition to [30,32,38]; none of these are self-citations of the present authors, and none is used to forbid alternatives. The main weaknesses—the sigmoid function and overlap ratios selected on the same Helmholtz/Poisson benchmarks later reported, and the conclusion conceding higher training cost than MLP ('Although the model requires higher training costs than traditional MLP architectures...')—undercut the abstract's cost-saving claim, but they are experimental-support and consistency issues, not circular reductions. Therefore no significant circularity is found.

Axiom & Free-Parameter Ledger

8 free parameters · 6 axioms · 0 invented entities

No invented physical entities are introduced. The architecture borrows KAN, MLP, and FBPINN machinery; the central free choices are per-problem hyperparameters (δ, subdomain count, sampling density, S-function choice) that the authors tune on the same benchmarks they report.

free parameters (8)
  • Overlap ratio δ = 3.3 (Helmholtz); 2.9 (2D Poisson, reaction-diffusion, Allen-Cahn); 1.9 (5D Poisson)
    Manually selected per benchmark 'based on our tests'; controls subdomain overlap and affects accuracy and efficiency.
  • Number of subdomains N = 20×20 for Helmholtz and Allen-Cahn; 2^(m-1)×2^(m-1) for 2D Poisson; 2^5 for 5D Poisson; 10 for reaction-diffusion
    Chosen by the authors according to problem complexity; scaling subdomains is part of the method and strongly affects accuracy.
  • Sampling point counts = 260×260 (Helmholtz, Allen-Cahn); (10·2^(m-1))² (2D Poisson); 10^5 (5D Poisson); 200 (reaction-diffusion)
    Set per problem to match complexity; not justified by convergence tests.
  • S-shaped weighting function S(α) = sigmoid
    Selected after comparing six S-shaped functions on the Helmholtz benchmarks; sigmoid gave the best final L2 error on those same problems.
  • Fourier KAN frequency count K = 4
    Fixed for all experiments; not swept or justified.
  • Hard-constraint steepness σ = 0.2 (Helmholtz)
    Controls tanh steepness in the boundary-constrained ansatz; chosen by hand.
  • Network width = 16 neurons per layer for modified HPKM; 32 for MLP and KAN baselines
    Architecture choice; the unequal widths make the comparison to baselines uneven.
  • Learning rate = 1e-4 (Adam)
    Fixed across all runs; standard choice, not tuned.
axioms (6)
  • standard math Kolmogorov-Arnold representation theorem, and the KAN approximation built on it
    Invoked in §2.1.1 to justify KAN as a compositional approximator for multivariate functions.
  • domain assumption Minimizing the PINN residual with hard-constrained boundary/initial terms yields the PDE solution
    The loss in §2.2.2 combines residual and data terms; no convergence or consistency theorem is proved.
  • domain assumption Partition-of-unity window functions produce a valid continuous global solution from local subnetworks
    Window functions in §2.2.1 satisfy Σω=1; continuity and PDE consistency are assumed following FBPINN references.
  • domain assumption Input normalization inside each subdomain lowers effective frequency and mitigates spectral bias
    Stated in §2.2.1 without proof; this is a load-bearing premise for why domain decomposition helps high-frequency problems.
  • domain assumption Fourier-KAN basis functions approximate high-frequency components better than B-spline KAN or MLP
    The choice of Fourier KAN in §2.1.1 is motivated by a reference, not by a comparative mathematical result.
  • ad hoc to paper A bounded convex combination S(α)∈[0,1] improves optimization stability
    The paper asserts in §2.1.2 that bounding α with an S-shaped function stabilizes training, but offers no analysis of how the bound changes the loss landscape.

pith-pipeline@v1.3.0-alltime-deepseek · 15886 in / 13882 out tokens · 113909 ms · 2026-08-03T22:14:03.020855+00:00 · methodology

0 comments
read the original abstract

In this work, we propose a modified Hybrid Parallel Kolmogorov--Arnold Network and Multilayer Perceptron Physics-Informed Neural Network to overcome the high-frequency and multiscale challenges inherent in Physics-Informed Neural Networks. This proposed model features a trainable weighting parameter to optimize the convex combination of outputs from the Kolmogorov--Arnold Network and the Multilayer Perceptron, thus maximizing the networks' capabilities to capture different frequency components. Furthermore, we adopt an overlapping domain decomposition technique to decompose complex problems into subproblems, which alleviates the challenge of global optimization. Benchmark results demonstrate that our method reduces training costs and improves computational efficiency compared with manual hyperparameter tuning in solving high-frequency multiscale problems.

Figures

Figures reproduced from arXiv: 2511.11228 by Qiumei Huang, Xu Wang, Yu Zhao.

Figure 1
Figure 1. Figure 1: Modified HPKM architecture. 2.2 Overlapping domain decomposition technique in modified HPKM-PINN In this subsection, we apply the modified HPKM architecture to the PINN framework, introducing a modified HPKM-PINN that incorporates the overlapping domain decomposition technique. 2.2.1 Overlapping domain decomposition The overlapping domain decomposition technique has advantages in handling multiscale and hi… view at source ↗
Figure 2
Figure 2. Figure 2: Domain decomposition of the 1D domain Ω = [0, l], l = 1. Following the overlapping domain decomposition process, we train separate neural networks on each subdomain Ωi and construct a global solution by blending their outputs with window functions. In this work, the window function ωi is given by ωi(x) = ωˆi(x) PN j=1 ωˆj (x) , x = (x1, . . . , xd) T ∈ Ω, i = 1, . . . , N, where for N = 1, ωˆi(x) ≡ 1, x ∈ … view at source ↗
Figure 3
Figure 3. Figure 3: Graphical abstract of the modified HPKM-PINN with overlapping domain decomposition. [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Helmholtz equation (ω = 1): (a) Exact solution; (b) Predicted solution; (c) The convergence curve of L2 error with different S-shaped functions; (d) The final L2 error after completing the training with different S-shaped functions. The results are illustrated in [PITH_FULL_IMAGE:figures/full_fig_p012_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: 20 random evolution curves of the weights [PITH_FULL_IMAGE:figures/full_fig_p013_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Helmholtz equation (ω = 32): (a) Exact solution; (b) Predicted solution; (c) The convergence curve of L2 error with different S-shaped functions; (d) The final L2 error after completing the training with different S-shaped functions. 3.2 Poisson equation The Poisson equation describes the spatial distribution of potential fields generated by given source terms, effectively capturing the interplay between d… view at source ↗
Figure 7
Figure 7. Figure 7: 20 random evolution curves of the weights [PITH_FULL_IMAGE:figures/full_fig_p015_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Helmholtz equation with ω = 16, 32, 48. Top: Exact solutions. Middle: Predicted solutions obtained by the modified HPKM-PINN. Bottom: L2 error comparison between KAN, MLP and the modified HPKM architectures. 16 [PITH_FULL_IMAGE:figures/full_fig_p016_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Two-dimensional Possion equation with m = 2, 3, 4, 5, 6. Top: Exact solutions. Middle: Predicted solutions obtained by the modified HPKM-PINN. Bottom: L2 error comparison between KAN, MLP and modified HPKM architectures. f(x) in (3.4) with x = (x1, x2, x3, x4, x5) T is given by f(x) = −5 Y 5 i=1 sin(πxi), and the corresponding exact solution is given by u(x) = − 1 π 2 Y 5 i=1 sin(πxi). In solving the five-… view at source ↗
Figure 10
Figure 10. Figure 10: Five-dimensional Poisson equation: (a) A two-dimensional slice of the exact solution in the [PITH_FULL_IMAGE:figures/full_fig_p018_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Steady-State reaction-diffusion equation: (a) The exact and predicted solution obtained by the [PITH_FULL_IMAGE:figures/full_fig_p019_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Allen–Cahn equation: (a) Exact solution; (b) Predicted solution obtained by the modified HPKM [PITH_FULL_IMAGE:figures/full_fig_p020_12.png] view at source ↗

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

43 extracted references · 3 linked inside Pith

  1. [1]

    Raissi, M., Perdikaris, P., Karniadakis, G. E. (2019). Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations. Journal of Computational physics, 378, 686-707

    2019

  2. [2]

    Sharma, R., Shankar, V. (2022). Accelerated training of physics-informed neural networks (PINNs) using meshless discretizations. Advances in Neural Information Processing Systems, 35, 1034-1046

    2022

  3. [3]

    D., Karniadakis, G

    Mao, Z., Jagtap, A. D., Karniadakis, G. E. (2020). Physics-informed neural networks for high-speed flows. Computer Methods in Applied Mechanics and Engineering, 360, 112789

    2020

  4. [4]

    A., Ullmann, E., Peters, N

    Herrero Martin, C., Oved, A., Chowdhury, R. A., Ullmann, E., Peters, N. S., et al. (2022). EP- PINNs: Cardiac electrophysiology characterisation using physics-informed neural networks. Frontiers in Cardiovascular Medicine, 8, 768419

    2022

  5. [5]

    P., Sun, Z

    Hou, Q., Li, Y., Singh, V. P., Sun, Z. (2024). Physics-informed neural network for diffusive wave model. Journal of Hydrology, 637, 131261

    2024

  6. [6]

    Tancik, M., Srinivasan, P., Mildenhall, B., Fridovich-Keil, S., Raghavan, N., et al. (2020). Fourier features let networks learn high frequency functions in low dimensional domains. Advances in Neural Information Processing Systems, 33, 7537-7547

    2020

  7. [7]

    Lee, K., Trask, N.A., Patel, R.G., Gulian, M.A., Cyr, E.C. (2021). Partition of unity networks: Deep hp-approximation. AAAI 2021 spring symposium on combining artificial intelligence and machine learning with physical sciences, AAAI-MLPS 2021, vol. 2964, No. 180

    2021

  8. [8]

    X., Gu, Q

    Cao, Y., Fang, Z., Wu, Y., Zhou, D. X., Gu, Q. (2021). Towards Understanding the Spectral Bias of Deep Learning. The 30th International Joint Conference on Artificial Intelligenc, IJCAI 2021, 2205–2211

    2021

  9. [9]

    Jacot, A., Gabriel, F., Hongler, C. (2018). Neural tangent kernel: Convergence and generalization in neural networks. Advances in Neural Information Processing Systems, 31

    2018

  10. [10]

    Xu, Z. Q. J., Zhang, Y., Luo, T. (2024). Overview frequency principle/spectral bias in deep learning. Communications on Applied Mathematics and Computation, 1-38

    2024

  11. [11]

    Xu, Y., Yuan, S., Zeng, H., Yuan, J., Yu, Y., et al. (2025). Frequency-dependent multiscale network for seismic high-resolution processing. Geophysics, 90(4), 297-312

    2025

  12. [12]

    Gao, Z., Cui, X., Wan, W., Gu, Z. (2019). Recognition of emotional states using multiscale information analysis of high frequency EEG oscillations. Entropy, 21(6), 609

    2019

  13. [13]

    W., Abhayapala, T

    Hanlen, L. W., Abhayapala, T. D. (2007). Space-time-frequency degrees of freedom: Fundamental limits for spatial information. In 2007 IEEE International Symposium on Information Theory, 701-705. IEEE

    2007

  14. [14]

    Krizhevsky, A., Sutskever, I., Hinton, G. E. (2012). Imagenet classification with deep convolutional neural networks. Advances in Neural Information Processing Systems, 25

    2012

  15. [15]

    C., Hagenbuchner, M., Monfardini, G

    Scarselli, F., Gori, M., Tsoi, A. C., Hagenbuchner, M., Monfardini, G. (2008). The graph neural network model. IEEE transactions on neural networks, 20(1), 61-80

    2008

  16. [16]

    Liu, Z., Wang, Y., Vaidya, S., Ruehle, F., Halverson, J., et al. (2025). KAN: Kolmogorov–arnold networks. 13th International Conference on Learning Representations, ICLR 2025, 66342–66388. 22

    2025

  17. [17]

    KODAMA, S. (2011). A version of Kolmogorov–Arnold representation theorem for differentiable functions of several variables. Journal of Nonlinear Analysis and Optimization: Theory and Applications, 2(2), 253-257

    2011

  18. [18]

    Wang, J., Zhao, X., Feng, X., Xu, H. (2024). An improved physics-informed neural network with adaptive weighting and mixed differentiation for solving the incompressible Navier-Stokes equations. Nonlinear Dynamics, 112(18), 16113-16134

    2024

  19. [19]

    J., Toscano, J

    Anagnostopoulos, S. J., Toscano, J. D., Stergiopulos, N., Karniadakis, G. E. (2024). Residual-based attention in physics-informed neural networks. Computer Methods in Applied Mechanics and Engineering, 421, 116805

    2024

  20. [20]

    A., Stinis, P

    Chen, W., Howard, A. A., Stinis, P. (2025). Self-adaptive weights based on balanced residual decay rate for physics-informed neural networks and deep operator networks. Journal of Computational Physics, 542, 114226

    2025

  21. [21]

    A., Murphy, S

    Howard, A. A., Murphy, S. H., Ahmed, S. E., Stinis, P. (2025). Stacked networks improve physics- informed training: Applications to neural networks and deep operator networks. Foundations of Data Science, 7(1), 134–162

    2025

  22. [22]

    Guo, J., Wang, H., Gu, S., Hou, C. (2024). TCAS-PINN: Physics-informed neural networks with a novel temporal causality-based adaptive sampling method. Chinese Physics B, 33(5), 050701

    2024

  23. [23]

    Meng, X., Karniadakis, G. E. (2020). A composite neural network that learns from multi-fidelity data: Application to function approximation and inverse PDE problems. Journal of Computational Physics, 401, 109020

    2020

  24. [24]

    Suro, F., Ferber, J., Stratulat, T., Michel, F. (2021). A hierarchical representation of behaviour supporting open ended development and progressive learning for artificial agents. Autonomous Robots, 45, 245-264

    2021

  25. [25]

    E., Kevrekidis, I

    Karniadakis, G. E., Kevrekidis, I. G., Lu, L., Perdikaris, P., Wang, S., et al. (2021). Physics-informed machine learning. Nature Reviews Physics, 3(6), 422-440

    2021

  26. [26]

    Wesel, F., Batselier, K. (2021). Large-scale learning with fourier features and tensor decompositions. Advances in Neural Information Processing Systems, 34, 17543-17554

    2021

  27. [27]

    A., Heniche, M., Glowinski, R

    Rivera, C. A., Heniche, M., Glowinski, R. (2010). Parallel finite element simulations of incompressible viscous fluid flow by domain decomposition with Lagrange multipliers. Journal of Computational Physics, 229(13), 5123-5143

    2010

  28. [28]

    Guo, J., Yao, Y., Wang, H., Gu, T. (2023). Pre-training strategy for solving evolution equations based on physics-informed neural networks. Journal of Computational Physics, 489, 112258

    2023

  29. [29]

    D., Karniadakis, G

    Jagtap, A. D., Karniadakis, G. E. (2020). 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

  30. [30]

    Moseley, B., Markham, A., Nissen-Meyer, T. (2023). Finite basis physics-informed neural networks (FBPINNs): a scalable domain decomposition approach for solving differential equations. Advances in Computational Mathematics, 49(4), 62

    2023

  31. [31]

    A., Beecroft, D., Stinis, P

    Heinlein, A., Howard, A. A., Beecroft, D., Stinis, P. (2024). Multifidelity domain decomposition-based physics-informed neural networks for time-dependent problems. arXiv preprint arXiv:2401.07888. 23

    Pith/arXiv arXiv 2024

  32. [32]

    Dolean, V., Heinlein, A., Mishra, S., Moseley, B. (2024). Multilevel domain decomposition-based architectures for physics-informed neural networks. Computer Methods in Applied Mechanics and Engineering, 429, 117116

    2024

  33. [33]

    Hornik, K., Stinchcombe, M., White, H. (1989). Multilayer feedforward networks are universal approxi- mators. Neural Networks, 2(5), 359-366

    1989

  34. [34]

    Ta, H. T. (2024). BSRBF-KAN: a combination of B-splines and radial basis functions in Kolmogorov– Arnold networks. In International Symposium on Information and Communication Technology. Singapore: Springer Nature Singapore, 3-15

    2024

  35. [35]

    S., Pato, M., Datia, N

    Costa, S. S., Pato, M., Datia, N. (2024). An empirical study on the application of KANs for classification. In Proceedings of the 2024 8th International Conference on Advances in Artificial Intelligence, 308-314

    2024

  36. [36]

    Xu, J., Chen, Z., Li, J., Yang, S., Wang, W., et al. (2024). FourierKAN-GCF: Fourier Kolmogorov-Arnold Network–An Effective and Efficient Feature Transformation for Graph Collaborative Filtering. arXiv preprint arXiv:2406.01034

    Pith/arXiv arXiv 2024

  37. [37]

    Xu, Z., Lv, B. (2025). Enhancing Physics-Informed Neural Networks with a Hybrid Parallel Kolmogorov– Arnold and MLP Architecture. arXiv preprint arXiv:2503.23289

    Pith/arXiv arXiv 2025

  38. [38]

    A., Jacob, B., Murphy, S

    Howard, A. A., Jacob, B., Murphy, S. H., Heinlein, A., Stinis, P. (2024). Finite basis kolmogorov- arnold networks: domain decomposition for data-driven and physics-informed problems. arXiv preprint arXiv:2406.19662

    arXiv 2024

  39. [39]

    E., Likas, A., Fotiadis, D

    Lagaris, I. E., Likas, A., Fotiadis, D. I. (1998). Artificial neural networks for solving ordinary and partial differential equations. IEEE Transactions on Neural Networks, 9(5), 987-1000

    1998

  40. [40]

    Ji, Y., Zha, Y., Yeh, T. C. J., Shi, L., Wang, Y. (2024). Groundwater inverse modeling: Physics-informed neural network with disentangled constraints and errors. Journal of Hydrology, 640, 131703

    2024

  41. [41]

    A., Agarwal, P., Elsayed, E

    Juraev, D. A., Agarwal, P., Elsayed, E. E., Targyn, N. (2024). Helmholtz equations and their applications in solving physical problems. Advanced Engineering Science, 4, 54-64

    2024

  42. [42]

    Satnoianu, R. A. (2003). Coexistence of stationary and traveling waves in reaction-diffusion-advection systems. Physical Review E, 68(3), 032101

    2003

  43. [43]

    Kang, S., Hwang, Y., Kim, J. (2025). Unconditionally stable method for the high-order Allen–Cahn equation. Journal of Computational Science, 102636. 24

    2025