Pith. sign in

REVIEW 4 major objections 4 minor 32 references

Adaptive Basis-inspired Deep Neural Network for Solving Partial Differential Equations with Localized Features

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

Pith's one-line read This paper claims that a neural network whose first layer is made of finite-element-style "basis-inspired blocks"—each initialized to reproduce a hat function—can capture sharp gradients and singularities in PDEs far better than a…

desk verdict A genuinely new FEM-inspired block architecture with real gains in the reported tests, but the adaptive placement mechanism is not yet supported in the one nonlinear case and the experiments need repeats; worth refereeing, not ready to crown. read the letter →

arxiv 2412.00636 v1 pith:VAPEY6ZT submitted 2024-12-01 math.NA cs.NA

classification math.NAcs.NA MSC 65M5068T9935Q6835J75
keywords Adaptivebasis-inspireddeepneuralnetworkphysics-informednetworkspartialdifferentialequationssingularityfiniteelementbasisfunctionKolmogorovSuperpositionTheoremlocalizedfeatures
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 is trying to establish that the weakness of physics-informed neural networks on localized features—cusp singularities, steep peaks, shocks—is fixable at the architecture level. It introduces the BI-block, a small subnetwork whose weights are initialized from a nodal position and its neighboring interval lengths so that the block behaves like a linear finite element basis function; a network assembled from these blocks is BI-DNN. Building on the Kolmogorov Superposition Theorem, the paper extends the construction from one dimension to several without tensor-product explosion. It then wraps BI-DNN in an adaptive "solve, estimate, mark, enhance" loop, ABI-DNN, in which new BI-blocks are added at clusters of high PDE residual. The reported experiments support the claim that BI-DNN reaches relative errors roughly one to two orders of magnitude below PINN with comparable trainable parameters, and that ABI-DNN reaches a prescribed tolerance with an architecture whose error matches or beats a fixed BI-DNN of the same size.

What carries the argument

The carrying object is the Basis-inspired Block (BI-block): a short subnetwork defined by equations (3.2) (ReLU) and (3.5) (Tanh), whose weights and biases are initialized from a nodal coordinate $x_j$ and adjacent interval lengths $h_{j-1}, h_j$ so that its output is a finite element hat function $\varphi_j(x)$—exactly for ReLU, approximately for Tanh, which is used for smoothness in PINN training. Because each block's attention region is set by its initialization, the network designer knows which part of the domain a block is responsible for. BI-DNN stacks these blocks and feeds them into fully connected layers; the multi-dimensional version follows the Kolmogorov Superposition Theorem to avoid tensor products. ABI-DNN repeats solve-estimate-mark-enhance: it evaluates the residual indicator $\eta_p = |\mathcal{L} u_{NN}(x_p) - f(x_p)|$ on a fixed sample set, marks points above a fraction $\gamma$ of the maximum, clusters them with DBSCAN, and inserts new BI-blocks at cluster centroids with radius-scaled supports. This machinery is what turns "add neurons" from an empirical gamble into a localized enrichment operation.

What would settle it

In the Burgers setting, compute the overlap between the clusters marked by equation (4.1) and the top-decile set of pointwise testing errors after each adaptive iteration; the paper already displays non-overlap in its Figures 22 and 23. A run where the final ABI-DNN, trained with the same epochs and initial model as a fixed BI-DNN of the final size, fails to beat that fixed BI-DNN—or where moving the clusters to the true high-error regions changes the outcome—would falsify the claim that residual-guided block insertion is what produces the gain.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that translating the finite element hat function into a trainable neural block gives the network a built-in notion of locality: the block's attention region is known from its initialization, so adding a block at a specified location is a controlled act of local enrichment rather than a blind widening of a hidden layer. In one dimension the block is exact for ReLU and approximate for Tanh; in higher dimensions the blocks are assembled per coordinate under the Kolmogorov Superposition Theorem, and the remaining layers mix their outputs. The same trainable node positions give a form of r-adaptivity, and the adaptive loop adds blocks where the residual marks trouble, a form of h-adaptivity. The paper's numerical section claims this combination yields one-to-two-order-of-magnitude relative $L^2$ error reductions over standard PINN on cusped and high-frequency function fitting, Poisson problems with one and two steep peaks, a re-entrant corner problem, and the viscous Burgers equation, with the adaptive version typically matching or improving on the fixed BI-DNN at the same final parameter count.

Load-bearing premise

The load-bearing assumption is that the residual indicator $\eta_p = |\mathcal{L} u_{NN}(x_p) - f(x_p)|$ evaluated on the fixed sample set identifies the regions where the approximation truly needs new capacity; the paper's own Burgers experiment shows the marked clusters can lie away from the high pointwise-error regions, so if the indicator misfires the added blocks cannot deliver the claimed improvement.

Editorial extensions

If this is right

  • For localized-feature problems—cusp singularities, steep Gaussian peaks, re-entrant corners, shock-like Burgers profiles—BI-DNN with a similar parameter count to PINN is claimed to reduce relative $L^2$ error by one to two orders of magnitude.
  • ABI-DNN can start from a small network and grow it until a tolerance is met, so the user no longer has to guess the width in advance; the final network's error is comparable to or better than a fixed BI-DNN with the same architecture.
  • Because BI-block weights encode nodal positions, training the network is also moving the mesh (r-adaptivity), while adding blocks is mesh refinement (h-adaptivity); both mechanisms are available inside one optimization.
  • The KST-based assembly gives a coordinate-wise construction for multivariate problems, so the local-enrichment idea does not require tensor-product basis grids that grow exponentially with dimension.

Reading between the lines

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

  • Editorial inference: the BI-block front end can be read as a learnable mesh with known support, which suggests a natural diagnostic—monitor the trained nodal positions and block supports to visualize where the network actually concentrated its approximation power, a tool the paper does not build.
  • Editorial inference: the paper's own Burgers experiment shows the residual indicator's marked clusters can miss the regions of largest pointwise error, so replacing equation (4.1) with a gradient- or feature-aware indicator (or a dual-weighted residual) is the most direct testable upgrade; the paper names this as future work.
  • Editorial inference: to isolate whether the gain is architectural rather than due to extra training or parameter tuning, one could compare ABI-DNN against PINN with the same adaptive-sampling markers but no BI-blocks; the paper compares against fixed PINN only.
  • Editorial inference: the method's philosophy suggests a bridge to existing adaptive-sampling PINNs—use their residual-based sampling to place BI-blocks instead of just collocation points—which would combine both lines of work.
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

4 major / 4 minor

Summary. The paper introduces Basis-inspired Blocks (BI-blocks), which are small subnetworks designed to reproduce or approximate finite-element hat functions, and builds two architectures on them: BI-DNN, a fixed network with BI-blocks in the first layer per input dimension, and ABI-DNN, which repeatedly solves a PINN loss, marks collocation points with large PDE residual via Eq. (4.1), clusters them with DBSCAN, and inserts new BI-blocks at cluster centroids. The authors report numerical experiments on singular and high-frequency function fitting, Poisson equations with peaks and a re-entrant corner, and the viscous Burgers equation, claiming that BI-DNN outperforms PINNs with comparable parameter counts and that ABI-DNN automatically grows an architecture with error comparable to or better than the fixed BI-DNN.

Significance. If the performance claims hold, the work offers a practical way to inject FEM-like locality into PINNs, with an interpretable analogue of h-refinement. The BI-block construction is transparent, and the adaptive insertion via residual marking and density clustering is a clear algorithmic contribution. The paper is honest about the Burgers indicator limitation. However, the significance is tempered by missing statistical rigor (single runs), a theoretical gap for the Tanh block, and the absence of a control that isolates targeted placement from capacity increase.

major comments (4)
  1. [§5.4, Figs. 22-23] The paper's own Burgers experiment shows that the residual indicator (4.1) marks clusters near x=0 that do not overlap the high pointwise-error regions, and the text concedes that the indicator 'may not be sufficiently effective' in that problem. Since targeted placement is the central novelty of ABI-DNN, this counterexample undercuts the claim that the adaptive mechanism, rather than added capacity and warm-start retraining, is responsible for the gains in Table 8. A control experiment that adds the same number of BI-blocks at random or uniformly spaced locations and retrains with the same epochs is needed to support the adaptive-placement claim.
  2. [Tables 1-8] All numerical results are from a single run per configuration; no seeds, repeats, or error bars are reported. The statements 'significantly lower' and 'comparable' rest on point comparisons that could be within run-to-run variance, especially for the cases where ABI-DNN improves on fixed BI-DNN by factors of 2-4 (e.g., Tables 1 and 5). The authors should either provide repeated runs with statistics or clearly restrict their claims to the particular trajectories shown.
  3. [§3.1.2, Eq. (3.5)] The Tanh BI-block is obtained by replacing the piecewise-linear saturation s(x) with tanh(x) after a first-order Taylor expansion near x=0, but no error bound is given for the discrepancy between the Tanh block and the intended FEM basis function, nor for the resulting residual in the PINN loss. Because all numerical experiments use the Tanh block, the exactness of the ReLU representation (3.2) does not justify the architecture actually deployed; a quantitative approximation statement is needed to ground the 'basis-inspired' interpretation.
  4. [Table 4] For Problem (5.2) with 5 initial blocks, Table 4 reports 39 adaptive iterations, while Section 5.1 states that the maximum number of adaptive iterations is J=10 in all experiments. This inconsistency must be resolved, and the stopping criterion (η ≤ ηtol vs. J cap) clarified, because it affects reproducibility and the claim that ABI-DNN stops once a prescribed tolerance is met.
minor comments (4)
  1. [Algorithm 1] Algorithm 1 line 3 uses 'while η > ϵ', but ϵ is not defined; the tolerance is called ηtol elsewhere in the paper.
  2. [§5.1] The DBSCAN parameters ε=0.1 and MinPts=1 are fixed across all problem domains (e.g., [0,1], [−1,1]^2, and the space-time domain), and no sensitivity study is given; a brief discussion of how these choices scale with domain size would help.
  3. [§3.2.3] The d-dimensional BI-DNN (3.13) is a fully connected generalization of the KST-inspired form (3.12), but the text does not discuss whether (3.13) retains any approximation guarantees; the motivation should be stated more cautiously.
  4. [§3.1.1] In Eq. (3.3), W2^1 is a diagonal matrix while W1^1 is a vector; a brief note on the intended dimensions would improve readability.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: empirical comparisons use holdout test errors, the adaptive indicator is residual-based rather than fitted, and the Burgers limitation is a reliability issue, not a circular reduction.

full rationale

This paper is not circular. The central empirical claims are comparisons of relative L2 errors on holdout uniform test grids (Sec 5.1), so the reported performance is not constructed from the quantities used to build or adapt the network. The adaptive indicator in Eq. (4.1) is a residual evaluated on training collocation points; it is not calibrated to test errors, and the final ABI-DNN error is sometimes worse than the fixed BI-DNN with the same architecture (e.g., Table 2: 8.00e-3 vs 4.22e-3; Table 6: 5.36e-3 vs 1.87e-3), which shows the claimed 'comparable error' is an empirical outcome rather than an identity. The KST connection is presented as inspiration, not as a theorem that forces the architecture: Eq. (3.13) is explicitly called 'a generalization' of the KST-motivated form (3.12), so the multidimensional BI-DNN is a heuristic ansatz with a plausible motivation, which is a rigor/correctness concern rather than circularity. The vectorized KST notation attributed to the authors' submitted [29] is a direct rewriting of Theorem 1 and is not load-bearing. The paper also candidly discloses the limitation of its indicator in Sec 5.4: points marked by (4.1) "do not precisely overlap with the regions of high pointwise error," and "the error indicator (4.1) may not be sufficiently effective in complex problems like the Burgers equation." This is a reliability concern for the adaptive mechanism, but it is the opposite of circularity: the method's failure mode is empirically checked against an independent reference solution, not defined into success. No load-bearing step reduces by construction to its inputs, and the only self-citation ([29], for vectorized notation) is non-essential.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The central method relies on a handful of hand-chosen hyperparameters, a standard theorem (KST), and two unproved modeling assumptions: the accuracy of replacing s by tanh, and the persistence of locality of BI-blocks after training. No new physical entities are introduced.

free parameters (5)
  • Marking threshold gamma = 0.5
    Controls the fraction of residual points marked for refinement in Eq. (4.2); chosen by hand.
  • DBSCAN parameters epsilon and MinPts = epsilon = 0.1, MinPts = 1
    Control clustering of marked points in Sec. 5.1; MinPts=1 prevents any point from being labeled noise.
  • BI-block width scaling s = 2
    New blocks are initialized with radius rs = s*r in Eq. (4.3); s=2 is chosen by hand to enhance flexibility.
  • Loss penalty beta = 1000
    Weights boundary and initial residuals in loss (2.3); fixed by hand for all experiments.
  • Training schedule per adaptive iteration = lr 5e-3 decay 0.9 per 2500 steps, 10000 or 15000 epochs
    Chosen by hand in Sec. 5.1; affects how well each enhanced network is optimized.
assumptions (4)
  • standard math Kolmogorov Superposition Theorem with Lipschitz inner functions
    Invoked in Sec. 3.2.2 as the theoretical basis for the multi-dimensional BI-DNN construction.
  • ad hoc to paper Tanh approximates the piecewise-linear saturation function s(x)
    Eq. (3.5) replaces s(x) with tanh(x) to build smooth BI-blocks; no quantitative error bound is provided for this replacement.
  • domain assumption BI-blocks initialized at cluster centroids remain localized after training
    The adaptive strategy assumes a block inserted at a marked region will focus on that region as weights evolve; the paper provides only empirical illustrations, not a proof.
  • ad hoc to paper The multi-dimensional network (3.13) inherits the approximation properties of the KST-inspired sparse form (3.12)
    The dense fully-connected generalization with added activations is heuristic; the paper gives no theorem or error bound connecting (3.13) to KST.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Adaptive Basis-inspired Deep Neural Network for Solving Partial Differential Equations with Localized Features." pith.science (2026). https://pith.science/paper/VAPEY6ZT

@misc{pith2026241200636,
  author       = {Pith},
  title        = {Pith review of: Adaptive Basis-inspired Deep Neural Network for Solving Partial Differential Equations with Localized Features},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/VAPEY6ZT}},
  note         = {Machine review of arXiv:2412.00636}
}
read the original abstract

This paper proposes an Adaptive Basis-inspired Deep Neural Network (ABI-DNN) for solving partial differential equations with localized phenomena such as sharp gradients and singularities. Like the adaptive finite element method, ABI-DNN incorporates an iteration of "solve, estimate, mark, enhancement", which automatically identifies challenging regions and adds new neurons to enhance its capability. A key challenge is to force new neurons to focus on identified regions with limited understanding of their roles in approximation. To address this, we draw inspiration from the finite element basis function and construct the novel Basis-inspired Block (BI-block), to help understand the contribution of each block. With the help of the BI-block and the famous Kolmogorov Superposition Theorem, we first develop a novel fixed network architecture named the Basis-inspired Deep Neural Network (BI-DNN), and then integrate it into the aforementioned adaptive framework to propose the ABI-DNN. Extensive numerical experiments demonstrate that both BI-DNN and ABI-DNN can effectively capture the challenging singularities in target functions. Compared to PINN, BI-DNN attains significantly lower relative errors with a similar number of trainable parameters. When a specified tolerance is set, ABI-DNN can adaptively learn an appropriate architecture that achieves an error comparable to that of BI-DNN with the same structure.

Figures

Figures reproduced from arXiv: 2412.00636 by the authors.

Figure 1
Figure 1. Illustration of basis-inspired blocks with activation functions: (a) ReLU (Equation ( [PITH_FULL_IMAGE:figures/full_fig_p006_1.png] view at source ↗
Figure 2
Figure 2. Image of basis-inspired blocks with (a) ReLU and (b) Tanh activation functions, respec [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗
Figure 3
Figure 3. Illustration of the architecture of a one-dimensional BI-DNN consisting of three BI [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figures from the paper (20 more)
Figure 4
Figure 4. Figure 4: Illustration of the architecture of a two-dimensional BI-DNN consisting of two BI-blocks [PITH_FULL_IMAGE:figures/full_fig_p011_4.png]
Figure 5
Figure 5. Figure 5: Comparison of relative errors between frozen BI-DNNs and BI-DNNs on Problem ( [PITH_FULL_IMAGE:figures/full_fig_p016_5.png]
Figure 6
Figure 6. Figure 6: The change of relative errors of DNN and BI-DNN along with the number of trainable [PITH_FULL_IMAGE:figures/full_fig_p016_6.png]
Figure 7
Figure 7. Figure 7: Illustration of the target function and approximating functions on problem ( [PITH_FULL_IMAGE:figures/full_fig_p018_7.png]
Figure 8
Figure 8. Figure 8: Illustration of clustering results of ABI-DNN after 1st, 2nd, and 4th adaptive iterations [PITH_FULL_IMAGE:figures/full_fig_p019_8.png]
Figure 9
Figure 9. Figure 9: Illustration of the target function and approximating functions on problem ( [PITH_FULL_IMAGE:figures/full_fig_p020_9.png]
Figure 10
Figure 10. Figure 10: Illustration of clustering results of ABI-DNN after 1st, 3nd, and 5th adaptive iterations [PITH_FULL_IMAGE:figures/full_fig_p021_10.png]
Figure 11
Figure 11. Figure 11: Convergence processes of DNN, BI-DNN and ABI-DNN on Problem ( [PITH_FULL_IMAGE:figures/full_fig_p021_11.png]
Figure 12
Figure 12. Figure 12: The change of relative L 2 errors of PINN and BI-DNN along with the number of trainable parameters on Problem (5.4). after 45000 epochs of training. Notably, the final ABI-DNN(b= [12,12]) achieves an error lower than the fixed BI-DNN with the same architecture and app…
Figure 13
Figure 13. Figure 13: Exact solution and pointwise absolute errors of PINN(w=19), BI-DNN(b=[12,12]) and [PITH_FULL_IMAGE:figures/full_fig_p024_13.png]
Figure 14
Figure 14. Figure 14: Illustration of clustering results of ABI-DNN after 1st and 2nd adaptive iterations on [PITH_FULL_IMAGE:figures/full_fig_p024_14.png]
Figure 15
Figure 15. Figure 15: illustrates the error comparison between the PINN and the BI-DNN. Consistent with the trends in [PITH_FULL_IMAGE:figures/full_fig_p025_15.png]
Figure 16
Figure 16. Figure 16: Exact solution and pointwise absolute errors of PINN(w=30), BI-DNN(b=[20,20]) and [PITH_FULL_IMAGE:figures/full_fig_p026_16.png]
Figure 17
Figure 17. Figure 17: Illustration of clustering results of ABI-DNN after 1st, 3rd and 5th adaptive iterations [PITH_FULL_IMAGE:figures/full_fig_p027_17.png]
Figure 18
Figure 18. Figure 18: The change of relative L 2 errors of PINN and BI-DNN along with the number of trainable parameters on Problem (5.6) [PITH_FULL_IMAGE:figures/full_fig_p028_18.png]
Figure 19
Figure 19. Figure 19: Exact solution and pointwise absolute errors of PINN(w=21), BI-DNN(b=[14,14]) and [PITH_FULL_IMAGE:figures/full_fig_p029_19.png]
Figure 20
Figure 20. Figure 20: Illustration of clustering results of ABI-DNN after 1st and 2nd adaptive iterations on [PITH_FULL_IMAGE:figures/full_fig_p029_20.png]
Figure 21
Figure 21. Figure 21: The change of relative L 2 errors of PINN and BI-DNN along with the number of trainable parameters on Problem (5.7) [PITH_FULL_IMAGE:figures/full_fig_p031_21.png]
Figure 22
Figure 22. Figure 22: The reference solution and pointwise absolute errors of PINN, BI-DNN, and ABI-DNN [PITH_FULL_IMAGE:figures/full_fig_p032_22.png]
Figure 23
Figure 23. Figure 23: Illustration of clustering results of ABI-DNN after 1st and 2nd adaptive iterations on [PITH_FULL_IMAGE:figures/full_fig_p032_23.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

32 extracted references · 28 canonical work pages

  1. [17]

    Self-adaptive deep neural network: Numerical approx- imation to functions and PDEs

    Zhiqiang Cai, Jingshuang Chen, and Min Liu. Self-adaptive deep neural network: Numerical approx- imation to functions and PDEs. Journal of Computational Physics, 455:111021, 2022

  2. [1]

    The deep Ritz method: a deep learning-based numerical algorithm for solving variational problems

    Weinan E and Bing Yu. The deep Ritz method: a deep learning-based numerical algorithm for solving variational problems. Communications in Mathematics and Statistics, 6(1):1–12, 2018

  3. [2]

    Karniadakis

    Maziar Raissi, Paris Perdikaris, and George E. Karniadakis. Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differ- ential equations. Journal of Computational Physics, 378:686–707, 2019

  4. [3]

    DGM: A deep learning algorithm for solving partial differential equations

    Justin Sirignano and Konstantinos Spiliopoulos. DGM: A deep learning algorithm for solving partial differential equations. Journal of Computational Physics, 375:1339–1364, 2018

  5. [4]

    Physics- informed neural networks (PINNs) for fluid mechanics: A review

    Shengze Cai, Zhiping Mao, Zhicheng Wang, Minglang Yin, and George Em Karniadakis. Physics- informed neural networks (PINNs) for fluid mechanics: A review. Acta Mechanica Sinica , 37(12):1727–1738, 2021

  6. [5]

    Artificial neural network mixed model for large eddy simulation of compressible isotropic turbulence

    Chenyue Xie, Jianchun Wang, Hui Li, Minping Wan, and Shiyi Chen. Artificial neural network mixed model for large eddy simulation of compressible isotropic turbulence. Physics of Fluids , 31(8):085112, 2019

  7. [6]

    Modeling subgrid-scale forces by spatial artificial neural networks in large eddy simulation of turbulence

    Chenyue Xie, Jianchun Wang, and Weinan E. Modeling subgrid-scale forces by spatial artificial neural networks in large eddy simulation of turbulence. Physical Review Fluids, 5(5):054606, 2020

  8. [7]

    A composite neural network that learns from multi-fidelity data: Application to function approximation and inverse PDE problems

    Xuhui Meng and George Em Karniadakis. 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

Show all 32 references
  1. [8]

    fPINNs: Fractional physics-informed neural networks

    Guofei Pang, Lu Lu, and George Em Karniadakis. fPINNs: Fractional physics-informed neural networks. SIAM Journal on Scientific Computing, 41(4):A2603–A2626, 2019

  2. [9]

    A comprehensive study of non- adaptive and residual-based adaptive sampling for physics-informed neural networks

    Chenxi Wu, Min Zhu, Qinyang Tan, Yadhu Kartha, and Lu Lu. A comprehensive study of non- adaptive and residual-based adaptive sampling for physics-informed neural networks. Computer Methods in Applied Mechanics and Engineering, 403:115671, 2023

  3. [10]

    Failure-informed adaptive sampling for PINNs

    Zhiwei Gao, Liang Yan, and Tao Zhou. Failure-informed adaptive sampling for PINNs. SIAM Journal on Scientific Computing, 45(4):A1971–A1994, 2023

  4. [11]

    Failure-informed adaptive sampling for PINNs, part II: combining with re-sampling and subset simulation.Communications on Applied Mathematics and Computation, 6(3):1720–1741, 2024

    Zhiwei Gao, Tao Tang, Liang Yan, and Tao Zhou. Failure-informed adaptive sampling for PINNs, part II: combining with re-sampling and subset simulation.Communications on Applied Mathematics and Computation, 6(3):1720–1741, 2024. 34

  5. [12]

    Jagtap, Kenji Kawaguchi, and George Em Karniadakis

    Ameya D. Jagtap, Kenji Kawaguchi, and George Em Karniadakis. Adaptive activation functions accelerate convergence in deep and physics-informed neural networks. Journal of Computational Physics, 404:109136, 2020

  6. [13]

    Jagtap, Kenji Kawaguchi, and George Em Karniadakis

    Ameya D. Jagtap, Kenji Kawaguchi, and George Em Karniadakis. Locally adaptive activation func- tions with slope recovery for deep and physics-informed neural networks. Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences, 476(2239):20200334, 2020

  7. [14]

    Self-adaptive physics-informed neural networks using a soft attention mechanism

    Levi McClenny and Ulisses Braga-Neto. Self-adaptive physics-informed neural networks using a soft attention mechanism. Journal of Computational Physics, 474:111722, 2023

  8. [15]

    Self-adaptive loss balanced physics-informed neural networks

    Zixue Xiang, Wei Peng, Xu Liu, and Wen Yao. Self-adaptive loss balanced physics-informed neural networks. Neurocomputing, 496:11–34, 2022

  9. [16]

    Taylor, Manuela Bastidas, Victor M

    Jamie M. Taylor, Manuela Bastidas, Victor M. Calo, and David Pardo. Adaptive deep fourier resid- ual method via overlapping domain decomposition. Computer Methods in Applied Mechanics and Engineering, 427:116997, 2024

  10. [18]

    Adaptive two-layer ReLU neural network: II

    Min Liu and Zhiqiang Cai. Adaptive two-layer ReLU neural network: II. Ritz approximation to elliptic PDEs. Computers & Mathematics with Applications, 113:103–116, 2022

  11. [19]

    Adaptive two-layer relu neural network: I

    Min Liu, Zhiqiang Cai, and Jingshuang Chen. Adaptive two-layer relu neural network: I. best least- squares approximation. Computers & Mathematics with Applications, 113:34–44, 2022

  12. [20]

    Kingma and Jimmy Lei Ba

    Diederik P. Kingma and Jimmy Lei Ba. Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980, 2014

  13. [21]

    Influence of activation functions on the convergence of physics-informed neural networks for 1d wave equation

    Paweł Maczuga and Maciej Paszy ´nski. Influence of activation functions on the convergence of physics-informed neural networks for 1d wave equation. International Conference on Computational Science. Cham: Springer Nature Switzerland, page 74–88, 2023

  14. [22]

    On the representation of continuous functions of many variables by superposition of continuous functions of one variable and addition

    Andrei Nikolaevich Kolmogorov. On the representation of continuous functions of many variables by superposition of continuous functions of one variable and addition. Dokl. Akad. Nauk SSSR , 114(5):953–956, 1957

  15. [23]

    George G. Lorentz. Metric entropy, widths, and superpositions of functions. The American Mathe- matical Monthly, 69(6):469–485, 1962

  16. [24]

    George G. Lorentz. Approximation of functions.-holt, rinehart and wilson. Inc., New York, 1966

  17. [25]

    Sprecher and Sorin Draghici

    David A. Sprecher and Sorin Draghici. Space-filling curves and kolmogorov superposition-based neural networks. Neural Networks, 15(1):57–67, 2002

  18. [26]

    The kolmogorov superposition theorem can break the curse of dimensionality when approximating high dimensional functions

    Ming-Jun Lai and Zhaiming Shen. The kolmogorov superposition theorem can break the curse of dimensionality when approximating high dimensional functions. arXiv: 2112. 09963v4, 2023

  19. [27]

    Kolmogorov’s theorem and multilayer neural networks

    V ˇera K ˚urkov´a. Kolmogorov’s theorem and multilayer neural networks. Neural Networks , 5(3):501–506, 1992

  20. [28]

    The kolmogorov-arnold representation theorem revisited

    Johannes Schmidt-Hieber. The kolmogorov-arnold representation theorem revisited. Neural net- works, 137:119–126, 2021

  21. [29]

    A kol- mogorov high order deep neural network for high frequency partial differential equations in high dimensions

    Yaqin Zhang, Ke Li, Zhipeng Chang, Xuejiao Liu, Yunqing Huang, and Xueshuang Xiang. A kol- mogorov high order deep neural network for high frequency partial differential equations in high dimensions. submitted, 2024

  22. [30]

    Selected topics in finite element methods

    Zhiming Chen and Haijun Wu. Selected topics in finite element methods . Beijing: Science Press, 2010

  23. [31]

    DBSCAN: Density-based spatial clustering of appli- cations with noise

    Fangming Bi, Weikui Wang, and Long Chen. DBSCAN: Density-based spatial clustering of appli- cations with noise. Journal of Nanjing University(Natural Sciences), 48(4):491–498, 2012

  24. [32]

    Moving sampling physics-informed neural networks induced by moving mesh PDE

    Yu Yang, Qihong Yang, Yangtao Deng, and Qiaolin He. Moving sampling physics-informed neural networks induced by moving mesh PDE. Neural Networks, 180:106706, 2024

Pith tools

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