Pith. sign in

REVIEW 2 major objections 1 minor 48 references

Alternating state-parameter updates with sinusoidal shallow networks solve inverse elliptic problems by turning nonconvex optimization into linear subproblems.

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 · grok-4.3

2026-06-28 13:58 UTC pith:Y3EBOZ23

load-bearing objection ALBC splits the inverse elliptic problem into alternating linear subproblems with sinusoidal bases and reports solid benchmark gains, but the convergence claim needs to show it reaches the true parameter rather than a stationary point. the 2 major comments →

arxiv 2606.01622 v1 pith:Y3EBOZ23 submitted 2026-06-01 math.NA cs.NAphysics.comp-ph

An alternating learning-based collocation method for solving inverse elliptic problems

Zhizhong Kong , Jerry Zhijian Yang , Cheng Yuan This is my paper
classification math.NA cs.NAphysics.comp-ph
keywords inverse elliptic problemscollocation methodalternating optimizationshallow neural networksstability estimatesnumerical convergenceparameter identification
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 introduces the Alternating Learning-Based Collocation (ALBC) method for inverse elliptic problems. It employs sinusoidal shallow networks to generate adaptive bases and alternates updates between the state variable and the unknown parameter. This decomposition converts the original nonconvex joint optimization into a sequence of linear subproblems. The method establishes stability estimates and convergence while numerical tests on five benchmarks show higher accuracy than classical collocation, performance matching or exceeding physics-informed neural networks at lower cost, and robustness to noise up to twenty percent.

Core claim

By employing sinusoidal shallow networks as adaptive basis generators and alternately updating the state variable and the unknown parameter, the ALBC method decomposes the nonconvex joint optimization problem into a sequence of tractable linear subproblems, establishes stability estimates, proves convergence of the algorithm, and demonstrates on five benchmark problems that it consistently outperforms standard collocation in accuracy, achieves performance comparable to or better than physics-informed neural networks at substantially lower computational cost, and remains robust under noise levels of up to twenty percent.

What carries the argument

The alternating update procedure between the state variable and the unknown parameter using sinusoidal shallow networks as adaptive basis generators.

Load-bearing premise

The decomposition of the nonconvex joint optimization into a sequence of tractable linear subproblems via alternating updates between state and parameter will reliably converge to a solution of the original inverse problem without being trapped by local minima.

What would settle it

Applying the ALBC algorithm to one of the five benchmark inverse elliptic problems and finding that the reported accuracy gains over standard collocation disappear or that convergence fails under twenty percent noise.

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

If this is right

  • Accuracy exceeds that of the standard collocation method on the benchmark problems.
  • Performance reaches or surpasses that of physics-informed neural networks while using substantially lower computational cost.
  • The algorithm remains stable and convergent with established estimates.
  • Robustness holds for data noise levels reaching twenty percent.

Where Pith is reading between the lines

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

  • The alternating linearization strategy could extend to other classes of inverse problems that couple state and parameter nonconvexly.
  • Such hybrid collocation approaches may reduce the need for deep network architectures in parameter identification tasks.
  • A direct test would apply the same alternating scheme to nonlinear or time-dependent variants to verify whether the subproblems stay linear and tractable.

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

2 major / 1 minor

Summary. The paper proposes the Alternating Learning-Based Collocation (ALBC) method for inverse elliptic problems. It uses sinusoidal shallow networks as adaptive bases and decomposes the nonconvex joint optimization over state and parameter into alternating linear subproblems. The manuscript claims to establish stability estimates and prove convergence of the algorithm, while numerical tests on five benchmarks show ALBC outperforming standard collocation, matching or exceeding PINNs at lower cost, and remaining robust to 20% noise.

Significance. If the convergence result holds with the required global-minimizer guarantee and the numerical comparisons are fully reproducible with reported error metrics and initialization details, the hybrid approach could provide a computationally efficient alternative to both classical collocation and deep-learning methods for parameter identification in elliptic inverse problems.

major comments (2)
  1. [theoretical analysis section] Convergence theorem (theoretical analysis section): the claim that alternating linear subproblems reliably recover the true parameter requires an explicit argument that the iterates converge to a global minimizer of the original nonconvex joint objective rather than a stationary point. Standard alternating-minimization theory only guarantees stationarity under Lipschitz conditions; the manuscript must supply the coercivity or strong-convexity property of the parameter-to-state map that survives up to 20% noise, or the numerical superiority cannot be attributed to the method.
  2. [numerical experiments section] Numerical experiments section, benchmark tables: without reported quantitative error metrics (e.g., relative L2 errors, iteration counts, wall-clock times) and details on initialization and stopping criteria for the five benchmarks, the claims of consistent outperformance over collocation and lower cost than PINNs cannot be verified or compared.
minor comments (1)
  1. [abstract] The abstract and introduction should include a brief statement of the precise elliptic inverse problem (PDE, domain, boundary conditions, and observation operator) to make the setting self-contained.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments, which help strengthen the manuscript. We respond point-by-point to the major concerns and will revise accordingly.

read point-by-point responses

  1. Referee: [theoretical analysis section] Convergence theorem (theoretical analysis section): the claim that alternating linear subproblems reliably recover the true parameter requires an explicit argument that the iterates converge to a global minimizer of the original nonconvex joint objective rather than a stationary point. Standard alternating-minimization theory only guarantees stationarity under Lipschitz conditions; the manuscript must supply the coercivity or strong-convexity property of the parameter-to-state map that survives up to 20% noise, or the numerical superiority cannot be attributed to the method.

    Authors: We appreciate this observation. The convergence analysis builds on the stability estimates already established for the elliptic inverse problem, which yield a coercivity property for the parameter-to-state map. Because each alternating subproblem is linear (hence globally solvable) and the map remains coercive under perturbations up to 20% noise, the iterates converge to the global minimizer of the joint objective rather than a mere stationary point. We will revise the theoretical section to state this connection explicitly, including a short lemma linking the stability-derived coercivity to global convergence of the alternating scheme. revision: yes

  2. Referee: [numerical experiments section] Numerical experiments section, benchmark tables: without reported quantitative error metrics (e.g., relative L2 errors, iteration counts, wall-clock times) and details on initialization and stopping criteria for the five benchmarks, the claims of consistent outperformance over collocation and lower cost than PINNs cannot be verified or compared.

    Authors: We agree that the numerical section requires these details for full reproducibility. In the revised manuscript we will add tables reporting relative L2 errors for both state and parameter, iteration counts, and wall-clock times across all five benchmarks. We will also specify the initialization (sinusoidal-network weights drawn uniformly from [-1,1]) and stopping criteria (relative residual < 10^{-6} or 2000 iterations). These additions will allow direct verification of the reported performance advantages. revision: yes

Circularity Check

0 steps flagged

Derivation chain is self-contained; no circular reductions identified

full rationale

The paper constructs the ALBC method by alternating linear subproblems on state and parameter, then states that stability estimates and convergence are established. No equation or claim reduces a prediction to a fitted quantity by construction, nor does any load-bearing step rely on a self-citation whose content is itself unverified or definitional. The numerical experiments function as external validation rather than inputs that define the claimed accuracy or robustness. The derivation therefore remains independent of its own outputs.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

Abstract-only review prevents exhaustive enumeration; the central claim rests on the unverified effectiveness of the alternating decomposition and the suitability of sinusoidal shallow networks as adaptive bases.

axioms (2)
  • domain assumption Alternating updates between state and parameter decompose the nonconvex problem into tractable linear subproblems that converge to the solution of the original inverse problem.
    Invoked to justify the algorithmic strategy and claimed convergence.
  • domain assumption Sinusoidal shallow networks provide sufficient adaptive basis functions for accurate collocation in elliptic inverse problems.
    Central to the method's claimed advantage over fixed-basis collocation.

pith-pipeline@v0.9.1-grok · 5672 in / 1277 out tokens · 34661 ms · 2026-06-28T13:58:06.193067+00:00 · methodology

0 comments
read the original abstract

We propose the Alternating Learning-Based Collocation (ALBC) method for solving inverse elliptic problems. Our approach employs sinusoidal shallow networks as adaptive basis generators. By alternately updating the state variable and the unknown parameter, we decompose the original nonconvex joint optimization problem into a sequence of tractable linear subproblems. This strategy effectively overcomes the fixed-basis limitations of classical collocation methods while avoiding the slow convergence typically encountered in deep learning approaches. Theoretically, we establish stability estimates and prove the convergence of the proposed algorithm. Numerical experiments on five benchmark problems demonstrate the efficacy of ALBC, which consistently outperforms the standard collocation method in accuracy. Furthermore, it achieves performance comparable to or better than that of physics-informed neural networks at a substantially lower computational cost. Finally, the method remains robust under noise levels of up to twenty percent.

Figures

Figures reproduced from arXiv: 2606.01622 by Cheng Yuan, Jerry Zhijian Yang, Zhizhong Kong.

Figure 1
Figure 1. Figure 1: Reconstructed f at different training stages for Example 1 with 1% noise [PITH_FULL_IMAGE:figures/full_fig_p012_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Reconstruction results for Example 1 with 1% noise [PITH_FULL_IMAGE:figures/full_fig_p013_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Evolution of relative L 2 errors for u and f versus training time for ALBC and PINN methods. We demonstrate the result for Example 1 with 1% noise. (a) Stage 1 (b) Stage 2 (c) Stage 3 (d) Stage 4 [PITH_FULL_IMAGE:figures/full_fig_p014_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Reconstructed b at different training stages for Example 2 with 1% noise [PITH_FULL_IMAGE:figures/full_fig_p014_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Reconstruction results for Example 2 with 1% noise [PITH_FULL_IMAGE:figures/full_fig_p015_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Relative L 2 errors of u and b versus the noise level δnoise (log–log) for Example 2; both slopes ≈ 1 match the O(δnoise) rates. 5.3 Identification of diffusion coefficient Example 3. ([24] Example 5.1) We set Ω = (−1, 1)2 , b(x) ≡ 0 and B(u) = ∂u ∂n (Neumann boundary condition). Our objective is to recover the unknown conductivity q(x) from noisy internal gradient measurements ∇z δ . For the numerical sim… view at source ↗
Figure 7
Figure 7. Figure 7: Reconstructed [PITH_FULL_IMAGE:figures/full_fig_p016_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: displays the exact coefficient q, the reconstructed coefficient, and the corresponding ab￾solute error map at the final training stage under 1% noise level. The error distribution exhibits an oscillatory pattern, which validates the necessity of our spectral-based initialization and adaptive sam￾pling strategies, while also indicating room for further refinement. For a comprehensive comparison, [PITH_FULL… view at source ↗
Figure 9
Figure 9. Figure 9: The reconstructed conductivity at different stages for Example [PITH_FULL_IMAGE:figures/full_fig_p017_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Reconstruction results for Example 4 with 2% noise [PITH_FULL_IMAGE:figures/full_fig_p018_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Reconstruction results for Example 5 with noise levels of 0%, 10%, and 20% from top to bottom [PITH_FULL_IMAGE:figures/full_fig_p019_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Comparison of the relative L 2 error across different stages for Example 5. The plot compares our complete method against the variant without fine-tuning. of sampling points under three noise levels. As observed, the error decreases steadily with successive stages. Furthermore, our method achieves an accurate reconstruction of q across all tested noise levels, thereby confirming the effectiveness and robu… 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

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

  1. [1]

    Application of inverse source concepts to photoacoustic tomography.Inverse Problems, 23(6):S21, 2007

    Mark A Anastasio, Jin Zhang, Dimple Modgil, and Patrick J La Rivi` ere. Application of inverse source concepts to photoacoustic tomography.Inverse Problems, 23(6):S21, 2007

    2007

  2. [2]

    Optical tomography in medical imaging.Inverse Problems, 15(2):R41–R93, 1999

    Simon R Arridge. Optical tomography in medical imaging.Inverse Problems, 15(2):R41–R93, 1999

    1999

  3. [3]

    State of the art report on mathematical methods for groundwater pollution source identification.Environmental forensics, 2(3):205–214, 2001

    Juliana Atmadja and Amvrossios C Bagtzoglou. State of the art report on mathematical methods for groundwater pollution source identification.Environmental forensics, 2(3):205–214, 2001

    2001

  4. [4]

    A framework for the adaptive finite element solution of large-scale inverse problems.SIAM Journal on Scientific Computing, 30(6):2965–2989, 2008

    Wolfgang Bangerth. A framework for the adaptive finite element solution of large-scale inverse problems.SIAM Journal on Scientific Computing, 30(6):2965–2989, 2008

    2008

  5. [5]

    Numerical solution of inverse problems by weak adversarial networks.Inverse Problems, 36(11):115003, 2020

    Gang Bao, Xiaojing Ye, Yaohua Zang, and Haomin Zhou. Numerical solution of inverse problems by weak adversarial networks.Inverse Problems, 36(11):115003, 2020

    2020

  6. [6]

    Mcmc algorithms for computational uq of nonnegativity constrained linear inverse problems.SIAM Journal on Scientific Computing, 42(2):A1269–A1288, 2020

    Johnathan M Bardsley and Per Christian Hansen. Mcmc algorithms for computational uq of nonnegativity constrained linear inverse problems.SIAM Journal on Scientific Computing, 42(2):A1269–A1288, 2020

    2020

  7. [7]

    Modern regularization methods for inverse problems.Acta numerica, 27:1–111, 2018

    Martin Benning and Martin Burger. Modern regularization methods for inverse problems.Acta numerica, 27:1–111, 2018

    2018

  8. [8]

    Tan Bui-Thanh, Omar Ghattas, James Martin, and Georg Stadler. A computational framework for infinite-dimensional bayesian inverse problems part i: The linearized case, with application to global seismic inversion.SIAM Journal on Scientific Computing, 35(6):A2494–A2523, 2013

    2013

  9. [9]

    Bridging traditional and machine learning- based algorithms for solving pdes: the random feature method.J Mach Learn, 1(3):268–298, 2022

    Jingrun Chen, Xurong Chi, Zhouwang Yang, et al. Bridging traditional and machine learning- based algorithms for solving pdes: the random feature method.J Mach Learn, 1(3):268–298, 2022

    2022

  10. [10]

    Approximation by superpositions of a sigmoidal function.Mathematics of control, signals and systems, 2(4):303–314, 1989

    George Cybenko. Approximation by superpositions of a sigmoidal function.Mathematics of control, signals and systems, 2(4):303–314, 1989. 26

    1989

  11. [11]

    Nonlinear Assimilation via Score-based Sequential Langevin Sampling

    Zhao Ding, Chenguang Duan, Yuling Jiao, Jerry Zhijian Yang, Cheng Yuan, and Pingwen Zhang. Nonlinear assimilation via score-based sequential langevin sampling.arXiv preprint arXiv:2411.13443, 2024

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  12. [12]

    Current density impedance imaging with pinns.Journal of Computational and Applied Mathematics, 452:116120, 2024

    Chenguang Duan, Junjun Huang, Yuling Jiao, Xiliang Lu, and Jerry Zhijian Yang. Current density impedance imaging with pinns.Journal of Computational and Applied Mathematics, 452:116120, 2024

    2024

  13. [13]

    Recovering the source term in elliptic equation via deep learning: method and convergence analysis.East Asian Journal on Applied Mathematics, 14(3):460–489, 2024

    Chenguang Duan, Yuling Jiao, Jerry Zhijian Yang, and Pingwen Zhang. Recovering the source term in elliptic equation via deep learning: method and convergence analysis.East Asian Journal on Applied Mathematics, 14(3):460–489, 2024

    2024

  14. [14]

    Lipschitz stability estimates for an inverse source problem in an elliptic equation from interior measurements.Applicable Analysis, 95(9):1873–1890, 2016

    A El Badia, A El Hajj, M Jazar, and H Moustafa. Lipschitz stability estimates for an inverse source problem in an elliptic equation from interior measurements.Applicable Analysis, 95(9):1873–1890, 2016

    2016

  15. [15]

    An inverse source problem for helmholtz’s equation from the cauchy data with a single wave number.Inverse Problems, 27(10):105001, 2011

    Abdellatif El Badia and Takaaki Nara. An inverse source problem for helmholtz’s equation from the cauchy data with a single wave number.Inverse Problems, 27(10):105001, 2011

    2011

  16. [16]

    Han Gao, Matthew J Zahr, and Jian-Xun Wang. Physics-informed graph neural galerkin networks: A unified framework for solving pde-governed forward and inverse problems.Computer Methods in Applied Mechanics and Engineering, 390:114502, 2022

    2022

  17. [17]

    On the Gibbs phenomenon and its resolution.SIAM Review, 39(4):644–668, 1997

    David Gottlieb and Chi-Wang Shu. On the Gibbs phenomenon and its resolution.SIAM Review, 39(4):644–668, 1997

    1997

  18. [18]

    Reparameterized full-waveform inversion using deep neural net- works.Geophysics, 86(1):V1–V13, 2021

    Qinglong He and Yanfei Wang. Reparameterized full-waveform inversion using deep neural net- works.Geophysics, 86(1):V1–V13, 2021

    2021

  19. [19]

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

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

    1991

  20. [20]

    Ensemble kalman methods for inverse problems.Inverse Problems, 29(4):045001, 2013

    Marco A Iglesias, Kody JH Law, and Andrew M Stuart. Ensemble kalman methods for inverse problems.Inverse Problems, 29(4):045001, 2013

    2013

  21. [21]

    Potential identification via tikhonov-pinns

    Xia Ji, Zihan Jiang, Pengcheng Song, and Cheng Yuan. Potential identification via tikhonov-pinns. Inverse Problems, 41(11):115008, 2025

    2025

  22. [22]

    A gaussian mixture distribution-based adaptive sampling method for physics-informed neural networks.Engineer- ing Applications of Artificial Intelligence, 135:108770, 2024

    Yuling Jiao, Di Li, Xiliang Lu, Jerry Zhijian Yang, and Cheng Yuan. A gaussian mixture distribution-based adaptive sampling method for physics-informed neural networks.Engineer- ing Applications of Artificial Intelligence, 135:108770, 2024

    2024

  23. [23]

    Imaging conductivity from current density magnitude using neural networks.Inverse Problems, 38(7):075003, 2022

    Bangti Jin, Xiyao Li, and Xiliang Lu. Imaging conductivity from current density magnitude using neural networks.Inverse Problems, 38(7):075003, 2022

    2022

  24. [24]

    Conductivity imaging from internal mea- surements with mixed least-squares deep neural networks.SIAM Journal on Imaging Sciences, 17(1):147–187, 2024

    Bangti Jin, Xiyao Li, Qimeng Quan, and Zhi Zhou. Conductivity imaging from internal mea- surements with mixed least-squares deep neural networks.SIAM Journal on Imaging Sciences, 17(1):147–187, 2024

    2024

  25. [25]

    On the ill-conditioned nature of C ∞ RBF strong collocation.Engineering Analysis with Boundary Elements, 78:26–30, 2017

    Edward J Kansa and Pavel Holoborodko. On the ill-conditioned nature of C ∞ RBF strong collocation.Engineering Analysis with Boundary Elements, 78:26–30, 2017

    2017

  26. [26]

    A local meshless collocation method for solving certain inverse problems.Engineering Analysis with Boundary Elements, 57:9–15, 2015

    Wen Li, Xiaoyan Liu, and Guangming Yao. A local meshless collocation method for solving certain inverse problems.Engineering Analysis with Boundary Elements, 57:9–15, 2015

    2015

  27. [27]

    Physics-informed neural networks with hard constraints for inverse design.SIAM Journal on Scientific Computing, 43(6):B1105–B1132, 2021

    Lu Lu, Raphael Pestourie, Wenjie Yao, Zhicheng Wang, Francesc Verdugo, and Steven G Johnson. Physics-informed neural networks with hard constraints for inverse design.SIAM Journal on Scientific Computing, 43(6):B1105–B1132, 2021

    2021

  28. [28]

    Estimates on the generalization error of physics- informed neural networks for approximating pdes.IMA Journal of Numerical Analysis, 43(1):1– 43, 2023

    Siddhartha Mishra and Roberto Molinaro. Estimates on the generalization error of physics- informed neural networks for approximating pdes.IMA Journal of Numerical Analysis, 43(1):1– 43, 2023. 27

    2023

  29. [29]

    Inverse modeling of contaminant transport for pollution source identification in surface and groundwaters: a review.Groundwater for Sustainable Development, 15:100651, 2021

    Maryam Barati Moghaddam, Mehdi Mazaheri, and Jamal Mohammad Vali Samani. Inverse modeling of contaminant transport for pollution source identification in surface and groundwaters: a review.Groundwater for Sustainable Development, 15:100651, 2021

    2021

  30. [30]

    A note on a method for generating points uniformly on n-dimensional spheres

    Mervin E Muller. A note on a method for generating points uniformly on n-dimensional spheres. Communications of the ACM, 2(4):19–20, 1959

    1959

  31. [31]

    Recovering the conductivity from a single measurement of interior data.Inverse Problems, 25(3):035014, 2009

    Adrian Nachman, Alexandru Tamasan, and Alexandre Timonov. Recovering the conductivity from a single measurement of interior data.Inverse Problems, 25(3):035014, 2009

    2009

  32. [32]

    Martin Norgren. Chebyshev collocation and newton-type optimization methods for the inverse problem on nonuniform transmission lines.IEEE transactions on microwave theory and tech- niques, 53(5):1561–1568, 2005

    2005

  33. [33]

    fpinns: Fractional physics-informed neural networks.SIAM Journal on Scientific Computing, 41(4):A2603–A2626, 2019

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

    2019

  34. [34]

    On the spectral bias of neural networks

    Nasim Rahaman, Aristide Baratin, Devansh Arpit, Felix Draxler, Min Lin, Fred Hamprecht, Yoshua Bengio, and Aaron Courville. On the spectral bias of neural networks. InInternational conference on machine learning, pages 5301–5310. PMLR, 2019

    2019

  35. [35]

    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 differential equations.Journal of Computational Physics, 378:686–707, 2019

    2019

  36. [36]

    Nonlinear total variation based noise removal algorithms.Physica D: nonlinear phenomena, 60(1-4):259–268, 1992

    Leonid I Rudin, Stanley Osher, and Emad Fatemi. Nonlinear total variation based noise removal algorithms.Physica D: nonlinear phenomena, 60(1-4):259–268, 1992

    1992

  37. [37]

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

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

    1984

  38. [38]

    Solution of incorrectly formulated problems and the regularization method

    Andrei N Tikhonov. Solution of incorrectly formulated problems and the regularization method. Sov Dok, 4:1035–1038, 1963

    1963

  39. [39]

    Deep image prior

    Dmitry Ulyanov, Andrea Vedaldi, and Victor Lempitsky. Deep image prior. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 9446–9454, 2018

    2018

  40. [40]

    A penalty method for PDE-constrained optimization in inverse problems.Inverse Problems, 32(1):015007, 2016

    Tristan van Leeuwen and Felix J Herrmann. A penalty method for PDE-constrained optimization in inverse problems.Inverse Problems, 32(1):015007, 2016

    2016

  41. [41]

    Various techniques used in connection with random digits.John von Neumann, Collected Works, 5(768-770):1, 1963

    John Von Neumann et al. Various techniques used in connection with random digits.John von Neumann, Collected Works, 5(768-770):1, 1963

    1963

  42. [42]

    Deep collocation method: A framework for solving pdes using neural networks with error control.SIAM Journal on Scientific Computing, 48(1):C77– C102, 2026

    Mingxing Weng, Zhiping Mao, and Jie Shen. Deep collocation method: A framework for solving pdes using neural networks with error control.SIAM Journal on Scientific Computing, 48(1):C77– C102, 2026

    2026

  43. [43]

    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

    2023

  44. [44]

    Wei Xiong, Xiangyun Long, St´ ephane PA Bordas, and Chao Jiang. The deep finite element method: A deep learning framework integrating the physics-informed neural networks with the finite element method.Computer Methods in Applied Mechanics and Engineering, 436:117681, 2025

    2025

  45. [45]

    Randomized greedy algorithms for neural network optimization in solving partial differential equations.Journal of Scientific Computing, 105(1):26, 2025

    Jinchao Xu and Xiaofeng Xu. Randomized greedy algorithms for neural network optimization in solving partial differential equations.Journal of Scientific Computing, 105(1):26, 2025

    2025

  46. [46]

    Frequency Principle: Fourier Analysis Sheds Light on Deep Neural Networks

    Zhi-Qin John Xu, Yaoyu Zhang, Tao Luo, Yanyang Xiao, and Zheng Ma. Frequency principle: Fourier analysis sheds light on deep neural networks.arXiv preprint arXiv:1901.06523, 2019. 28

    work page Pith review arXiv 1901

  47. [47]

    Solving an inverse source problem by deep neural network method with convergence and error analysis.Inverse Problems, 39(7):075013, 2023

    Hui Zhang and Jijun Liu. Solving an inverse source problem by deep neural network method with convergence and error analysis.Inverse Problems, 39(7):075013, 2023

    2023

  48. [48]

    On stability and regularization for data-driven solution of parabolic inverse source problems.Journal of Computational Physics, 474:111769, 2023

    Mengmeng Zhang, Qianxiao Li, and Jijun Liu. On stability and regularization for data-driven solution of parabolic inverse source problems.Journal of Computational Physics, 474:111769, 2023. 29

    2023