Pith. sign in

REVIEW 2 major objections 4 minor 65 references

Treating displacement, strain, and equilibrium residuals as Laplace variables with scales learned from data lets a physics-informed network recover heterogeneous Young's modulus and Poisson's ratio maps from noisy displacement observations

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 ↗

A probabilistic PINN with adaptive Laplace-scale weighting and a B-spline displacement prior estimates heterogeneous Young's modulus and Poisson's ratio from noisy low-resolution displacement data more robustly than IE-PINN and other baselines.

T0 review reviewed 2026-08-02 challenge →

load-bearing objection PIE-PINN is a credible engineering extension for inverse elasticity from noisy low-resolution displacement data, but the absolute Young's modulus headline depends on an unstated reference mean-modulus Ec and the evidence is figures-only. the 2 major comments →

arxiv 2607.14563 v1 pith:PMHI4WCJ submitted 2026-07-16 cs.LG stat.ML

Probabilistic Physics-Informed Neural Networks for Estimating Heterogeneous Elastic Properties from Low-Resolution and Noisy Displacement Data

Tatthapong Srikitrungruang , Jaesung Lee This is my paper
classification cs.LG stat.ML
keywords inverse elasticityphysics-informed neural networksheterogeneous elastic propertiesYoung's modulusPoisson's ratiolow-resolution displacement dataadaptive loss weightinghorseshoe+ prior
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 reading

Each residual in the inverse elasticity loss—displacement fitting, strain consistency, and equilibrium force—is modeled as a Laplace-distributed variable, and the scale of each residual is estimated during training instead of being set by hand. For displacement errors the scales are pointwise and are wrapped in a horseshoe+ hierarchical prior, a global–local shrinkage model that lets a few large residuals stand out while suppressing small ones; the predicted displacement field is a B-spline global surface plus a neural-network correction. The paper shows that on finite-element-simulated benchmarks this scheme recovers the spatial patterns of Young's modulus and Poisson's ratio from observations downsampled to 12.5% of the grid and corrupted to SNR 20, where IE-PINN, Self-Adaptive IE-PINN, ElastNet, EI-UNet, and gPINN degrade sharply. If correct, it means elastography-style material property maps can be reconstructed from sparse, noisy measurements without manual tuning of loss weights.

Core claim

The central claim is that an inverse elasticity PINN can remain accurate under degraded observations only if the residual scales are estimated from data rather than prescribed or maximized. PIE-PINN therefore treats the observed-displacement residual, the strain-discrepancy residual, and the equilibrium-force residual as Laplace random variables: the displacement scales are pointwise with a horseshoe+ hierarchical half-Cauchy prior that shrinks small errors while permitting large ones, and the displacement mean is represented as a B-spline surface plus a neural-network correction. An alternating maximum-likelihood procedure updates the mean functions by weighted residual minimization (E-step

What carries the argument

The load-bearing object is the joint negative log-likelihood built from three Laplace residual models—Laplace here meaning a heavy-tailed error distribution that tolerates outliers better than a Gaussian—for displacement (pointwise scales bu,i), strain discrepancy (global scales bε), and equilibrium force (global scales bf), plus a deterministic mean-modulus constraint. The displacement term is wrapped in a horseshoe+ hierarchy, a global–local shrinkage prior on the pointwise Laplace scales that prevents low-resolution underfitting by allowing a few large residuals while pulling most scales toward zero. The displacement network is hybrid: a tensor-product B-spline represents the smooth globa

Load-bearing premise

The load-bearing premise is that the prescribed reference mean modulus Ec is available or correctly chosen: since the equilibrium equations are unchanged when Young's modulus is multiplied by a constant, displacement data alone identify only relative modulus, and the paper does not state how Ec is obtained in practice.

What would settle it

Run PIE-PINN on the same low-resolution noisy displacement dataset twice, with the mean-modulus constraint Ec set to the true mean and to 1.1 times the true mean. If the identifiability argument holds, the estimated Young's modulus maps should differ by the same 10% global factor while displacement, strain, and equilibrium residuals remain essentially unchanged; the reported absolute MAE would then be shown to depend on benchmark construction rather than on information contained in the displacement data.

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

If this is right

  • Young's modulus and Poisson's ratio maps can be recovered from displacement data downsampled to 12.5% of the original grid and corrupted to SNR 20, a regime in which the paper shows IE-PINN, ElastNet, EI-UNet, gPINN, and Self-Adaptive IE-PINN degrade sharply.
  • The same training procedure yields denoised, high-resolution displacement, strain, and stress fields as by-products, because fitting operates on the latent B-spline-plus-network mean rather than on the raw noisy observations.
  • Loss balancing between data fidelity and physics consistency becomes an automated maximum-likelihood task: the learned inverse scale 1/b for each residual is the loss weight, removing manual weight selection from the pipeline.
  • The reported accuracy holds across several distinct spatial elasticity distributions, suggesting the mechanism transfers beyond any single benchmark pattern.

Where Pith is reading between the lines

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

  • Editorial inference: reported absolute errors for Young's modulus depend on the prescribed reference mean modulus Ec in Eq. (20); because the equilibrium equation is invariant under uniform rescaling of E, displacement data alone identify only relative modulus. Re-running the benchmark with Ec offset by 10% should scale the estimated modulus map by the same factor while leaving residuals unchanged
  • Editorial inference: after convergence, the learned Laplace scales and horseshoe+ shrinkage factors encode per-point residual uncertainty, which could seed approximate confidence intervals on the recovered property maps; the paper leaves full uncertainty quantification to future work.
  • Editorial inference: the split into a linear B-spline global field and a nonlinear local correction suggests that, for a fixed spline basis, the global component could be updated very cheaply, potentially enabling near-real-time elastography if the local correction is retrained on small patches; the paper does not explore this.
  • Editorial inference: the contrast between minimax self-adaptive weighting, which amplifies noisy residuals, and this scale-based downweighting suggests a general lesson for PINNs with noisy data: weights should shrink, not amplify, residuals that are dominated by measurement noise.
Share X Bluesky LinkedIn Reddit HN

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 / 4 minor

Summary. The paper proposes PIE-PINN, a probabilistic physics-informed neural network for estimating heterogeneous Young's modulus and Poisson's ratio from noisy, low-resolution displacement observations. Displacement fidelity, strain-discrepancy, and equilibrium residuals are modeled with Laplace distributions; the displacement field is represented by a B-spline-guided neural network; and a hierarchical half-Cauchy prior is placed on pointwise displacement Laplace scales. Training alternates between updating mean-function parameters (weighted residual minimization) and scale parameters (adaptive loss weights). The method is evaluated on the ElastNet dragon/dog benchmark at various resolutions and SNRs, with comparisons to IE-PINN, ElastNet, EI-UNet, gPINN, and Self-Adaptive IE-PINN, plus ablations of the three main components. The central claim is that PIE-PINN consistently outperforms existing methods across noise levels and observation resolutions.

Significance. If the empirical claim holds, the method would be a practically useful contribution to inverse elasticity estimation from clinically realistic observations: the probabilistic formulation provides a principled route to adaptive loss weighting, the B-spline-plus-neural-network representation is well motivated for low-resolution data, and the EM-style training is an elegant way to stabilize joint estimation. The paper also includes multiple datasets and ablation studies. However, the evidence is currently mostly qualitative and the absolute Young's modulus results depend on an unstated reference mean modulus, Ec. As presented, the contribution is best understood as a method for recovering relative modulus and Poisson's ratio, with absolute-scale recovery requiring an additional calibration step that is not described.

major comments (2)
  1. [Section 5, Eq. (20)] The absolute Young's modulus results depend on the prescribed reference mean modulus Ec. The equilibrium equation (3) is invariant under uniform scaling of E, so displacement observations alone cannot determine the absolute scale of E; only the relative field E/Ē is identifiable. The manuscript inherits the mean-modulus constraint LE from IE-PINN but, unlike the Phase 2 calibration described for IE-PINN in Section 2.2, it never states how Ec is obtained in practice or in the benchmark experiments. If Ec is set to the true mean of the ElastNet ground-truth field, the reported MAE against absolute E is not a self-contained estimate. Please provide a practical rule for selecting Ec (e.g., from boundary traction data) or explicitly reframe the results as relative-modulus recovery, and report the Ec values used in every experiment.
  2. [Section 5, Figures 4–14] The central claim of consistent outperformance is supported only by qualitative field plots. No numerical MAE values, confidence intervals, or repeated-run statistics are reported anywhere in the main text or supplementary material. For example, Figure 7 shows error comparisons across patterns without axis values, and Figures 8–14 show estimated fields but no quantitative metrics. To substantiate 'consistently outperforms existing methods,' please include tables of MAE (mean ± std over multiple random seeds) for all methods and all tested SNRs/resolutions, including baselines. The current evidence is insufficient for a benchmark claim.
minor comments (4)
  1. [Eqs. (12), (17); Sections 3, 5.4] The hierarchical prior is repeatedly called 'horseshoe+' but Eq. (12) specifies the standard horseshoe prior: b ~ C+(0, ητ/√2), η ~ C+(0,1), τ ~ C+(0,1), with no additional local shrinkage parameter. Horseshoe+ would require a second local parameter. Please correct the terminology or extend the model.
  2. [Table S1] The activation function is reported as 'SIREN' with citation to Vaswani et al. (2017), which is the Attention paper; the correct SIREN reference should be cited. Also, 'linearity' should be 'activation function'.
  3. [Algorithm 1 / Eq. (23)] The M-step is stated as 'min_Φ L_total(Φ|u, Θ)', but the mathematical description in Eq. (23) minimizes only C(Φ). Please clarify how L_total(Φ|u,Θ) is defined, or restate the M-step as minimizing C(Φ).
  4. [Section 5 / Supplementary Table S1] The description 'number of basis functions dimension of observation' is unclear; please specify the B-spline knot count and grid size. Also clarify what 'all probabilistic loss components are initially assigned weights equal to 1.0' means in terms of the initial scale parameters b_u, b_ε, b_f.

Circularity Check

1 steps flagged

No significant circularity in the main derivation; minor scale-level circularity in absolute Young's modulus via the prescribed Ec constraint.

specific steps
  1. other [Section 4.1, Eq. (20); Section 5.1 MAE definition]
    "the mean-modulus constraint is not treated as a random residual. It is retained as a deterministic constraint term, LE = λE | ¯E − Ec | ... The primary role of this loss term is to constrain the average predicted Young's modulus to a prescribed value, thereby alleviating the ill-posedness of the inverse problem and preventing the predicted modulus field from collapsing to the trivial zero solution."

    The equilibrium equation (3) is invariant under uniform rescaling of E, so displacement data fix E only up to a global multiplicative constant. Eq. (20) drives the predicted mean Ebar toward an externally supplied Ec. In the benchmarks, MAE is measured against the known ground-truth E (Section 5.1), and the paper gives no rule for obtaining Ec (e.g., the boundary-condition calibration described for IE-PINN in Section 2.2). Thus the absolute scale of the reported Young's modulus estimate is injected by an input rather than inferred from data. The spatial heterogeneity and Poisson's ratio are still learned independently, so this circularity is confined to the global scale.

full rationale

The core PIE-PINN derivation, from Laplace residuals to the alternating maximum-likelihood E/M-step, is self-contained: the displacement, strain, and equilibrium residuals are all expressed as probabilistic models, and the inverse-scale weights arise from maximizing the likelihood, not from fitting the target output. The B-spline and horseshoe+ components are architectural/prior additions that are ablated independently. Comparisons are made against independent FEM-generated ground truth from the ElastNet dataset, so the central robustness claim does not reduce to a fitted target. The main caveat is the mean-modulus constraint in Eq. (20): because the equilibrium PDE is scale-invariant in E, the absolute level of Young's modulus is fixed by the user-supplied Ec rather than learned. The paper does not disclose how Ec is set in the experiments and lacks a calibration phase for PIE-PINN analogous to IE-PINN's Phase 2, so the reported absolute-MAE numbers are not fully self-contained. The self-citations to the authors' IE-PINN are used as baseline and design inheritance, but they are not load-bearing for the main result, which is independently benchmarked. Overall, this is a minor, scale-level circularity rather than a collapse of the derivation.

Axiom & Free-Parameter Ledger

5 free parameters · 5 axioms · 0 invented entities

The method introduces no new physical entities. The main external inputs are the linear-elasticity forward model, the prescribed mean modulus Ec, and a set of hand-chosen architecture/optimization hyperparameters. The adaptive scales are learned from residuals but their shrinkage properties are asserted rather than analyzed.

free parameters (5)
  • Reference mean Young's modulus Ec = not stated; presumably true mean from dataset
    Used in Eq. (20) mean-modulus constraint to anchor absolute modulus scale; without known Ec only relative modulus is identifiable.
  • Mean-modulus constraint weight λE = 0.1 (Table S1)
    Hand-specified; paper claims insensitivity based on prior IE-PINN work but includes no sensitivity study in this paper.
  • Network architecture hyperparameters = 16x128 SIREN layers; B-spline degree 2; learning rate 1e-3
    Chosen without a systematic tuning study; could affect the reported robustness results.
  • Initial probabilistic loss weights = 1.0 for all probabilistic components
    All probabilistic loss components initially assigned weight 1.0; adaptive scales later alter them.
  • Laplace scale parameters b_u,i, b_ε,k, b_f,j = estimated during M-step
    These set the adaptive loss weights and are learned from residuals under Eqs. (16)-(19), not externally known.
axioms (5)
  • domain assumption Plane-stress, small-strain, isotropic linear elasticity equations (Eqs. 1-3) are the correct forward model
    Used to define strain and equilibrium residuals; real tissue and materials may be nonlinear, anisotropic, or three-dimensional.
  • ad hoc to paper The mean Young's modulus Ec is known or prescribable
    Eq. (20) requires Ec to prevent collapse and to set absolute scale; inherited from IE-PINN and not derived from data.
  • domain assumption Residuals follow Laplace distributions with scales b; strains and stresses can be treated as latent random variables
    Equations (10)-(14); a modeling choice for robustness, not derived from measurement physics.
  • ad hoc to paper The horseshoe-style half-Cauchy hierarchy in Eq. (12) induces the claimed sparsity on residual scales
    The prior structure is imposed to improve robustness; the paper provides no theoretical analysis of its shrinkage behavior in this inverse setting.
  • domain assumption FEM-generated ElastNet displacement data adequately represent noisy low-resolution observations
    All experiments use synthetic downsampled, noisy finite-element displacement; no experimental or clinical data validation.

reviewed 2026-08-02 · how reviews work

0 comments
Cite this review

Pith. "Pith review of Probabilistic Physics-Informed Neural Networks for Estimating Heterogeneous Elastic Properties from Low-Resolution and Noisy Displacement Data." pith.science (2026). https://pith.science/paper/PMHI4WCJ

@misc{pith2026260714563,
  author       = {Pith},
  title        = {Pith review of: Probabilistic Physics-Informed Neural Networks for Estimating Heterogeneous Elastic Properties from Low-Resolution and Noisy Displacement Data},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/PMHI4WCJ}},
  note         = {Machine review of arXiv:2607.14563}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Estimating spatially heterogeneous elastic properties from low-resolution displacement measurements is a severely ill-posed inverse elasticity problem because low resolution obscures spatial details needed to distinguish heterogeneous property variations, and small measurement perturbations or fitting errors are amplified through inverse estimation. Existing inverse methods often rely on high-fidelity observations and manually prespecified loss weights, limiting their adaptability and making them sensitive to noise and resolution degradation. We propose a Probabilistic Inverse Elasticity Physics-Informed Neural Network (PIE-PINN) framework for robust estimation of Young's modulus and Poisson's ratio from noisy, low-resolution displacement data. PIE-PINN models displacement observation, strain-discrepancy, and equilibrium residuals using Laplace distributions within a unified probabilistic model. To improve robustness, the framework combines a B-spline-guided displacement network with a hierarchical half-Cauchy model for displacement residual scales. The B-spline provides a smooth global representation of the displacement field, while the neural network correction captures local variations. The hierarchical scale model adaptively downweights severe displacement fitting errors, enabling more robust recovery of the latent mean displacement field. An alternating maximum-likelihood training strategy updates the mean through weighted residual minimization and updates the scales to adjust the loss weights. Systematic case studies across varying noise levels and observation resolutions demonstrate the robustness of PIE-PINN.

Figures

Figures reproduced from arXiv: 2607.14563 by Jaesung Lee, Tatthapong Srikitrungruang.

Figure 1
Figure 1. Figure 1: Governing relationships of the linear elasticity model, including strain-displacement, constitutive, and [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Conceptual flow of the two-phase strategy used in the IE-PINN framework for estimating spatially heteroge [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Framework Probabilistic Physics-Informed Neural Networks (PIE-PINN) for heterogeneous elasticity [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Comparison of Young’s modulus (E) estimates from different models inferred from noisy, low-resolution displacement data (50% resolution, SNR = 100) [PITH_FULL_IMAGE:figures/full_fig_p014_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Comparison of Poisson’s ratio (ν) estimates from different models inferred from noisy, low-resolution displacement data (50% resolution, SNR = 100). 5.2 Robustness to noise & resolution The presence of noise is known to degrade estimation accuracy, as demonstrated in previous studies (Srikitrungruang et al. 2025), and low spatial resolution further exacerbates the difficulty of inverse elasticity estimatio… view at source ↗
Figure 6
Figure 6. Figure 6: Related mechanical fields in the proposed PIE-PINN under noisy, low-resolution displacement data (50% [PITH_FULL_IMAGE:figures/full_fig_p015_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Elasticity estimation errors across multiple elastic distribution patterns obtained from noisy, low-resolution [PITH_FULL_IMAGE:figures/full_fig_p016_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Elasticity maps estimated by the proposed PIE-PINN model using 50% low-resolution displacement data [PITH_FULL_IMAGE:figures/full_fig_p016_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Elasticity maps estimated by the proposed PIE-PINN under noisy, low-resolution displacement data at SNR = [PITH_FULL_IMAGE:figures/full_fig_p017_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Comparison of elasticity maps inferred from noisy, low-resolution displacement data (25% resolution, SNR [PITH_FULL_IMAGE:figures/full_fig_p017_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Comparison of elasticity maps inferred from noisy, low-resolution displacement data (50% resolution, SNR [PITH_FULL_IMAGE:figures/full_fig_p017_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Evolution of Young’s modulus estimates inferred from noisy, low-resolution displacement data (50% spatial [PITH_FULL_IMAGE:figures/full_fig_p018_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Evolution of Poisson’s ratio estimates inferred from noisy, low-resolution displacement data (50% spatial [PITH_FULL_IMAGE:figures/full_fig_p018_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Comparison of elasticity maps inferred from noisy, low-resolution displacement data (50% spatial resolution, [PITH_FULL_IMAGE:figures/full_fig_p019_14.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

65 extracted references

  1. [1]

    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 Engineering421, 116805

    2024

  2. [2]

    Arda, K., Ciledag, N., Aktas, E., Arıbas, B. K. & Köse, K. (2011), ‘Quantitative assessment of normal soft-tissue elasticity using shear-wave ultrasound elastography’,American Journal of Roentgenology197(3), 532–536

    2011

  3. [3]

    A., Oberai, A

    Babaniyi, O. A., Oberai, A. A. & Barbone, P. E. (2017), ‘Direct error in constitutive equation formulation for plane stress inverse elasticity problem’,Computer Methods in Applied Mechanics and Engineering314, 3–18

    2017

  4. [4]

    & Smołka, M

    Barabasz, B., Gajda-Zagórska, E., Migórski, S., Paszy´nski, M., Schaefer, R. & Smołka, M. (2014), ‘A hybrid algorithm for solving inverse problems in elasticity’,International Journal of Applied Mathematics and Computer Science 24(4), 865–886

    2014

  5. [5]

    Barbone, P. E. & Oberai, A. A. (2007), ‘Elastic modulus imaging: some exact solutions of the compressible elastography inverse problem’,Physics in Medicine and Biology52(6), 1577–1593

    2007

  6. [6]

    & Sisman, Y

    Bektas, S. & Sisman, Y . (2010), ‘The comparison of l1 and l2-norm minimization methods’,International Journal of the Physical Sciences5, 1721–1727

    2010

  7. [7]

    S., FitzGerald, E

    Berko, N. S., FitzGerald, E. F., Amaral, T. D., Payares, M. & Levin, T. L. (2013), ‘Ultrasound elastography in children: Establishing the normal range of muscle elasticity’,Pediatric Radiology44(2), 158–163

    2013

  8. [8]

    Boutouyrie, P., Chowienczyk, P., Humphrey, J. D. & Mitchell, G. F. (2021), ‘Arterial stiffness and cardiovascular risk in hypertension’,Circulation research128(7), 864–886

    2021

  9. [9]

    Buntin, C. M. & Silver, F. H. (1990), ‘Noninvasive assessment of mechanical properties of peripheral arteries’,Annals of Biomedical Engineering18(5), 549–566

    1990

  10. [10]

    Chen, C. & Gu, G. X. (2023), ‘Physics-informed deep-learning for elasticity: Forward, inverse, and mixed problems’, Advanced Science10(18)

    2023

  11. [11]

    Chen, C.-T. & Gu, G. X. (2021), ‘Learning hidden elasticity with deep neural networks’,Proceedings of the National Academy of Sciences118(31)

    2021

  12. [12]

    & Stinis, P

    Chen, W., Howard, A. & Stinis, P. (2026), ‘Self-adaptive weighting and sampling for physics-informed neural networks’, Machine Learning: Science and Technology7(2), 025050

    2026

  13. [13]

    A., Heinrich, S

    Choren, J. A., Heinrich, S. M. & Silver-Thorn, M. B. (2013), ‘Young’s modulus and volume porosity relationships for additive manufacturing applications’,Journal of Materials Science48(15), 5103–5112

    2013

  14. [14]

    T., Dufera, T

    Deresse, A. T., Dufera, T. T. & Firdi, M. D. (2025), ‘Leveraging enhanced physics-informed neural networks based on adaptive weight loss for solving inverse problems of nonlinear sine-gordon equation’,The European Physical Journal Plus140(12)

    2025

  15. [15]

    M., Meaney, P

    Doyley, M. M., Meaney, P. M. & Bamber, J. C. (2000), ‘Evaluation of an iterative reconstruction method for quantitative elastography’,Physics in Medicine and Biology45(6), 1521–1540

    2000

  16. [16]

    & De Lorenzis, L

    Flaschel, M., Kumar, S. & De Lorenzis, L. (2021), ‘Unsupervised discovery of interpretable hyperelastic constitutive laws’,Computer Methods in Applied Mechanics and Engineering381, 113852

    2021

  17. [17]

    Gao, B., Yao, R. & Li, Y . (2025), ‘Physics-informed neural networks with adaptive loss weighting algorithm for solving partial differential equations’,Computers & Mathematics with Applications181, 216–227

    2025

  18. [18]

    Gao, H., Zahr, M. J. & Wang, J.-X. (2022), ‘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

  19. [19]

    E., Jiang, J., Hall, T

    Goenezen, S., Dord, J.-F., Sink, Z., Barbone, P. E., Jiang, J., Hall, T. J. & Oberai, A. A. (2012), ‘Linear and nonlinear elastic modulus imaging: An application to breast cancer diagnosis’,IEEE Transactions on Medical Imaging 31(8), 1628–1637

    2012

  20. [20]

    & Courville, A

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

    2016

  21. [21]

    & Chand, L

    Gupta, M., Taneja, H. & Chand, L. (2018), ‘Performance enhancement and analysis of filters in ultrasound image denoising’,Procedia Computer Science132, 643–652

    2018

  22. [22]

    & Juanes, R

    Haghighat, E., Raissi, M., Moure, A., Gomez, H. & Juanes, R. (2021), ‘A physics-informed deep learning framework for inversion and surrogate modeling in solid mechanics’,Computer Methods in Applied Mechanics and Engineering 379, 113741

    2021

  23. [23]

    M., Long, K

    Hamel, C. M., Long, K. N. & Kramer, S. L. B. (2022), ‘Calibrating constitutive models with full-field data via physics informed neural networks’,Strain59(2). 19 Probabilistic Physics-Informed Neural Networks (PIE-PINN)A PREPRINT

    2022

  24. [24]

    Kabanikhin, S. I. (2008), ‘Definitions and examples of inverse and ill-posed problems’,Journal of Inverse and Ill-posed Problems16(4)

    2008

  25. [25]

    & Laksari, K

    Kamali, A. & Laksari, K. (2024), ‘Physics-informed unets for discovering hidden elasticity in heterogeneous materials’, Journal of the Mechanical Behavior of Biomedical Materials150, 106228

    2024

  26. [26]

    & Laksari, K

    Kamali, A., Sarabian, M. & Laksari, K. (2023), ‘Elasticity imaging using physics-informed neural networks: Spatial discovery of elastic modulus and poisson’s ratio’,Acta Biomaterialia155, 400–409

    2023

  27. [27]

    & V osburgh, K

    Krissian, K., Kikinis, R., Westin, C.-F. & V osburgh, K. (2005), Speckle-constrained filtering of ultrasound images,in ‘2005 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR’05)’, V ol. 2, IEEE, p. 547–552

    2005

  28. [28]

    Lang, S. B. (1970), ‘Ultrasonic method for measuring elastic coefficients of bone and results on fresh and dried bovine bones’,IEEE Transactions on Biomedical EngineeringBME-17(2), 101–105

    1970

  29. [29]

    & Adjouadi, M

    Mafi, M., Tabarestani, S., Cabrerizo, M., Barreto, A. & Adjouadi, M. (2018), ‘Denoising of ultrasound images affected by combined speckle and gaussian noise’,IET Image Processing12(12), 2346–2351

    2018

  30. [30]

    V ., Ehman, R

    Manduca, A., Bayly, P. V ., Ehman, R. L., Kolipaka, A., Royston, T. J., Sack, I., Sinkus, R. & Van Beers, B. E. (2021), ‘Mr elastography: Principles, guidelines, and terminology’,Magnetic resonance in medicine85(5), 2377–2390

    2021

  31. [31]

    McClenny, L. D. & Braga-Neto, U. M. (2023), ‘Self-adaptive physics-informed neural networks’,Journal of Computa- tional Physics474, 111722

    2023

  32. [32]

    (1996), ‘The expectation-maximization algorithm’,IEEE Signal Processing Magazine13(6), 47–60

    Moon, T. (1996), ‘The expectation-maximization algorithm’,IEEE Signal Processing Magazine13(6), 47–60

    1996

  33. [33]

    & Davies, H

    Mouloodi, S., Rahmanpanah, H., Burvill, C. & Davies, H. M. (2020), ‘Prediction of load in a long bone using an artificial neural network prediction algorithm’,journal of the mechanical behavior of biomedical materials102, 103527

    2020

  34. [34]

    F., Manduca, A

    Muthupillai, R., Lomas, D., Rossman, P., Greenleaf, J. F., Manduca, A. & Ehman, R. L. (1995), ‘Magnetic resonance elastography by direct visualization of propagating acoustic strain waves’,science269(5232), 1854–1857

    1995

  35. [35]

    Nguyen, L. T. P. & Liu, B. H. (2022), ‘Machine learning framework for determination of elastic modulus without contact model fitting’,International Journal of Solids and Structures256, 111976

    2022

  36. [36]

    & Gao, H

    Ni, B. & Gao, H. (2021), ‘A deep learning approach to the inverse problem of modulus identification in elasticity’,MRS Bulletin46(1), 19–25

    2021

  37. [37]

    & Pharr, G

    Oliver, W. & Pharr, G. (1992), ‘An improved technique for determining hardness and elastic modulus using load and displacement sensing indentation experiments’,Journal of Materials Research7(6), 1564–1583

    1992

  38. [38]

    & Pharr, G

    Oliver, W. & Pharr, G. (2004), ‘Measurement of hardness and elastic modulus by instrumented indentation: Advances in understanding and refinements to methodology’,Journal of Materials Research19(1), 3–20

    2004

  39. [39]

    & Kim, K

    Quan, F.-S. & Kim, K. S. (2016), ‘Medical applications of the intrinsic mechanical properties of single cells’,Acta Biochimica et Biophysica Sinica48(10), 865–871

    2016

  40. [40]

    & Batmanghelich, K

    Ragoza, M. & Batmanghelich, K. (2023), Physics-informed neural networks for tissue elasticity reconstruction in magnetic resonance elastography,in‘Medical Image Computing and Computer Assisted Intervention – MICCAI 2023’, Springer Nature Switzerland, p. 333–343

    2023

  41. [41]

    & Karniadakis, G

    Raissi, M., Perdikaris, P. & Karniadakis, G. (2019), ‘Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations’,Journal of Computational Physics378, 686–707

    2019

  42. [42]

    & Karniadakis, G

    Raissi, M., Yazdani, A. & Karniadakis, G. E. (2020), ‘Hidden fluid mechanics: Learning velocity and pressure fields from flow visualizations’,Science367(6481), 1026–1030

    2020

  43. [43]

    & Balasubramanian, G

    Roy, A., Babuska, T., Krick, B. & Balasubramanian, G. (2020), ‘Machine learned feature identification for predicting phase and young’s modulus of low-, medium- and high-entropy alloys’,Scripta Materialia185, 152–158

    2020

  44. [44]

    & Popp, A

    Sahin, T., von Danwitz, M. & Popp, A. (2024), ‘Solving forward and inverse problems of contact mechanics using physics-informed neural networks’,Advanced Modeling and Simulation in Engineering Sciences11(1)

    2024

  45. [45]

    D., Blackshire, J

    Shukla, K., Jagtap, A. D., Blackshire, J. L., Sparkman, D. & Em Karniadakis, G. (2022), ‘A physics-informed neural network for quantifying the microstructural properties of polycrystalline nickel using ultrasound data: A promising approach for solving inverse problems’,IEEE Signal Processing Magazine39(1), 68–77

    2022

  46. [46]

    & Liu, D

    Smyl, D., Bossuyt, S. & Liu, D. (2019), ‘Openqsei: A matlab package for quasi static elasticity imaging’,SoftwareX 9, 73–76

    2019

  47. [47]

    & Alkhalifah, T

    Song, C. & Alkhalifah, T. A. (2022), ‘Wavefield reconstruction inversion via physics-informed neural networks’,IEEE Transactions on Geoscience and Remote Sensing60, 1–12. 20 Probabilistic Physics-Informed Neural Networks (PIE-PINN)A PREPRINT

    2022

  48. [48]

    Song, Y ., Wang, H., Yang, H., Taccari, M. L. & Chen, X. (2024), ‘Loss-attentional physics-informed neural networks’, Journal of Computational Physics501, 112781

    2024

  49. [49]

    & Zhou, Y

    Srikitrungruang, T., Aghaee Dabaghan Fard, S., Lemon, M., Lee, J. & Zhou, Y . (2025), ‘Robust physics-informed neural network approach for estimating heterogeneous elastic properties from noisy displacement data’,Advanced Science12(48)

    2025

  50. [50]

    (2012), ‘A review of numerical methods for nonlinear partial differential equations’,Bulletin of the American Mathematical Society49(4), 507–554

    Tadmor, E. (2012), ‘A review of numerical methods for nonlinear partial differential equations’,Bulletin of the American Mathematical Society49(4), 507–554

    2012

  51. [51]

    N., Kaiser, L

    Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A. N., Kaiser, L. & Polosukhin, I. (2017), ‘Attention is all you need’

    2017

  52. [52]

    Wang, J., Zhou, B., Parkinson, I., Thomas, C. D. L., Clement, J. G., Fazzalari, N. & Guo, X. E. (2013), ‘Trabecular plate loss and deteriorating elastic modulus of femoral trabecular bone in intertrochanteric hip fractures’,Bone research 1(1), 346–354

    2013

  53. [53]

    & Perdikaris, P

    Wang, S., Teng, Y . & Perdikaris, P. (2021), ‘Understanding and mitigating gradient flow pathologies in physics-informed neural networks’,SIAM Journal on Scientific Computing43(5), A3055–A3081

    2021

  54. [54]

    & Perdikaris, P

    Wang, S., Yu, X. & Perdikaris, P. (2022), ‘When and why pinns fail to train: A neural tangent kernel perspective’, Journal of Computational Physics449, 110768

    2022

  55. [55]

    Wang, X., Yi, S., Gu, H., Xu, J. & Xu, W. (2025), ‘Wf-pinns: solving forward and inverse problems of burgers equation with steep gradients using weak-form physics-informed neural networks’,Scientific Reports15(1)

    2025

  56. [56]

    & Dong, F

    Wu, H., Liang, W., Jiao, Y ., Song, H., Peng, Q., Luo, H., Zhang, Y ., Xu, J. & Dong, F. (2019), ‘A preliminary comparative study of young’s modulus versus shear modulus in the diagnosis of breast cancer’,Ultrasound Quarterly 35(1), 88–92

    2019

  57. [57]

    & Yao, W

    Xiang, Z., Peng, W., Liu, X. & Yao, W. (2022), ‘Self-adaptive loss balanced physics-informed neural networks’, Neurocomputing496, 11–34

    2022

  58. [58]

    & Cook, R

    Xu, D., Harvey, T., Begiristain, E., Domínguez, C., Sánchez-Abella, L., Browne, M. & Cook, R. B. (2022), ‘Measuring the elastic modulus of soft biomaterials using nanoindentation’,Journal of the Mechanical Behavior of Biomedical Materials133, 105329

    2022

  59. [59]

    K., Natarajan, S

    Yadav, G. K., Natarajan, S. & Srinivasan, B. (2022), ‘Distributed pinn for linear elasticity — a unified approach for smooth, singular, compressible and incompressible media’,International Journal of Computational Methods19(08)

    2022

  60. [60]

    & Zhang, H

    Yan, H., Li, Q., Yang, B., Yang, Y ., Wang, Y . & Zhang, H. (2024), ‘Modeling and prediction method for young’s moduli of ti alloys based on residual muti-layer perceptron’,JOM77(1), 76–90

    2024

  61. [61]

    & Karniadakis, G

    Yang, L., Meng, X. & Karniadakis, G. E. (2021), ‘B-pinns: Bayesian physics-informed neural networks for forward and inverse pde problems with noisy data’,Journal of Computational Physics425, 109913

    2021

  62. [62]

    & Hwang, S

    Yoon, J. & Hwang, S. J. (2017), Combined group and exclusive sparsity for deep neural networks,inD. Precup & Y . W. Teh, eds, ‘Proceedings of the 34th International Conference on Machine Learning’, V ol. 70 ofProceedings of Machine Learning Research, PMLR, pp. 3958–3966

    2017

  63. [63]

    & Karniadakis, G

    Yu, J., Lu, L., Meng, X. & Karniadakis, G. E. (2022), ‘Gradient-enhanced physics-informed neural networks for forward and inverse pde problems’,Computer Methods in Applied Mechanics and Engineering393, 114823

    2022

  64. [64]

    & Vinyals, O

    Zhang, C., Bengio, S., Hardt, M., Recht, B. & Vinyals, O. (2016), ‘Understanding deep learning requires rethinking generalization’

    2016

  65. [65]

    −1 0 1 −1 0 1 −1 0 1 # ,w yy =

    Zhou, Y . & Du, J. (2022), ‘Atomic force microscopy (afm) and its applications to bone-related research’,Progress in Biophysics and Molecular Biology176, 52–66. 21 Probabilistic Physics-Informed Neural Networks (PIE-PINN)A PREPRINT SUPPLEMENTARY MA TERIAL Supplementary Notes Supplementary Note S1: Finite difference The finite difference method serves as a...

    2022

This paper was first reviewed by deepseek-v4-flash on August 2, 2026.