Pith. sign in

REVIEW 4 major objections 5 minor 300 references

A pair of neural operators, PANO and CANO, learns to map full-field displacement data and reaction forces directly to a hyperelastic material's strain-energy density function, returning the model in a single forward pass without solving an

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

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

T0 review · deepseek-v4-flash

2026-08-02 00:17 UTC pith:FSRJEZQI

load-bearing objection A useful, mostly credible operator-learning framework for parametric hyperelastic identification, but the 'discovery' claim is only demonstrated within the six-parameter separable cubic family that the CANO trunk encodes by construction. the 4 major comments →

arxiv 2607.15049 v1 pith:FSRJEZQI submitted 2026-07-16 cs.CE cond-mat.mtrl-sci

Neural operators solve inverse problems for constitutive model discovery

Moritz Flaschel , Burigede Liu , Ellen Kuhl This is my paper
classification cs.CE cond-mat.mtrl-sci
keywords neural operatorsconstitutive model discoveryhyperelasticitystrain-energy densityinverse problemsphysics-augmented neural networksLaplacian eigenfunction encodingmaterial characterization
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper proposes two neural operator architectures, PANO and CANO, that learn a map from measurable quantities—surface displacement fields and net reaction forces—to the infinite-dimensional space of admissible hyperelastic strain-energy density functions. In contrast to standard material characterization, which solves an optimization problem each time, a trained operator returns the material model in one forward pass. The output is constrained to be physically admissible (objective, stress-free, polyconvex, etc.) by construction. The authors demonstrate robustness to noise, missing data, different discretizations, and specimen size, and present a numerical identifiability analysis showing that within the six-parameter cubic family of strain-energy functions the standardized experiment yields no non-identifiable parameter directions. The load-bearing claim is that the operator captures the physics-to-data map accurately enough for near-instantaneous material characterization without retraining.

Core claim

The central claim is that the inverse problem of constitutive model discovery can be reformulated as supervised operator learning: a neural operator approximates the conditional inverse operator that maps an observation tuple—surface displacements and net reaction force—to the physically admissible strain-energy density function. The operator is trained on thousands of simulated experiments for strain-energy functions sampled from a six-parameter cubic family. Its branch net encodes the displacement field through Laplacian eigenfunctions, which makes the prediction independent of the measurement grid and robust to noise, while its trunk net outputs latent features designed to enforce objecti

What carries the argument

The load-bearing construction is the branch-and-trunk operator architecture. The branch net takes the flattened coefficients of the displacement field's Laplacian-eigenfunction expansion together with the normalized reaction-force history, and outputs a non-negative latent vector. The trunk net maps the isochoric invariants to a latent feature vector: in PANO these features are learned convex monotone functions (input-convex networks); in CANO they are the six fixed monomials of a cubic generalized Mooney–Rivlin expansion. The strain-energy density is the inner product of the two latent vectors scaled by the reaction-force norm. The Laplacian-eigenfunction projection in the branch's first la

Load-bearing premise

The operator is trained and tested only on strain-energy functions of the six-parameter separable cubic form, and the CANO trunk features are exactly those six monomials; so the demonstrated accuracy establishes recovery of six coefficients within a pre-chosen model family, not discovery of a general constitutive function.

What would settle it

Generate synthetic displacement and reaction data from a strain-energy function outside the six-parameter separable cubic family—e.g., one with a coupling term C11(¯I1−3)(¯I2−3^(3/2))—and feed it to the trained operator. If the predicted energy deviates substantially from the truth over the sampled invariant domain, the claim that the operator maps into the infinite-dimensional space is refuted.

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

If this is right

  • Trained operators characterize a new material without any optimization loop, so material identification becomes near-instantaneous and could be run in real time during experiments.
  • Because the branch input is a spectral encoding rather than raw grid values, the same operator works for different measurement resolutions, partial fields, and specimens of different side lengths and thicknesses, subject to the stated scaling relations.
  • The predicted models are admissible by construction, so they can be plugged directly into finite element simulations without additional regularization or stability checks.
  • The SVD-based identifiability analysis provides a quantitative way to judge whether a given experiment, material class, and set of measured quantities can uniquely determine material parameters, which can guide experimental design.
  • The framework reframes simulation databases as reusable assets: data generated for one-off calibration can instead train operators that serve many future inverse problems.

Where Pith is reading between the lines

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

  • Editorial inference: The current demonstration is essentially coefficient identification within a known six-parameter family, so the more ambitious 'discovery' claim needs a test on out-of-family functions; if the operator were trained on a mixture of model forms, it might interpolate across families in ways the fixed-feature CANO cannot.
  • Editorial inference: The Laplacian-encoding trick generalizes beyond this setting: representing input fields in the eigenbasis of the problem domain regularizes any inverse operator learning task, potentially reducing the data needed for other ill-posed identification problems.
  • Editorial inference: Conditioning the operator on the loading protocol or geometry descriptor (rather than a fixed standardized experiment) would be a natural next step and is implicitly supported by the scaling-invariance results, since the operator must currently be retrained for fundamentally different boundary conditions.
  • Editorial inference: The identifiability matrix could be used as a cheap a priori experiment-design tool: maximizing its smallest singular value over candidate loading protocols would give an optimization criterion for designing tests that most reliably separate material parameters.

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

4 major / 5 minor

Summary. The paper proposes two neural operator architectures, PANO and CANO, that map full-field displacement measurements and net reaction forces directly to the hyperelastic strain-energy density function, eliminating the need to solve iterative inverse calibration problems. The operators combine a DeepONet-style branch net, which encodes displacement fields via Laplacian eigenfunctions and force data, with a physics-constrained trunk net based on PANN or CANN architectures. Training data are generated from 3000 finite element simulations of a thin plate with a central hole, using the six-parameter separable cubic strain-energy family of Eq. (24). The authors test on unseen samples, noisy displacement data, missing spatial data, different discretizations, and scaled geometries, and they analyze identifiability within the six-parameter CANO class via singular values of the weak-form equilibrium system. The central claim is that, after training, a single forward pass yields a physically admissible strain-energy density from a new experimental dataset.

Significance. If the claimed generalization holds, the framework is a useful step toward amortized inverse material characterization: it avoids per-material retraining and iterative optimization, builds physical admissibility (polyconvexity, objectivity, stress-free reference state) directly into the output, and introduces a discretization-robust encoding via Laplacian eigenfunctions. The simulated-data study is carefully designed within its family, and the SVD identifiability analysis in Section 4.2 is a valuable diagnostic. However, the demonstrated scope is narrower than the title and abstract suggest: all training and test materials lie in a six-parameter polynomial family, and CANO's trunk uses exactly that basis. The evidence therefore supports coefficient identification within a preselected model family rather than discovery of general constitutive functions in the infinite-dimensional space W of Eq. (2).

major comments (4)
  1. [§2.3.2, §2.5, Eq. (24); Abstract] All training and test labels are generated from the six-parameter separable cubic family in Eq. (24), and the CANO trunk in Eq. (23) is exactly the monomial basis of that family. The learned operator therefore cannot represent any strain-energy function outside this family, and the empirical results demonstrate six-coefficient identification within a known basis rather than discovery of a general function in the infinite-dimensional space W of Eq. (2). This undercuts the abstract's "infinite-dimensional output space" claim and the Introduction's statement that the framework "reduces a priori assumptions about material behavior" and "can predict a variety of functional forms" (Section 1). PANO is architecturally more flexible but is trained and tested only on this same family, so it provides no out-of-family evidence. Either add out-of-family experiments or rephrase the claims to in-famil
  2. [§3.1–3.3] The quantitative evaluation is limited to histograms and selected best/median/worst curves; no numerical MSE values, quantiles, or confidence intervals are reported anywhere. This makes it impossible to assess the frequency of outliers and the magnitude of the "excellent agreement" claimed in Sections 3.1–3.2. For instance, the PANO worst-case MSE is dismissed as an "exceptional outlier" without reporting its rank or frequency. Please report numeric error statistics (e.g., median, 95th percentile, max MSE, or relative L2 errors) over the full test set, and if feasible confidence intervals across repeated training runs.
  3. [§3.3; Eq. (21)] The noise-robustness study perturbs only the displacement fields; the scalar reaction force R_data, which enters the branch net and provides the absolute scale through Eq. (21), is left unchanged. Since the abstract claims robustness to "noisy data" generally and the load-cell signal is an equally important measurement, the current experiment does not fully support the claim. Please add tests with reaction-force noise (and combined noise), or explicitly restrict the robustness claim to displacement noise.
  4. [§4.2, Eqs. (32)–(38)] The identifiability analysis is an honest and useful check, but its scope is narrow: it assumes plane stress, exact displacement data, and that the true model lies in CANO's six-parameter feature space. The reported σ_min/σ_max ≳ 0.15 is only for "representative datasets," and the tolerance τ is not specified. Because the conclusion "no non-identifiable directions" is used to argue that the framework regularizes the inverse problem, the analysis should be documented across all 3000 simulations and for noisy or missing-data cases; otherwise the conclusion should be stated more cautiously.
minor comments (5)
  1. [§2.5] In the parameter-scaling sentence, "C01 = ¯C10/5" appears to be a typo; it should be "C01 = ¯C01/5".
  2. [Eq. (21)] The scaling by ∥R_data∥ is used both in normalizing the branch-net input and in rescaling the final strain energy. Please clarify this two-step use in the notation to avoid confusion.
  3. [Figures 5–6] The histograms in Figs. 5a and 6a lack axis labels in the captions; the x-axis presumably is MSE, but this should be stated explicitly.
  4. [§3.4] The geometry-scaling test is performed for a single representative material model. This demonstration is suggestive, but a statement about the robustness across the full test set would strengthen the claim of geometry invariance.
  5. [§4.2, Eq. (27)] The determinant expression 1/det(I+∇u_t(X)) is written with a scalar denominator; the notation would be clearer if the inverse of the 2×2 block and the scalar determinant were distinguished explicitly.

Circularity Check

0 steps flagged

No significant circularity: the neural operators are trained on independently simulated FE data; the CANO basis overlap with the training family is a scope limitation, not a circular derivation.

full rationale

The central mapping from displacement/reaction-force data to strain-energy densities is learned in a standard supervised setting: labeled pairs (u_data, R_data, W_data) come from independent finite-element simulations (Sections 2.4–2.5), and test accuracy is evaluated on held-out simulations (Sections 3.1–3.2). Nothing in that loop is fitted to the test set or renamed from a fitted parameter. The main limitation, correctly identified by the reader, is that CANO's trunk features (Eq. 23) are exactly the six monomials of the training family (Eq. 24), so CANO cannot represent out-of-family constitutive functions and the reported tests demonstrate in-family coefficient identification rather than general functional discovery. However, this is an expressiveness/validation-scope issue, not circularity: the output basis is an explicit architectural choice, and the branch coefficients are learned from data rather than defined as the target. PANO's trunk is more flexible but is likewise only trained and evaluated on the same six-parameter family, so the 'infinite-dimensional discovery' claim is under-supported; again, this is an overclaim, not a definitional reduction. The identifiability analysis in Section 4.2 is an independent SVD sensitivity check, and the paper itself restricts its conclusion to the six-parameter class and to exact plane-stress displacement data. Self-citations (e.g., Flaschel et al. 2021, 2026a,b,c) concern implementation details, prior database-based Material Fingerprinting, and EUCLID-style matrix assembly; they are not used as the justification for the neural operator result. Universal-approximation citations are used only as background, and the paper explicitly notes these guarantees do not apply to the finite-capacity networks used in practice. Overall, the derivation chain is self-contained against independent simulation data, and no load-bearing step reduces by construction or by self-citation to its own input.

Axiom & Free-Parameter Ledger

4 free parameters · 5 axioms · 0 invented entities

All central claims rest on the restricted constitutive class and the assumption that synthetic FE data stand in for experimental reality. The free parameters are mostly design choices for the training distribution and network capacity. No invented physical entities appear; the latent features psi are computational coordinates, not new physics.

free parameters (4)
  • C_ij sampling ranges and scaling = [0,1]^6 with per-term scaling
    Defines the six-parameter separable cubic model family in Eq. (24) used to generate all training and test data; chosen by hand, not derived.
  • N_phi = number of Laplacian eigenfunctions = 100
    Chosen hyperparameter for displacement encoding; balances compression vs. fidelity; no convergence study is reported.
  • N_b = latent feature dimension = 12 (PANO), 6 (CANO)
    Determines expressive capacity of the operator output; tuned via optuna on validation for some variants, otherwise heuristic.
  • I_sample invariant sampling domain = I*_2 <= I*_1 + 3 (heuristic)
    Chosen to approximate invariants encountered in training simulations; authors state that alternative sampling domains did not significantly affect the results.
axioms (5)
  • domain assumption Material response is incompressible, isotropic, hyperelastic, with separable strain energy W(I1*,I2*) in C^2 satisfying Eq. (2).
    The entire function space and training data are restricted to this class; the inverse problem and identifiability analysis are only meaningful inside it. Eq. (2), Section 2.1.1.
  • domain assumption Forward FE simulations in FEniCSx exactly solve the incompressible finite-strain BVP for sampled models.
    Training data are treated as ground truth; no verification of FE convergence or experimental validation is provided. Appendix A.
  • domain assumption For identifiability analysis, plane stress conditions hold and displacement fields fully determine F via Eq. (27) with P_33 = 0.
    Section 4.2 uses the reduced plane-stress assumption to derive A and the SVD rank; thin-plate approximation is stated but not quantified.
  • domain assumption Geometry rescaling rules (thickness scales force by s3; side length scales displacements and force by s1) remain valid for the tested scaling factors.
    Used to apply the operator to different-sized specimens (Sections 2.4 and 3.4) without retraining; relies on thin-plate plane-stress validity.
  • standard math Monotone convex functions of the lifted polyconvex invariants I1* and I2* = I2^(3/2) - 3^(3/2) are polyconvex, per Hartmann and Neff.
    Basis for claiming polyconvexity by construction in Eqs. (21)-(23) and in the PANO trunk design.

pith-pipeline@v1.3.0-alltime-deepseek · 30192 in / 13528 out tokens · 146227 ms · 2026-08-02T00:17:41.353995+00:00 · methodology

0 comments
read the original abstract

Characterizing the mechanical response of materials traditionally requires solving optimization problems in which model parameters are calibrated or trained to minimize the discrepancy between model predictions and experimental data. This process can be computationally expensive and time-consuming. To overcome this limitation, we propose two neural operator architectures that directly map experimentally measured data to the constitutive functions governing the mechanical response of the material: Physics-Augmented Neural Operators (PANO) and Constitutive Artificial Neural Operators (CANO). The proposed neural operators approximate the mapping between the infinite-dimensional input space of full-field displacement measurements and net reaction forces, and the infinite-dimensional output space of hyperelastic strain-energy density functions. The displacement fields are encoded through Laplacian eigenfunctions to obtain discretization-independent and noise-robust predictions. Our framework constrains the output space to physically admissible material models that satisfy fundamental physical requirements by design. The neural operators are trained on simulated data tuples of displacement fields and reaction forces for a range of material models. Once trained, the neural operators enable near-instantaneous material characterization and require only a single forward pass to infer the strain-energy density function from a given experimental dataset. We test the predictive power of the neural operators for unseen data, noisy data, data with missing information, data from different spatial discretizations, and data from geometries of different sizes.

Figures

Figures reproduced from arXiv: 2607.15049 by Burigede Liu, Ellen Kuhl, Moritz Flaschel.

Figure 1
Figure 1. Figure 1: Graphical abstract: The Physics-Augmented Neural Operator (PANO) and Constitutive Artificial Neural Operator (CANO) are composed of a trunk and branch net. The trunk net maps the deformation gradient to a latent space. The Physics-Augmented Neural Network (PANN) or Constitutive Artificial Neural Network (CANN) architecture chosen for the trunk net guarantee physical admissibility of the material model. The… view at source ↗
Figure 2
Figure 2. Figure 2: Physics-Augmented Neural Operator (PANO) architecture. [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Constitutive Artificial Neural Operator (CANO) architecture. [PITH_FULL_IMAGE:figures/full_fig_p012_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Specimen geometry (left) and symmetry-reduced computational domain with applied boundary conditions (right). [PITH_FULL_IMAGE:figures/full_fig_p012_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: PANO results for unseen testing data. 3.1. PANO We use the Adam optimizer to train the PANO on 80% of the simulation data. 10% of the simulation data is used for validation during the training process. The corresponding validation loss is used to select the best model over all training epochs. Another 10% of the simulation data is held aside and is entirely unseen during training. These testing data are us… view at source ↗
Figure 6
Figure 6. Figure 6: CANO results for unseen testing data. A comparison of the PANO and CANO results reveals that the MSE distribution for CANO is shifted towards lower values, and the minimum and median MSE values are closer to one another than those obtained with PANO. This indi￾cates that CANO predicts the underlying material models more consistently and accurately than PANO. Importantly, the maximum MSE achieved by CANO is… view at source ↗
Figure 7
Figure 7. Figure 7: CANO results for the median MSE sample of the [PITH_FULL_IMAGE:figures/full_fig_p016_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: CANO results for the median MSE sample of the [PITH_FULL_IMAGE:figures/full_fig_p017_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: CANO results for unseen data stemming from a different discretization and different geometry dimensions. deliberately differs from that employed during the generation of the training data. Fig. 9a shows the spatial locations at which displacements are evaluated on the surface of the new specimen geometry. The resulting simulation data are then used as input to the neural operator. To this end, the displace… 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

300 extracted references · 108 canonical work pages

  1. [1]

    Strain , author =

    A. Strain , author =. 2017 , pages =. doi:10.1111/str.12229 , abstract =

    work page doi:10.1111/str.12229 2017

  2. [2]

    International Journal for Numerical Methods in Engineering , author =

    Adjoint-weighted variational formulation for the direct solution of inverse problems of general linear elasticity with full interior data:. International Journal for Numerical Methods in Engineering , author =. 2010 , pages =. doi:10.1002/nme.2760 , abstract =

    work page doi:10.1002/nme.2760 2010

  3. [3]

    Computer Methods in Applied Mechanics and Engineering , author =

    Solution of the nonlinear elasticity imaging inverse problem:. Computer Methods in Applied Mechanics and Engineering , author =. 2011 , pages =. doi:10.1016/j.cma.2010.12.018 , abstract =

    work page doi:10.1016/j.cma.2010.12.018 2011

  4. [4]

    Composite Structures , author =

    A new approach to inverse identification of mechanical properties of composite materials:. Composite Structures , author =. 2013 , pages =. doi:10.1016/j.compstruct.2013.04.025 , abstract =

    work page doi:10.1016/j.compstruct.2013.04.025 2013

  5. [5]

    Biomechanics and Modeling in Mechanobiology , author =

    Inverse identification of local stiffness across ascending thoracic aortic aneurysms , volume =. Biomechanics and Modeling in Mechanobiology , author =. 2019 , pages =. doi:10.1007/s10237-018-1073-0 , abstract =

    work page doi:10.1007/s10237-018-1073-0 2019

  6. [6]

    Annals of Biomedical Engineering , author =

    Identifying. Annals of Biomedical Engineering , author =. 2019 , pages =. doi:10.1007/s10439-019-02204-5 , abstract =

    work page doi:10.1007/s10439-019-02204-5 2019

  7. [7]

    Cardiovascular Engineering , author =

    A. Cardiovascular Engineering , author =. 2006 , pages =. doi:10.1007/s10558-006-9013-5 , abstract =

    work page doi:10.1007/s10558-006-9013-5 2006

  8. [8]

    Journal of Biomechanical Engineering , author =

    A. Journal of Biomechanical Engineering , author =. 2003 , pages =. doi:10.1115/1.1574333 , abstract =

    work page doi:10.1115/1.1574333 2003

  9. [9]

    Biomechanics and Modeling in Mechanobiology , author =

    Identifying heterogeneous anisotropic properties in cerebral aneurysms: a pointwise approach , volume =. Biomechanics and Modeling in Mechanobiology , author =. 2011 , pages =. doi:10.1007/s10237-010-0225-7 , language =

    work page doi:10.1007/s10237-010-0225-7 2011

  10. [10]

    Biomechanics and Modeling in Mechanobiology , author =

    Pointwise characterization of the elastic properties of planar soft tissues: application to ascending thoracic aneurysms , volume =. Biomechanics and Modeling in Mechanobiology , author =. 2015 , pages =. doi:10.1007/s10237-014-0646-9 , language =

    work page doi:10.1007/s10237-014-0646-9 2015

  11. [11]

    Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences , author =

    Digital image correlation-based point-wise inverse characterization of heterogeneous material properties of gallbladder. Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences , author =. 2014 , pages =. doi:10.1098/rspa.2014.0152 , abstract =

    arXiv 2014

  12. [12]

    Pointwise identification of elastic properties in nonlinear heterogeneous membranes, and application to soft tissues , url =

    Zhao, Xuefeng , month = may, year =. Pointwise identification of elastic properties in nonlinear heterogeneous membranes, and application to soft tissues , url =. doi:10.17077/etd.4qcg3dn5 , file =

    work page doi:10.17077/etd.4qcg3dn5

  13. [13]

    Computer Methods in Applied Mechanics and Engineering , author =

    Nonlinear material identification of heterogeneous isogeometric. Computer Methods in Applied Mechanics and Engineering , author =. 2022 , pages =. doi:10.1016/j.cma.2021.114442 , abstract =

    arXiv 2022

  14. [14]

    Computer Methods in Applied Mechanics and Engineering , author =

    Elastic properties of anisotropic vascular membranes examined by inverse analysis , volume =. Computer Methods in Applied Mechanics and Engineering , author =. 2009 , pages =. doi:10.1016/j.cma.2008.08.002 , abstract =

    work page doi:10.1016/j.cma.2008.08.002 2009

  15. [15]

    Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences , author =

    Estimation of the distributions of anisotropic, elastic properties and wall stresses of saccular cerebral aneurysms by inverse analysis , volume =. Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences , author =. 2008 , pages =. doi:10.1098/rspa.2007.0332 , abstract =

    arXiv 2008

  16. [16]

    Journal of Computational and Applied Mathematics , author =

    A numerical framework for material characterisation of inhomogeneous hyperelastic membranes by inverse analysis , volume =. Journal of Computational and Applied Mathematics , author =. 2010 , pages =. doi:10.1016/j.cam.2009.12.049 , abstract =

    work page doi:10.1016/j.cam.2009.12.049 2010

  17. [17]

    IEEE Transactions on Neural Networks , author =

    Universal approximation to nonlinear operators by neural networks with arbitrary activation functions and its application to dynamical systems , volume =. IEEE Transactions on Neural Networks , author =. 1995 , pages =. doi:10.1109/72.392253 , abstract =

    doi:10.1109/72.392253 1995

  18. [18]

    Arbitrary-

    Yu, Annan and Becquey, Chloé and Halikias, Diana and Mallory, Matthew Esmaili and Townsend, Alex , month = sep, year =. Arbitrary-

  19. [19]

    IEEE Transactions on Circuits and Systems I: Fundamental Theory and Applications , author =

    Approximations for nonlinear functionals , volume =. IEEE Transactions on Circuits and Systems I: Fundamental Theory and Applications , author =. 1992 , pages =. doi:10.1109/81.109247 , language =

    work page doi:10.1109/81.109247 1992

  20. [20]

    Journal of Machine Learning Research , author =

    On. Journal of Machine Learning Research , author =. 2021 , pages =

    2021

  21. [21]

    Li, Zongyi and Kovachki, Nikola and Azizzadenesheli, Kamyar and Liu, Burigede and Bhattacharya, Kaushik and Stuart, Andrew and Anandkumar, Anima , year =. Fourier. doi:10.48550/arXiv.2010.08895 , abstract =

    Pith/arXiv arXiv doi:10.48550/arxiv.2010.08895 2010

  22. [22]

    Optimal choice of spatially adaptive parameters in total generalized variation via bilevel optimization with applications to

    Sirotenko, Clemens , year =. Optimal choice of spatially adaptive parameters in total generalized variation via bilevel optimization with applications to

  23. [23]

    Journal of Mathematical Imaging and Vision , author =

    A. Journal of Mathematical Imaging and Vision , author =. 2011 , pages =. doi:10.1007/s10851-010-0251-1 , abstract =

    doi:10.1007/s10851-010-0251-1 2011

  24. [24]

    Physica D: Nonlinear Phenomena , author =

    Nonlinear total variation based noise removal algorithms , volume =. Physica D: Nonlinear Phenomena , author =. 1992 , pages =. doi:10.1016/0167-2789(92)90242-F , language =

    doi:10.1016/0167-2789(92)90242-f 1992

  25. [25]

    Journal of Applied Physics , author =

    Thermoelasticity and. Journal of Applied Physics , author =. 1956 , pages =. doi:10.1063/1.1722351 , abstract =

    work page doi:10.1063/1.1722351 1956

  26. [26]

    Journal of the Mechanics and Physics of Solids , author =

    Inelastic constitutive relations for solids:. Journal of the Mechanics and Physics of Solids , author =. 1971 , pages =. doi:10.1016/0022-5096(71)90010-X , language =

    doi:10.1016/0022-5096(71)90010-x 1971

  27. [27]

    Amos, Brandon and Xu, Lei and Kolter, J Zico , year =. Input. Proceedings of the 34th

  28. [28]

    Journal of Machine Learning Research , author =

    Deep. Journal of Machine Learning Research , author =. 2018 , pages =

    2018

  29. [29]

    2014 , pages =

    IEEE Transactions on Medical Imaging , author =. 2014 , pages =. doi:10.1109/TMI.2014.2337321 , abstract =

    arXiv 2014

  30. [30]

    Long, Zichao and Lu, Yiping and Ma, Xianzhong and Dong, Bin , month = jan, year =

  31. [31]

    Journal of Computational Physics , author =

    Data driven governing equations approximation using deep neural networks , volume =. Journal of Computational Physics , author =. 2019 , pages =. doi:10.1016/j.jcp.2019.06.042 , language =

    doi:10.1016/j.jcp.2019.06.042 2019

  32. [32]

    Kovachki, Nikola and Li, Zongyi and Liu, Burigede and Azizzadenesheli, Kamyar and Bhattacharya, Kaushik and Stuart, Andrew and Anandkumar, Anima , month = apr, year =. Neural

  33. [33]

    SIAM Review , author =

    Optimization. SIAM Review , author =. 2018 , pages =. doi:10.1137/16M1080173 , abstract =

    doi:10.1137/16m1080173 2018

  34. [34]

    Journal of Machine Learning Research , author =

    Adaptive. Journal of Machine Learning Research , author =. 2011 , pages =

    2011

  35. [35]

    and Ba, Jimmy , month = jan, year =

    Kingma, Diederik P. and Ba, Jimmy , month = jan, year =. Adam:

  36. [36]

    Mathematics of Control, Signals, and Systems , author =

    Approximation by superpositions of a sigmoidal function , volume =. Mathematics of Control, Signals, and Systems , author =. 1989 , pages =

    1989

  37. [37]

    1988 , pages =

    Capabilities of three-layered perceptrons , isbn =. 1988 , pages =. doi:10.1109/ICNN.1988.23901 , abstract =

    arXiv 1988

  38. [38]

    1988 , pages =

    There exists a neural network that does not make avoidable mistakes , isbn =. 1988 , pages =. doi:10.1109/ICNN.1988.23903 , abstract =

    arXiv 1988

  39. [39]

    Neural Networks , author =

    On the approximate realization of continuous mappings by neural networks , volume =. Neural Networks , author =. 1989 , pages =. doi:10.1016/0893-6080(89)90003-8 , abstract =

    doi:10.1016/0893-6080(89)90003-8 1989

  40. [40]

    2022 , file =

    Unsupervised discovery of pressure-sensitive plastic yield criteria with non-associative flow rules under convex constraints , author =. 2022 , file =

    2022

  41. [41]

    Acta Numerica , author =

    Approximation theory of the. Acta Numerica , author =. 1999 , pages =. doi:10.1017/S0962492900002919 , abstract =

    doi:10.1017/s0962492900002919 1999

  42. [42]

    IEEE Signal Processing Magazine , author =

    Using. IEEE Signal Processing Magazine , author =. 2018 , pages =. doi:10.1109/MSP.2017.2760358 , language =

    arXiv 2018

  43. [43]

    IEEE Signal Processing Magazine , author =

    Plug-and-. IEEE Signal Processing Magazine , author =. 2023 , pages =. doi:10.1109/MSP.2022.3199595 , language =

    arXiv 2023

  44. [44]

    IEEE Signal Processing Magazine , author =

    Algorithm. IEEE Signal Processing Magazine , author =. 2021 , pages =. doi:10.1109/MSP.2020.3016905 , language =

    arXiv 2021

  45. [45]

    Daya , editor =

    Reddy, B. Daya , editor =. Introductory. 1998 , doi =

    1998

  46. [46]

    Introductory functional analysis with applications , isbn =

    Kreyszig, Erwin , year =. Introductory functional analysis with applications , isbn =

  47. [47]

    Phase-regularized and displacement-regularized compressed sensing for fast magnetic resonance elastography , journal =

    Mohammed, Shahed , year =. Phase-regularized and displacement-regularized compressed sensing for fast magnetic resonance elastography , journal =

  48. [48]

    Journal of Physics: Conference Series , author =

    Identification of viscoelastic properties by magnetic resonance elastography , volume =. Journal of Physics: Conference Series , author =. 2007 , pages =. doi:10.1088/1742-6596/73/1/012009 , abstract =

    work page doi:10.1088/1742-6596/73/1/012009 2007

  49. [49]

    Medical Image Analysis , author =

    Magnetic. Medical Image Analysis , author =. 2021 , pages =. doi:10.1016/j.media.2021.102212 , language =

    arXiv 2021

  50. [50]

    Physics in Medicine & Biology , author =

    Ensemble. Physics in Medicine & Biology , author =. 2022 , pages =. doi:10.1088/1361-6560/ac9fa1 , abstract =

    work page doi:10.1088/1361-6560/ac9fa1 2022

  51. [51]

    Journal of Mathematical Analysis and Applications , author =

    Stability analysis for. Journal of Mathematical Analysis and Applications , author =. 2015 , pages =. doi:10.1016/j.jmaa.2015.05.029 , language =

    work page doi:10.1016/j.jmaa.2015.05.029 2015

  52. [52]

    SIAM Journal on Applied Mathematics , author =

    On an. SIAM Journal on Applied Mathematics , author =. 2011 , pages =. doi:10.1137/110832082 , abstract =

    work page doi:10.1137/110832082 2011

  53. [53]

    Nature Reviews Physics , author =

    Magnetic resonance elastography from fundamental soft-tissue mechanics to diagnostic imaging , volume =. Nature Reviews Physics , author =. 2022 , pages =. doi:10.1038/s42254-022-00543-2 , language =

    doi:10.1038/s42254-022-00543-2 2022

  54. [54]

    2019 , pages =

    SoftwareX , author =. 2019 , pages =. doi:10.1016/j.softx.2019.01.004 , abstract =

    work page doi:10.1016/j.softx.2019.01.004 2019

  55. [55]

    2020 , pages =

    IEEE Transactions on Systems, Man, and Cybernetics: Systems , author =. 2020 , pages =. doi:10.1109/TSMC.2017.2717481 , abstract =

    arXiv 2020

  56. [56]

    2013 , pages =

    Computers in Biology and Medicine , author =. 2013 , pages =. doi:10.1016/j.compbiomed.2013.04.005 , abstract =

    work page doi:10.1016/j.compbiomed.2013.04.005 2013

  57. [57]

    Soviet Math

    Solution of incorrectly formulated problems and the regularization method , volume =. Soviet Math. Dokl. , author =. 1963 , pages =

    1963

  58. [58]

    IEEE Transactions on Image Processing , author =

    Deterministic edge-preserving regularization in computed imaging , volume =. IEEE Transactions on Image Processing , author =. 1997 , pages =. doi:10.1109/83.551699 , abstract =

    doi:10.1109/83.551699 1997

  59. [59]

    Mathematical problems in image processing: partial differential equations and the calculus of variations , isbn =

    Aubert, Gilles and Kornprobst, Pierre , year =. Mathematical problems in image processing: partial differential equations and the calculus of variations , isbn =

  60. [60]

    Neural Networks , author =

    Multilayer feedforward networks are universal approximators , volume =. Neural Networks , author =. 1989 , pages =. doi:10.1016/0893-6080(89)90020-8 , abstract =

    doi:10.1016/0893-6080(89)90020-8 1989

  61. [61]

    Bulletin of Mathematical Biophysics , author =

    A logical calculus of the ideas immanent in nervous activity , volume =. Bulletin of Mathematical Biophysics , author =. 1948 , file =

    1948

  62. [62]

    SIAM Journal on Applied Mathematics , author =

    Total. SIAM Journal on Applied Mathematics , author =. 2004 , pages =. doi:10.1137/S0036139903422784 , abstract =

    doi:10.1137/s0036139903422784 2004

  63. [63]

    Tyrrell , year =

    Rockafellar, R. Tyrrell , year =. Convex analysis , isbn =

  64. [64]

    Machine Learning , author =

    Regularisation of neural networks by enforcing. Machine Learning , author =. 2021 , pages =. doi:10.1007/s10994-020-05929-w , abstract =

    doi:10.1007/s10994-020-05929-w 2021

  65. [65]

    Funktionalanalysis , author =

  66. [66]

    Magnetic Resonance in Medicine , author =

    Correction of gradient-induced phase errors in radial. Magnetic Resonance in Medicine , author =. 2014 , pages =. doi:10.1002/mrm.24643 , abstract =

    work page doi:10.1002/mrm.24643 2014

  67. [67]

    Magnetic Resonance in Medicine , author =

    Centering the projection reconstruction trajectory:. Magnetic Resonance in Medicine , author =. 2003 , pages =. doi:10.1002/mrm.10501 , language =

    work page doi:10.1002/mrm.10501 2003

  68. [68]

    Journal of Mathematical Imaging and Vision , author =

    Bilevel. Journal of Mathematical Imaging and Vision , author =. 2017 , pages =. doi:10.1007/s10851-016-0662-8 , abstract =

    doi:10.1007/s10851-016-0662-8 2017

  69. [69]

    Communications on Pure and Applied Mathematics , author =

    Stable signal recovery from incomplete and inaccurate measurements , volume =. Communications on Pure and Applied Mathematics , author =. 2006 , pages =. doi:10.1002/cpa.20124 , language =

    doi:10.1002/cpa.20124 2006

  70. [70]

    Yang, Yan and Li, Huibin and Sun, Jian and Xu, Zongben , year =. Deep

  71. [71]

    2011 , pages =

    IEEE Transactions on Medical Imaging , author =. 2011 , pages =. doi:10.1109/TMI.2010.2090538 , abstract =

    arXiv 2011

  72. [72]

    and Maierhofer, Georg and Williams, Guy and Schönlieb, Carola-Bibiane and Ehrhardt, Matthias J

    Sherry, Ferdia and Benning, Martin and Reyes, Juan Carlos De los and Graves, Martin J. and Maierhofer, Georg and Williams, Guy and Schönlieb, Carola-Bibiane and Ehrhardt, Matthias J. , month = jun, year =. Learning the

  73. [73]

    Neural Networks , author =

    Multilayer feedforward networks with a nonpolynomial activation function can approximate any function , volume =. Neural Networks , author =. 1993 , pages =. doi:10.1016/S0893-6080(05)80131-5 , abstract =

    doi:10.1016/s0893-6080(05)80131-5 1993

  74. [74]

    Dong, Guozhi and Hintermüller, Michael and Papafitsoros, Kostas , year =. Annual

  75. [75]

    Improvement of image processing by using homogeneous neural networks with fractional derivatives theorem , isbn =

    Gomolka, Zbigniew , year =. Improvement of image processing by using homogeneous neural networks with fractional derivatives theorem , isbn =. Conference. doi:10.3934/proc.2011.2011.505 , file =

    arXiv 2011

  76. [76]

    Wang, Bohan and Meng, Qi and Chen, Wei and Liu, Tie-Yan , year =. The

  77. [77]

    Gradient

    Lyu, Kaifeng and Li, Jian , month = dec, year =. Gradient

  78. [78]

    LeCun, Yann and Chopra, Sumit and Hadsell, Raia and Ranzato, Marc’Aurelio and Huang, Fu Jie , year =. A

  79. [79]

    Clinical Anatomy , author =

    Magnetic resonance elastography:. Clinical Anatomy , author =. 2010 , pages =. doi:10.1002/ca.21006 , language =

    work page doi:10.1002/ca.21006 2010

  80. [80]

    Acta Numerica , author =

    Solving inverse problems using data-driven models , volume =. Acta Numerica , author =. 2019 , pages =. doi:10.1017/S0962492919000059 , abstract =

    doi:10.1017/s0962492919000059 2019

Showing first 80 references.