Advancing Generalization in PINNs through Latent-Space Representations

Gao Huang; Honghui Wang; Shiji Song; Yifan Pu

REVIEW 4 major objections 7 minor 44 references

Advancing Generalization in PINNs through Latent-Space Representations

T0 review · 4 major / 7 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read PIDO claims that a physics-informed neural PDE solver can generalize across unseen initial conditions, PDE coefficients, and time horizons by auto-decoding solutions into a latent space and learning latent dynamics with a Neural ODE…

desk verdict Strong empirical proposal for a data-free latent-dynamics PDE solver, but a concrete math error in its main regularizer and an unproven identifiability claim mean the numbers should be treated with caution until fixed and reproduced. read the letter →

arxiv 2411.19125 v1 pith:TU3ZYG4J submitted 2024-11-28 cs.LG

Honghui Wang , Yifan Pu , Shiji Song , Gao Huang This is my paper · ORCID

classification cs.LG

keywords physics-informedneuralnetworkslatentspaceregularizationpartialdifferentialequationsspatial-temporaldynamicsmodelingordinaryauto-decodingtemporalextrapolationNavier-Stokes

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

PIDO is a physics-informed neural PDE solver built to generalize across configurations of the same PDE family: unseen initial conditions, unseen coefficients, and time horizons beyond those used in training. The paper’s central claim is that this can be achieved without any exact-solution data, using auto-decoding to project solutions into a low-dimensional latent space, a Neural ODE conditioned on the PDE coefficients to evolve the latent codes, and two latent-space regularizers to keep that evolution stable and drift-free. On five benchmark scenarios (1D Burgers, KdV, combined, and 2D Navier-Stokes), PIDO reports the lowest test-set $L^2$ relative errors, including temporal extrapolation errors far below the best baseline, for example 2.24% versus 9.32% on CE1 Out-t. If the claim holds, physics-informed solvers could serve whole families of problems rather than a single problem instance, and do so with far less data than neural operator methods typically require.

What carries the argument

The load-bearing object is the pair $(D, F)$ acting on a latent code $c_t$: $D$ is a grid-independent implicit neural representation mapping $(c_t, x)$ to the solution value at $x$, and $F$ is a Neural ODE mapping $(c_t, \alpha)$ to the time derivative of the latent code, so predictions come from integrating $c_0 = E(\phi)$ forward and decoding, where $E(\phi)$ denotes the auto-decoded latent fit to the initial condition. Auto-decoding replaces a trained encoder with a short gradient-descent fit of $c_0$ to the initial condition through $D$, which is what lets the model adapt to unseen initial conditions at test time. The argument is carried by two regularizers in latent space: Latent Dynamics Smoothing penalizes $\|F(c_t,\alpha)\|^2$ and the Jacobian norm $\|\nabla F\|_F^2$ to prevent the physics-informed loss from selecting wildly fluctuating dynamics, and Latent Dynamics Alignment penalizes $\|c_t - \tilde{c}_t\|^2$ where $\tilde{c}_t$ is obtained by re-encoding the predicted solution through the decoder, keeping unrolled embeddings inside the decoder’s well-behaved range. Together these regularizers are what transform the residual loss from an unstable, non-extrapolating objective into a trainable one.

What would settle it

Train the CE1 or NS1 scenario with auto-decoding replaced by a directly trained encoder that outputs $c_0$, keeping the PDE-residual loss and both regularizers unchanged: if In-t and Out-t errors stay at the reported levels, generalization to unseen initial conditions does not depend on the claimed auto-decoding mechanism. A second decisive check is to compare the code’s actual smoothing regularizer with Eq. (14), since the printed equality between $\|\nabla F\|_F^2$ and $E_{\varepsilon\sim\mathcal{N}(0,I)}(\varepsilon^T \nabla F \varepsilon)^2$ is not an identity; re-running Table IV with the printed objective would show which form produces the reported gains.

Watch

Extended reading notes

Core claim

The central discovery is that the PDE residual alone, evaluated on unrolled latent trajectories, can supervise a correct latent dynamics model provided those trajectories are regularized in latent space. PIDO encodes each initial condition by auto-decoding (fitting a latent vector with gradient descent through the decoder), integrates that vector with a Neural ODE whose right-hand side is conditioned on the coefficients $\alpha$, and decodes the result with an implicit neural representation to produce the solution at any spatial point. Training uses the initial-condition reconstruction loss plus the PDE residual and boundary loss, with no exact solution data. The paper identifies two failure modes in the latent trajectories—overly complex dynamics that destabilize training and latent embedding drift that degrades extrapolation—and counteracts them with Latent Dynamics Smoothing and Latent Dynamics Alignment. With both regularizers, PIDO reports test $L^2$ relative errors of 1.48% (CE1 In-t), 3.02% (CE2 In-t), 2.35% (NS1 In-t), and 4.59% (NS2 In-t), with Out-t errors of 2.24%, 7.15%, 8.08%, 5.43%, and 10.02%, respectively, beating PI-DeepONet, PINODE, and MAD in every setting.

Load-bearing premise

The load-bearing premise is empirical: minimizing the PDE residual over unrolled latent trajectories, together with the two hand-weighted regularizers, drives the dynamics network to learn the true latent evolution rather than a trivial or non-physical solution, and the paper gives no identifiability or convergence argument for this; the printed form of the smoothing regularizer also appears to rely on an equality that is not true in general.

Editorial extensions

If this is right

A single PIDO model trained on one benchmark family can be applied to unseen initial conditions without retraining; on CE1, CE2, and NS1 the In-t test error drops by 63%, 72%, and 76% relative to the second-best baseline.
The same model predicts beyond its training horizon: the Out-t L2 errors are 2.24% (CE1), 7.15% (CE2), 8.08% (CE3), 5.43% (NS1), and 10.02% (NS2), where every baseline is at least roughly twice as large.
Because training uses only the PDE residual, initial conditions, and boundary conditions, the method does not need exact solution data and is more sample-efficient than data-driven EDM variants such as DINO.
The learned latent representations transfer to downstream tasks: after fine-tuning, a pre-trained PIDO cuts accumulated long-term integration error by 77% and recovers PDE coefficients from as few as two snapshots far more accurately than a from-scratch PINN.
The method also extrapolates outside the training range of Reynolds numbers in NS2, with lower errors than MAD at both low (alpha=550) and high (alpha=1850) unseen values.

Reading between the lines

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

Editorial extension: because the smoothing regularizer penalizes the Jacobian norm of the dynamics, PIDO should be at its best on smooth, diffusion-dominated solutions; a natural test is to apply the same training recipe to shock-forming equations and anneal the smoothing weight over training.
Editorial extension: the alignment regularizer treats the decoder as a pseudo-inverse, so its effectiveness should degrade if the decoder is made deliberately non-injective (for example by increasing latent dimension); this yields a testable prediction about when drift control will or will not help.
Editorial extension: the same latent-space diagnosis could be applied to other physics-informed operator learners: constraining predicted latent trajectories to the decoder’s well-posed region may be a general cure for the optimization pathology of residual losses, not just for this architecture.
Editorial extension: the reported gains are all on periodic-boundary benchmarks, and the paper notes this limitation itself; the most informative next experiment is a non-periodic boundary or irregular geometry, where the grid-independent decoder should be an advantage if the mechanism is general.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note

read the letter

The two things to know: this is a genuine attempt at a data-free PDE solver that generalizes over initial conditions, coefficients, and time horizons in one model; and the benchmark numbers, if they hold up, are a real step forward for the latent-dynamics-plus-PINN line. But as printed, the paper has a concrete mathematical mistake in its key regularizer, and it does not secure its central claim that physics-informed residual training alone identifies the true latent dynamics.

What's actually new: the combination of auto-decoded implicit neural representations, coefficient-conditioned Neural ODE dynamics, and physics-informed loss with the two regularizers. The diagnosis of latent-space failure modes—overly complex dynamics causing instability, and embedding drift hurting extrapolation—is a nice addition, and the ablations show both regularizers matter. The evaluation spans Burgers, KdV, combined 1D equations, and 2D Navier-Stokes, with held-out test sets and downstream tasks. That is more than most PINN papers do.

Where the soft spots are. First, Eq. (14) is wrong as printed. The squared Frobenius norm of the Jacobian is not E_epsilon ||epsilon^T J epsilon||^2 for Gaussian epsilon; that expectation equals 2||J||_F^2 + (tr J)^2. So the regularizer is not fully specified. Since this regularizer is crucial for training stability, the error needs to be fixed and the implemented form stated. Minor related issue: Algorithm 2's auto-decoding step count S is never specified.

Second, the identifiability concern is real, though I would frame it as a gap rather than a demonstrated failure. The objective in Eq. (13) supervises D(c_t) through the initial condition at t=0 and through residual/boundary losses at discrete t. There is a plausible trivial branch where the trajectory moves to a decoder state that is near zero at all sampled t>0, and the discrete residual loss would not see it. Alignment is a self-consistency regularizer, not a cure for this degeneracy, and smoothing penalizes rough dynamics but does not add information. The paper gives visualizations, no convergence or identifiability argument. The empirical results are consistent with genuine learned dynamics, so I would ask the authors to address the gap explicitly rather than reject the paper.

Third, no code or error bars. Combined with the Eq. (14) issue, I cannot verify the reported losses are what was trained. This is a reproducibility burden the authors should lift.

The paper is for people working on physics-informed learning and neural operators, especially those interested in temporal extrapolation and parametric PDEs. The central idea is plausible and the experiments are extensive; the problems are fixable. I would send it to serious peer review with the requirement that the math error be corrected, code released, and the identifiability concern discussed. As published, I would not rely on the numbers yet, but it deserves referee time.

Referee Report

4 major / 7 minor

Summary. The paper proposes PIDO, a physics-informed neural PDE solver that combines auto-decoding of spatial fields into a latent space with a coefficient-conditioned Neural ODE for latent dynamics. Training minimizes initial-condition reconstruction, PDE residual, and boundary losses on unrolled latent trajectories, augmented by two latent-space regularizers: Latent Dynamics Smoothing (RS) and Latent Dynamics Alignment (RA). The authors evaluate on 1D combined equations (CE1-CE3) and 2D Navier-Stokes equations (NS1-NS2) with held-out initial conditions, coefficients, and temporal horizons, reporting large L2-relative-error margins over PI-DeepONet, PINODE, and MAD, as well as favorable comparisons with data-driven baselines. They also demonstrate transfer of the learned representations to long-term integration and inverse problems.

Significance. If the reported results are reproducible, PIDO is a meaningful advance: it is the only method among those compared that generalizes across initial conditions, PDE coefficients, and time horizon in a residual-only (data-free of exact solutions) setting. The held-out evaluation protocol is appropriate, and the reported margins are large, with ablations in Table IV showing that both regularizers contribute to the final performance. The downstream-task results in Table V provide additional evidence of representation transfer. The main weaknesses are that the printed definition of RS in Eq. (14) contains an incorrect identity, the latent-dynamics objective is only heuristically justified with no identifiability or convergence analysis, and no code or error bars are provided. These issues need to be addressed before the claimed generalization can be taken as fully established.

major comments (4)

[Section III-D, Eq. (14)] The second equality in Eq. (14) is not an identity: for J = ∇F(c_t, α), E_{ε∼N(0,I)}∥ε^T J ε∥² = (tr J)² + 2∥(J+J^T)/2∥_F², which is not equal to ∥J∥_F² in general. As printed, the regularizer is therefore underspecified: implementing the quadratic-form expectation gives a different loss from implementing the Frobenius norm. Please state the exact loss used in the experiments, correct Eq. (14), and clarify whether the implementation penalizes the true Jacobian norm or the quadratic-form expectation.
[Section III-C, Eq. (13); Section III-D] The training objective in Eq. (13) admits a trivial-solution branch: because the zero field is a solution of the benchmark PDEs with zero boundary loss, a trajectory that reconstructs φ at t = 0 and then drives D(c_t) to zero at the discrete collocation times t > 0 can make the residual and boundary terms vanish without approximating the true dynamics. Neither RS nor RA excludes this branch: RS penalizes F only at sampled times, and RA aligns c_t to an auto-decoding of the model's own output. The paper provides only qualitative visualizations in Section III-D rather than a quantitative argument. Please add a concrete diagnostic, such as evaluating D(c_t) at 5–10× finer temporal resolution than the training collocation grid for several test trajectories, and report sensitivity of the final errors to the residual time step and to the number of auto-decoding steps. This would demonstrate that the reported generalization is not an artifact of the discrete time grid.
[Section III-D.2, Fig. 4] RA is self-referential: the anchor c̃_t is produced by auto-decoding D(c_t), so the regularizer couples c_t to a pseudo-label that is a function of the same prediction. The claim that c̃_t 'does not exhibit drift' is supported only by one qualitative example (Figure 3), and the mechanism could in principle be a generic shrinkage of c_t toward a low-norm or low-variance region. Please provide quantitative statistics of c_t and c̃_t across the extrapolation horizon, such as distances to the training embedding distribution, and compare RA against simpler alternatives (a fixed pretrained encoder, a norm penalty, or a stop-gradient target) to isolate the effect of the proposed anchor mechanism.
[Tables II–V; Appendix C] All reported errors come from a single run and no code is released. Physics-informed training is known to be sensitive to initialization and loss weights, so the reported margins, especially the large Out-t gains, need multiple-seed results (mean ± std) for the main comparisons. Without error bars, one cannot assess whether the improvements are statistically robust; without code, the discrepancy between the printed Eq. (14) and the actual implementation cannot be resolved. Please provide at least three seeds per setting for Tables II–V and release the code or a detailed configuration.

minor comments (7)

[Eq. (14)] The notation ε ∼ N(0,1) should be ε ∼ N(0,I) when ε is a vector; as written, the quadratic form is dimensionally inconsistent.
[Algorithm 1] The sampling line uses {c̄_t} while the initialization and the alignment update use {c̃_t}; please unify the notation for the anchor embeddings.
[Eq. (10)] The notation c_t(x) in the chain rule is misleading because c_t is a latent vector independent of x; the term should be written as ∂ũ_t(x)/∂c_t · F(c_t, α).
[Section III-B, Eq. (6)] The symbol E is used both for the encoder and for expectation; consider using a different notation for the encoder to avoid confusion.
[Appendix C-A; Algorithm 2] The number S of test-time auto-decoding steps in Algorithm 2 is not reported; please state S for each benchmark along with the collocation-point sampling scheme.
[Table IV] The ablation of RS and RA is reported only for the NS2 scenario; reporting the same ablations on at least one 1D scenario would strengthen the claim that both regularizers are generally necessary.
[Table X] Computational cost is reported only for the CE1 setting; a brief statement for NS1/NS2 would help assess scalability.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: all headline test errors are evaluated against held-out exact solutions, and the self-referential alignment regularizer is an auxiliary consistency term, not a source of prediction.

full rationale

PIDO's central claims are benchmark numbers on test sets whose exact solutions are explicitly 'only used for evaluation'. No parameter is fitted to the test outputs, and the Out-t and coefficient-generalization numbers are computed against ground-truth simulations, so the reported predictions are not forced by construction. The Latent Dynamics Alignment term RA(ct, c~t) = ||ct - E(D(ct))|| is self-referential in that its anchor is generated from the model's own predictions, and the authors explicitly acknowledge that 'pseudo labels do not provide additional information in the data space'; however, this is an auxiliary regularizer, not a reported prediction, and it cannot by itself force the held-out L2 errors. There are no load-bearing self-citations: related-work citations are external baselines, and no uniqueness theorem is imported from the authors' prior work. The paper does contain a clear non-circular correctness issue in Eq. (14), where the stated equality between ||∇F||_F^2 and E||ε^T ∇F ε||^2 is not the standard identity for Gaussian ε; that is an implementation/derivation defect rather than a circular prediction. Consequently, no step in the reported generalization results reduces by construction to the paper's own inputs.

Assumptions & free parameters 5 free parameters · 6 assumptions · 2 invented entities

The method rests on a handful of learned capacity choices (code dimensionality, regularizer weights, optimization schedules) and on the unproved assumption that the physics-informed loss uniquely guides the latent dynamics. No external physical constants or new field entities are introduced; the invented entities are internal computational codes.

free parameters (5)

Latent code dimensionality = 64 (1D), 128 (2D)
Hand-chosen capacity of the latent representation; affects reconstruction and dynamics expressiveness.
Smoothing regularization weight = 0.01
Set globally across benchmarks; trades stability against high-frequency information, as acknowledged in the limitations.
Alignment regularization weight = 1
Set globally; controls how strongly unrolled codes are pulled toward auto-decoded anchors.
Test-time auto-decoding optimization steps = S (unreported)
Algorithm 2 requires S optimization steps to encode a new initial condition, but the paper does not give S or its learning rate.
Residual sampling time step = 1 s for 2D NS, 60 steps over 1.92 s for 1D
The time step size at which PDE residuals are sampled is chosen per problem; the authors use a larger step than baselines for NS and attribute this to smoothing.

assumptions (6)

domain assumption The PDE operator L_alpha and boundary operator B are differentiable and can be evaluated through automatic differentiation on the decoder output.
Required for the physics-informed loss in Eq. (11); standard for PINN methods but assumed for the FourierNet decoder.
domain assumption The decoder D has sufficient capacity to represent the full solution families from low-dimensional codes.
The model's ability to generalize depends on D(c, x) capturing the true solution manifold; no error bound is given.
ad hoc to paper Minimizing the PDE residual with F replacing dc/dt yields the correct latent dynamics; no identifiability or global-convergence guarantee is provided.
Core training objective in Eq. (13); the paper diagnoses failure modes empirically but does not prove this optimization reaches the true dynamics.
ad hoc to paper The equality in Eq. (14) relating the squared Frobenius norm of the Jacobian to a quadratic-form expectation is assumed; it is not correct as stated.
The smoothing regularizer is defined by this equation; if corrected to the standard Hutchinson estimator, the loss changes.
domain assumption Auto-decoding converges to an adequate inverse E after a bounded number of gradient steps, both during training and at test time.
Used to define initial embeddings, anchor embeddings, and new-condition encoding in Algorithms 1 and 2; no convergence certificate is given.
domain assumption Test initial conditions and coefficient ranges are drawn from the same family as training, namely periodic domains and smooth sinusoidal or reference-benchmark initial data.
The generalization claims are limited to this distribution by the benchmark design; boundary-condition diversity is explicitly excluded.

invented entities (2)

Latent embedding c_t
purpose: Low-dimensional state vector representing the PDE solution at time t; encodes initial conditions and is evolved by the Neural ODE.
It is a model-internal representation; no falsifiable prediction exists outside the trained network.
Anchor embedding c_tilde_t
purpose: Auto-decoded latent code of the model's own predicted solution; used as regularization target to reduce drift.
It is derived from the model's own output and decoder, so it carries no external information and cannot serve as independent ground truth.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Advancing Generalization in PINNs through Latent-Space Representations." pith.science (2026). https://pith.science/paper/TU3ZYG4J

@misc{pith2026241119125,
  author       = {Pith},
  title        = {Pith review of: Advancing Generalization in PINNs through Latent-Space Representations},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/TU3ZYG4J}},
  note         = {Machine review of arXiv:2411.19125}
}

read the original abstract

Physics-informed neural networks (PINNs) have made significant strides in modeling dynamical systems governed by partial differential equations (PDEs). However, their generalization capabilities across varying scenarios remain limited. To overcome this limitation, we propose PIDO, a novel physics-informed neural PDE solver designed to generalize effectively across diverse PDE configurations, including varying initial conditions, PDE coefficients, and training time horizons. PIDO exploits the shared underlying structure of dynamical systems with different properties by projecting PDE solutions into a latent space using auto-decoding. It then learns the dynamics of these latent representations, conditioned on the PDE coefficients. Despite its promise, integrating latent dynamics models within a physics-informed framework poses challenges due to the optimization difficulties associated with physics-informed losses. To address these challenges, we introduce a novel approach that diagnoses and mitigates these issues within the latent space. This strategy employs straightforward yet effective regularization techniques, enhancing both the temporal extrapolation performance and the training stability of PIDO. We validate PIDO on a range of benchmarks, including 1D combined equations and 2D Navier-Stokes equations. Additionally, we demonstrate the transferability of its learned representations to downstream applications such as long-term integration and inverse problems.

Figures

Figures reproduced from arXiv: 2411.19125 by the authors.

**Figure 2.** Training instability arises from overly complex dynamics. We randomly sample 3 dimensions from the 128-dim embeddings. We visualize the loss and [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗

**Figure 3.** Time extrapolation degradation arises from latent embedding drift. We take the [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗

Figures from the paper (6 more)

**Figure 4.** Figure 4: Regularization. We apply smoothing regularization [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]

**Figure 5.** Figure 5: Performance (y-axis) of various models on the [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]

**Figure 6.** Figure 6: Visualization of PIDO’s predictions in the long-term integration setting, spanning from t = 0 s to t = 45 s with a step size of 5 s. significant improvement (77% error reduction) through finetuning the dynamics model. This showcases its effectiveness in enhancing long…

**Figure 7.** Figure 7: CE3 test In-t performance vs. numbers of training pairs. E. Visualizations We visualize the predictions of PIDO for the CE3 test set in [PITH_FULL_IMAGE:figures/full_fig_p010_7.png]

**Figure 8.** Figure 8: Prediction of PIDO on 1D combined equations with different α. Ground Truth PIDO PI-DeepONet MAD (a) Reynolds number 950 (b) Reynolds number 1350 Ground Truth PIDO PI-DeepONet MAD [PITH_FULL_IMAGE:figures/full_fig_p011_8.png]

**Figure 9.** Figure 9: Prediction of PIDO on 2D NS equations with different Reynolds number. The last 5 frames are beyond the training horizon. APPENDIX B AUTO-DECODER VERSUS AUTO-ENCODER FOR PHYSICS-INFORMED EDM Since physics-informed training strategies for auto-encoders differ significant…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

44 extracted references · 28 canonical work pages

[1]

The deep ritz method: a deep learning-based numerical algo- rithm for solving variational problems,

B. Yu et al., “The deep ritz method: a deep learning-based numerical algo- rithm for solving variational problems,” Communications in Mathematics and Statistics, vol. 6, no. 1, pp. 1–12, 2018

work page 2018
[2]

Neural operator: Learning maps between function spaces,

N. Kovachki, Z. Li, B. Liu, K. Azizzadenesheli, K. Bhattacharya, A. Stuart, and A. Anandkumar, “Neural operator: Learning maps between function spaces,” arXiv preprint arXiv:2108.08481 , 2021

arXiv 2021
[3]

Message passing neural pde solvers,

J. Brandstetter, D. E. Worrall, and M. Welling, “Message passing neural pde solvers,” in International Conference on Learning Representations , 2021

work page 2021
[4]

Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations,

M. Raissi, P. Perdikaris, and G. 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, vol. 378, pp. 686–707, 2019

2019
[5]

Hidden fluid mechanics: Learning velocity and pressure fields from flow visualizations,

M. Raissi, A. Yazdani, and G. E. Karniadakis, “Hidden fluid mechanics: Learning velocity and pressure fields from flow visualizations,” Science, vol. 367, no. 6481, pp. 1026–1030, 2020

2020
[6]

Uncertainty quantification in estimating blood alcohol concentration from transdermal alcohol level with physics-informed neural networks,

C. Oszkinat, S. E. Luczak, and I. Rosen, “Uncertainty quantification in estimating blood alcohol concentration from transdermal alcohol level with physics-informed neural networks,” IEEE transactions on neural networks and learning systems , vol. 34, no. 10, pp. 8094–8101, 2022

work page 2022
[7]

Physics-informed neural networks with weighted losses by uncertainty evaluation for accurate and stable prediction of manufacturing systems,

J. Hua, Y . Li, C. Liu, P. Wan, and X. Liu, “Physics-informed neural networks with weighted losses by uncertainty evaluation for accurate and stable prediction of manufacturing systems,” IEEE Transactions on Neural Networks and Learning Systems , 2023

2023
[8]

Inherently interpretable physics-informed neural network for battery modeling and prognosis,

F. Wang, Q. Zhi, Z. Zhao, Z. Zhai, Y . Liu, H. Xi, S. Wang, and X. Chen, “Inherently interpretable physics-informed neural network for battery modeling and prognosis,” IEEE Transactions on Neural Networks and Learning Systems, 2023

work page 2023

Show all 44 references

[9]

Fourier neural operator for parametric partial differential equations,

Z. Li, N. B. Kovachki, K. Azizzadenesheli, K. Bhattacharya, A. Stuart, A. Anandkumar et al., “Fourier neural operator for parametric partial differential equations,” in International Conference on Learning Repre- sentations, 2020

2020
[10]

Learning nonlinear operators via deeponet based on the universal approximation theorem of operators,

L. Lu, P. Jin, G. Pang, Z. Zhang, and G. E. Karniadakis, “Learning nonlinear operators via deeponet based on the universal approximation theorem of operators,” Nature machine intelligence , vol. 3, no. 3, pp. 218–229, 2021

2021
[11]

Learning the solution operator of parametric partial differential equations with physics-informed deeponets,

S. Wang, H. Wang, and P. Perdikaris, “Learning the solution operator of parametric partial differential equations with physics-informed deeponets,” Science advances, vol. 7, no. 40, p. eabi8605, 2021

2021
[12]

Deepsdf: Learning continuous signed distance functions for shape representation,

J. J. Park, P. Florence, J. Straub, R. Newcombe, and S. Lovegrove, “Deepsdf: Learning continuous signed distance functions for shape representation,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , 2019, pp. 165–174

2019
[13]

Con- tinuous pde dynamics forecasting with implicit neural representations,

Y . Yin, M. Kirchmeyer, J.-Y . Franceschi, A. Rakotomamonjyet al., “Con- tinuous pde dynamics forecasting with implicit neural representations,” in The Eleventh International Conference on Learning Representations , 2022

2022
[14]

Evolve smoothly, fit consistently: Learning smooth latent dynamics for advection-dominated systems,

Z. Y . Wan, L. Zepeda-Nunez, A. Boral, and F. Sha, “Evolve smoothly, fit consistently: Learning smooth latent dynamics for advection-dominated systems,” in The Eleventh International Conference on Learning Repre- sentations, 2022

2022
[15]

Neu- ral ordinary differential equations,

R. T. Chen, Y . Rubanova, J. Bettencourt, and D. K. Duvenaud, “Neu- ral ordinary differential equations,” Advances in Neural Information Processing Systems, vol. 31, 2018

2018
[16]

Characterizing possible failure modes in physics-informed neural net- works,

A. Krishnapriyan, A. Gholami, S. Zhe, R. Kirby, and M. W. Mahoney, “Characterizing possible failure modes in physics-informed neural net- works,” Advances in Neural Information Processing Systems , vol. 34, pp. 26 548–26 560, 2021

2021
[17]

Understanding and mitigating gradient flow pathologies in physics-informed neural networks,

S. Wang, Y . Teng, and P. Perdikaris, “Understanding and mitigating gradient flow pathologies in physics-informed neural networks,” SIAM Journal on Scientific Computing , vol. 43, no. 5, pp. A3055–A3081, 2021

2021
[18]

Meta-auto-decoder for solving parametric partial differential equations,

X. Huang, Z. Ye, H. Liu, S. Ji, Z. Wang, K. Yang, Y . Li, M. Wang, H. Chu, F. Yuet al., “Meta-auto-decoder for solving parametric partial differential equations,” Advances in Neural Information Processing Systems , vol. 35, pp. 23 426–23 438, 2022

2022
[19]

Physics-informed neural ode (pinode): embedding physics into models using collocation points,

A. Sholokhov, Y . Liu, H. Mansour, and S. Nabi, “Physics-informed neural ode (pinode): embedding physics into models using collocation points,” Scientific Reports, vol. 13, no. 1, p. 10166, 2023

2023
[20]

Implicit neural representations with periodic activation functions,

V . Sitzmann, J. Martel, A. Bergman, D. Lindell, and G. Wetzstein, “Implicit neural representations with periodic activation functions,” Advances in Neural Information Processing Systems , vol. 33, pp. 7462– 7473, 2020

2020
[21]

Multiplicative filter networks,

R. Fathony, A. K. Sahu, D. Willmott, and J. Z. Kolter, “Multiplicative filter networks,” in International Conference on Learning Representations, 2020. 14

2020
[22]

A high-efficient hybrid physics-informed neural networks based on convolutional neural network,

Z. Fang, “A high-efficient hybrid physics-informed neural networks based on convolutional neural network,” IEEE Transactions on Neural Networks and Learning Systems , vol. 33, no. 10, pp. 5514–5526, 2021

2021
[23]

Rigorous a posteriori error bounds for pde-defined pinns,

B. Hillebrecht and B. Unger, “Rigorous a posteriori error bounds for pde-defined pinns,” IEEE Transactions on Neural Networks and Learning Systems, 2023

2023
[24]

Physics-informed neural networks for solving forward and inverse problems in complex beam systems,

T. Kapoor, H. Wang, A. N ´u˜nez, and R. Dollevoet, “Physics-informed neural networks for solving forward and inverse problems in complex beam systems,” IEEE Transactions on Neural Networks and Learning Systems, 2023

2023
[25]

Parallel solution of nonlinear projection equations in a multitask learning framework,

D. Wu and A. Lisser, “Parallel solution of nonlinear projection equations in a multitask learning framework,” IEEE Transactions on Neural Networks and Learning Systems , 2024

2024
[26]

When and why pinns fail to train: A neural tangent kernel perspective,

S. Wang, X. Yu, and P. Perdikaris, “When and why pinns fail to train: A neural tangent kernel perspective,” Journal of Computational Physics , vol. 449, p. 110768, 2022

2022
[27]

Respecting causality is all you need for training physics-informed neural networks,

S. Wang, S. Sankaran, and P. Perdikaris, “Respecting causality is all you need for training physics-informed neural networks,” arXiv preprint arXiv:2203.07404, 2022

2022 arXiv
[28]

Multiadam: Parameter- wise scale-invariant optimizer for multiscale training of physics-informed neural networks,

J. Yao, C. Su, Z. Hao, S. Liu, H. Su, and J. Zhu, “Multiadam: Parameter- wise scale-invariant optimizer for multiscale training of physics-informed neural networks,” in Proceedings of the 40th International Conference on Machine Learning , ser. ICML’23. JMLR.org, 2023

2023
[29]

Separable physics-informed neural networks,

J. Cho, S. Nam, H. Yang, S.-B. Yun, Y . Hong, and E. Park, “Separable physics-informed neural networks,” in Thirty-seventh Conference on Neural Information Processing Systems , 2023

2023
[30]

Multipole graph neural operator for parametric partial differential equations,

Z. Li, N. Kovachki, K. Azizzadenesheli, B. Liu, A. Stuart, K. Bhat- tacharya, and A. Anandkumar, “Multipole graph neural operator for parametric partial differential equations,” Advances in Neural Information Processing Systems, vol. 33, pp. 6755–6766, 2020

2020
[31]

Nonlinear reconstruction for operator learning of pdes with discontinuities,

S. Lanthaler, R. Molinaro, P. Hadorn, and S. Mishra, “Nonlinear reconstruction for operator learning of pdes with discontinuities,” in The Eleventh International Conference on Learning Representations , 2022

2022
[32]

Physics-informed neural operator for learning partial differential equations,

Z. Li, H. Zheng, N. Kovachki, D. Jin, H. Chen, B. Liu, K. Azizzadenesheli, and A. Anandkumar, “Physics-informed neural operator for learning partial differential equations,” arXiv preprint arXiv:2111.03794 , 2021

2021 arXiv
[33]

Learning to optimize multigrid pde solvers,

D. Greenfeld, M. Galun, R. Basri, I. Yavneh, and R. Kimmel, “Learning to optimize multigrid pde solvers,” in International Conference on Machine Learning. PMLR, 2019, pp. 2415–2423

2019
[34]

Snode: Spectral discretization of neural odes for system identification,

A. Quaglino, M. Gallieri, J. Masci, and J. Koutn ´ık, “Snode: Spectral discretization of neural odes for system identification,” in International Conference on Learning Representations , 2019

2019
[35]

Learning dynamical systems from partial observations,

I. Ayed, E. de B ´ezenac, A. Pajot, J. Brajard, and P. Gallinari, “Learning dynamical systems from partial observations,” arXiv preprint arXiv:1902.11136, 2019

1902 arXiv
[36]

Operator learning with neural fields: Tackling pdes on general geometries,

L. Serrano, L. L. Boudec, A. K. Koupa ¨ı, T. X. Wang, Y . Yin, J.-N. Vittaut, and P. Gallinari, “Operator learning with neural fields: Tackling pdes on general geometries,” arXiv preprint arXiv:2306.07266 , 2023

2023 arXiv
[37]

Reduced-order modeling for pa- rameterized pdes via implicit neural representations,

T. Wen, K. Lee, and Y . Choi, “Reduced-order modeling for pa- rameterized pdes via implicit neural representations,” arXiv preprint arXiv:2311.16410, 2023

2023 arXiv
[38]

Automatic differentiation in machine learning: a survey,

A. G. Baydin, B. A. Pearlmutter, A. A. Radul, and J. M. Siskind, “Automatic differentiation in machine learning: a survey,” Journal of machine learning research, vol. 18, no. 153, pp. 1–43, 2018

2018
[39]

How to train your neural ode: the world of jacobian and kinetic regularization,

C. Finlay, J.-H. Jacobsen, L. Nurbekyan, and A. Oberman, “How to train your neural ode: the world of jacobian and kinetic regularization,” in International conference on machine learning . PMLR, 2020, pp. 3154–3164

2020
[40]

Long-time integration of parametric evolution equations with physics-informed deeponets,

S. Wang and P. Perdikaris, “Long-time integration of parametric evolution equations with physics-informed deeponets,” Journal of Computational Physics, vol. 475, p. 111855, 2023

2023
[41]

Solving inverse-pde problems with physics-aware neural networks,

S. Pakravan, P. A. Mistani, M. A. Aragon-Calvo, and F. Gibou, “Solving inverse-pde problems with physics-aware neural networks,” Journal of Computational Physics, vol. 440, p. 110414, 2021

2021
[42]

Learning to solve pde- constrained inverse problems with graph networks,

Q. Zhao, D. B. Lindell, and G. Wetzstein, “Learning to solve pde- constrained inverse problems with graph networks,” arXiv preprint arXiv:2206.00711, 2022

2022 arXiv
[43]

Grids-net: Inverse shape design and identification of scatterers via geometric regularization and physics-embedded deep learning,

S. Nair, T. F. Walsh, G. Pickrell, and F. Semperlotti, “Grids-net: Inverse shape design and identification of scatterers via geometric regularization and physics-embedded deep learning,” Computer Methods in Applied Mechanics and Engineering , vol. 414, p. 116167, 2023

2023
[44]

torchdiffeq,

R. T. Q. Chen, “torchdiffeq,” 2018. [Online]. Available: https: //github.com/rtqichen/torchdiffeq

2018

Pith tools

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

[1] [1]

The deep ritz method: a deep learning-based numerical algo- rithm for solving variational problems,

B. Yu et al., “The deep ritz method: a deep learning-based numerical algo- rithm for solving variational problems,” Communications in Mathematics and Statistics, vol. 6, no. 1, pp. 1–12, 2018

work page 2018

[2] [2]

Neural operator: Learning maps between function spaces,

N. Kovachki, Z. Li, B. Liu, K. Azizzadenesheli, K. Bhattacharya, A. Stuart, and A. Anandkumar, “Neural operator: Learning maps between function spaces,” arXiv preprint arXiv:2108.08481 , 2021

arXiv 2021

[3] [3]

Message passing neural pde solvers,

J. Brandstetter, D. E. Worrall, and M. Welling, “Message passing neural pde solvers,” in International Conference on Learning Representations , 2021

work page 2021

[4] [4]

Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations,

M. Raissi, P. Perdikaris, and G. 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, vol. 378, pp. 686–707, 2019

2019

[5] [5]

Hidden fluid mechanics: Learning velocity and pressure fields from flow visualizations,

M. Raissi, A. Yazdani, and G. E. Karniadakis, “Hidden fluid mechanics: Learning velocity and pressure fields from flow visualizations,” Science, vol. 367, no. 6481, pp. 1026–1030, 2020

2020

[6] [6]

Uncertainty quantification in estimating blood alcohol concentration from transdermal alcohol level with physics-informed neural networks,

C. Oszkinat, S. E. Luczak, and I. Rosen, “Uncertainty quantification in estimating blood alcohol concentration from transdermal alcohol level with physics-informed neural networks,” IEEE transactions on neural networks and learning systems , vol. 34, no. 10, pp. 8094–8101, 2022

work page 2022

[7] [7]

Physics-informed neural networks with weighted losses by uncertainty evaluation for accurate and stable prediction of manufacturing systems,

J. Hua, Y . Li, C. Liu, P. Wan, and X. Liu, “Physics-informed neural networks with weighted losses by uncertainty evaluation for accurate and stable prediction of manufacturing systems,” IEEE Transactions on Neural Networks and Learning Systems , 2023

2023

[8] [8]

Inherently interpretable physics-informed neural network for battery modeling and prognosis,

F. Wang, Q. Zhi, Z. Zhao, Z. Zhai, Y . Liu, H. Xi, S. Wang, and X. Chen, “Inherently interpretable physics-informed neural network for battery modeling and prognosis,” IEEE Transactions on Neural Networks and Learning Systems, 2023

work page 2023