pith. sign in

arxiv: 2606.23834 · v1 · pith:6RT4BY7Qnew · submitted 2026-06-22 · 💻 cs.CE

Efficient implementation of graph autoencoders for model-order reduction of systems with sharp gradients

Liam K Magargal , Parisa Khodabakhshi This is my paper

Pith reviewed 2026-06-26 05:56 UTC · model grok-4.3

classification 💻 cs.CE
keywords graph autoencodersmodel-order reductionsharp gradientslatent space dynamics identificationoperator learninggraph neural networksreduced-order modelspoint cloud error metric
0
0 comments X

The pith

Graph autoencoders enable accurate model-order reduction for systems with sharp gradients where linear methods fail.

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

The paper demonstrates that graph autoencoders can perform nonlinear dimensionality reduction on high-dimensional dynamical systems featuring sharp gradients, for which proper orthogonal decomposition is inadequate. It introduces GNN-LaSDI, which pairs the autoencoder with an operator-learning step to directly approximate the time evolution of the resulting low-dimensional latent representation. Tests on two example problems show GNN-LaSDI achieves substantially higher accuracy than POD-LaSDI while remaining far cheaper to run than geometric deep least-squares Petrov-Galerkin projection. A new point cloud error metric is also presented that tracks the locations of sharp gradients more informatively than conventional norms.

Core claim

GNN-LaSDI employs graph autoencoders to obtain nonlinear low-dimensional representations of systems with sharp gradients and then applies operator learning to predict the temporal evolution of those representations. For the studied problems this yields significantly greater accuracy than POD-LaSDI at only modestly higher cost, while remaining substantially cheaper than GD-LSPG. The point cloud error metric supplies a more intuitive assessment of accuracy in the vicinity of sharp gradients than standard error measures.

What carries the argument

Graph autoencoder nonlinear dimensionality reduction combined with operator learning on the latent representation (GNN-LaSDI).

If this is right

  • GNN-LaSDI supplies a practical middle ground between the accuracy of geometric deep projection methods and the speed of POD-based latent dynamics identification.
  • The point cloud error metric evaluates reduced-order model performance on sharp-gradient locations more informatively than conventional L2-type norms.
  • Graph autoencoders succeed at capturing nonlinear features that defeat linear reduction techniques such as POD.
  • The overall framework balances predictive accuracy against computational speedup for the class of problems examined.

Where Pith is reading between the lines

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

  • The same graph-autoencoder reduction could be tested on three-dimensional problems or on systems whose sharp features move or interact.
  • Replacing the operator learner with a parametric version might allow the reduced model to handle families of problems without retraining.
  • The point cloud metric could be generalized to other localized solution features such as contact lines or material interfaces.

Load-bearing premise

The graph autoencoder learns a latent representation that preserves the essential nonlinear structure of sharp gradients well enough for the operator-learning step to produce accurate long-term predictions.

What would settle it

A new test problem containing sharp gradients in which GNN-LaSDI latent dynamics diverge from the full-order solution over time despite low autoencoder reconstruction error on training data.

Figures

Figures reproduced from arXiv: 2606.23834 by Liam K Magargal, Parisa Khodabakhshi.

Figure 1
Figure 1. Figure 1: The graph autoencoder employed in this study, adapted from [15]. The figure corresponds to the case [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Initial condition and parameter-space sampling for Kobayashi’s solidification model. The left figure presents the initial conditions for [PITH_FULL_IMAGE:figures/full_fig_p010_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Comparison of reconstruction and state prediction errors as a function of latent state dimension. The left figure shows the range of graph [PITH_FULL_IMAGE:figures/full_fig_p010_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: State prediction errors for GNN-LaSDI (left column), GD-LSPG (middle column), and POD-LaSDI (right column) across the test [PITH_FULL_IMAGE:figures/full_fig_p011_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Order parameter state solutions and corresponding point cloud representations of the solid-liquid interface obtained via Canny edge [PITH_FULL_IMAGE:figures/full_fig_p012_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Point cloud errors (17) for GNN-LaSDI, GD-LSPG, and POD-LaSDI as a function of time. Columns correspond to latent state [PITH_FULL_IMAGE:figures/full_fig_p013_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Range of speedup factors for GNN-LaSDI, GD-LSPG, and POD-LaSDI across all 16 test parameter sets as a function of latent state [PITH_FULL_IMAGE:figures/full_fig_p014_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Computational domain and boundary conditions for the 2D bow shock problem governed by the Euler equations and solved using a [PITH_FULL_IMAGE:figures/full_fig_p015_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Range of graph autoencoder and POD reconstruction errors (left) and state prediction errors for GNN-LaSDI, GD-LSPG, and [PITH_FULL_IMAGE:figures/full_fig_p017_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: State prediction errors plotted with respect to Mach number, [PITH_FULL_IMAGE:figures/full_fig_p017_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Pressure field solution and corresponding point cloud representations of the bow shock obtained from the Ducros sensor for the FOM [PITH_FULL_IMAGE:figures/full_fig_p019_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Point cloud errors (17) for GNN-LaSDI, GD-LSPG, and POD-LaSDI plotted with respect to time. The left, middle, and right columns [PITH_FULL_IMAGE:figures/full_fig_p020_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Pressure field solutions and corresponding point cloud representations of the bow shock for the test parameters yielding the largest state [PITH_FULL_IMAGE:figures/full_fig_p021_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Point cloud errors (17) plotted with respect to time for the GNN-LaSDI, GD-LSPG, and POD-LaSDI solutions at latent state [PITH_FULL_IMAGE:figures/full_fig_p022_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: Range of speedup factors for GNN-LaSDI, GD-LSPG, and POD-LaSDI across all 10 test parameters, plotted with respect to the latent [PITH_FULL_IMAGE:figures/full_fig_p022_15.png] view at source ↗
Figure 16
Figure 16. Figure 16: State prediction errors as functions of latent state dimension [PITH_FULL_IMAGE:figures/full_fig_p023_16.png] view at source ↗
read the original abstract

This study investigates the efficient deployment of graph autoencoders, a class of graph neural networks (GNNs), for model-order reduction (MOR) of high-dimensional dynamical systems. The proposed framework leverages graph autoencoders to perform nonlinear dimensionality reduction, enabling low-dimensional representations of systems characterized by sharp gradients for which conventional linear approximations, such as proper orthogonal decomposition (POD), are inadequate. Specifically, this study introduces graph neural network latent space dynamics identification (GNN-LaSDI). GNN-LaSDI employs an operator learning framework to directly approximate the temporal evolution of the graph autoencoder's latent representation. The performance of GNN-LaSDI is assessed against both geometric deep least-squares Petrov-Galerkin (GD-LSPG and POD latent space dynamics identification (POD-LaSDI), which combines POD-based dimensionality reduction with operator learning. In addition to standard error metrics, this work presents a novel point cloud error metric specifically tailored to evaluate the accuracy of the identified locations of sharp gradients within the solution. The effectiveness of the metric and the proposed MOR framework is demonstrated through two numerical experiments featuring sharp gradients. For the studied problems, GNN-LaSDI incurs a substantially lower computational cost than GD-LSPG, though it remains slightly more computationally expensive than POD-LaSDI. However, GNN-LaSDI achieves significantly greater accuracy than POD-LaSDI, thereby providing a balance between predictive accuracy and computational speedup. Additionally, the results indicate that the proposed point cloud error provides a more intuitive and informative measure of reduced-order model accuracy in regions with sharp gradients than conventional error metrics.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper proposes GNN-LaSDI, a model-order reduction framework that uses graph autoencoders for nonlinear dimensionality reduction of high-dimensional dynamical systems featuring sharp gradients, followed by an operator-learning step to identify the latent-space dynamics. It compares GNN-LaSDI against POD-LaSDI and GD-LSPG on two numerical test problems, reporting that GNN-LaSDI achieves substantially higher accuracy than POD-LaSDI while incurring only modestly higher online cost than POD-LaSDI and far lower cost than GD-LSPG. A novel point-cloud error metric is introduced to assess the fidelity of predicted sharp-gradient locations.

Significance. If the accuracy claims hold under the reported conditions, the work would be significant for MOR applications involving discontinuities or moving fronts, where linear bases such as POD are known to be inadequate. The point-cloud metric provides a domain-appropriate evaluation tool that conventional L2 or relative-error norms do not capture. The framework also demonstrates a practical trade-off between nonlinear reduction quality and computational expense.

major comments (2)
  1. [§3.2, Eq. (8)] §3.2, Eq. (8): The autoencoder loss is defined solely via reconstruction error plus latent regularization; no auxiliary term enforces that the encoder maps the physical location of a discontinuity or front to a consistent (approximately equivariant) latent coordinate. Consequently, small reconstruction errors can produce inconsistent latent trajectories whose long-time integration by the learned operator may not preserve the reported accuracy advantage.
  2. [Numerical experiments] Numerical experiments section: The accuracy advantage of GNN-LaSDI over POD-LaSDI is asserted via the point-cloud metric, yet the manuscript provides neither error bars across multiple random seeds nor an explicit statement of the data-exclusion or hyper-parameter selection protocol used to generate the tabulated results. Without these, it is impossible to determine whether the claimed superiority is statistically robust or sensitive to the particular realizations shown.
minor comments (2)
  1. The definition of the point-cloud error metric should be stated explicitly (including the matching threshold and normalization) rather than referred to only by name.
  2. Figure captions for the solution snapshots should indicate the time instants shown and whether the plotted fields are full-order or reconstructed latent solutions.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their insightful comments on our manuscript. We address each major comment below and outline the revisions we will make to strengthen the presentation and statistical rigor of the work.

read point-by-point responses

  1. Referee: [§3.2, Eq. (8)] §3.2, Eq. (8): The autoencoder loss is defined solely via reconstruction error plus latent regularization; no auxiliary term enforces that the encoder maps the physical location of a discontinuity or front to a consistent (approximately equivariant) latent coordinate. Consequently, small reconstruction errors can produce inconsistent latent trajectories whose long-time integration by the learned operator may not preserve the reported accuracy advantage.

    Authors: We appreciate this observation regarding the loss formulation. The graph autoencoder operates on a mesh that explicitly encodes spatial connectivity, and the training snapshots are generated from physics-consistent evolution of the sharp gradients. Consequently, the encoder learns a latent mapping in which front locations vary smoothly and consistently across the dataset, as required for stable operator learning. While an explicit equivariance penalty is not included, the combination of graph structure and data physics provides the necessary consistency, which is reflected in the reported accuracy gains. In the revision we will add a short paragraph in §3.2 clarifying this implicit mechanism and noting that future extensions could incorporate an auxiliary term if needed for other applications. revision: partial

  2. Referee: [Numerical experiments] Numerical experiments section: The accuracy advantage of GNN-LaSDI over POD-LaSDI is asserted via the point-cloud metric, yet the manuscript provides neither error bars across multiple random seeds nor an explicit statement of the data-exclusion or hyper-parameter selection protocol used to generate the tabulated results. Without these, it is impossible to determine whether the claimed superiority is statistically robust or sensitive to the particular realizations shown.

    Authors: We agree that reporting variability across seeds and documenting the experimental protocol would improve confidence in the results. In the revised manuscript we will (i) repeat all experiments with five independent random seeds, (ii) report means and standard deviations of the point-cloud metric, and (iii) add a dedicated subsection describing the train/validation/test split, hyper-parameter search procedure (grid search with cross-validation), and any data-exclusion rules applied. revision: yes

Circularity Check

0 steps flagged

No significant circularity in derivation chain

full rationale

The paper defines GNN-LaSDI as the combination of a graph autoencoder (with reconstruction + latent regularization loss) followed by operator learning on the latent coordinates, then validates performance via direct numerical comparison to POD-LaSDI and GD-LSPG on two test problems using both standard and point-cloud error metrics. No equation or step reduces a reported prediction or accuracy claim to a fitted parameter by construction, nor does any central premise rest on a self-citation chain that would make the result tautological. The framework is self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract provides no information on free parameters, axioms, or invented entities; all such elements remain unidentified.

pith-pipeline@v0.9.1-grok · 5826 in / 1076 out tokens · 16854 ms · 2026-06-26T05:56:15.882172+00:00 · methodology

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

68 extracted references · 2 canonical work pages · 2 internal anchors

  1. [1]

    Sirovich, Turbulence and the dynamics of coherent structures part I: coherent structures, Quarterly of Applied Mathematics 45 (3) (1987) 561–571

    L. Sirovich, Turbulence and the dynamics of coherent structures part I: coherent structures, Quarterly of Applied Mathematics 45 (3) (1987) 561–571

    1987

  2. [2]

    T. Lieu, C. Farhat, M. Lesoinne, Reduced-order fluid/structure modeling of a complete aircraft configuration, Computer Methods in Applied Mechanics and Engineering 195 (41) (2006) 5730–5742

    2006

  3. [3]

    C. R. Wentland, K. Duraisamy, C. Huang, Scalable projection-based reduced-order models for large multiscale fluid systems, AIAA Journal 61 (10) (2023) 4499–4523

    2023

  4. [4]

    S. A. McQuarrie, C. Huang, K. E. Willcox, Data-driven reduced-order models via regularised operator inference for a single-injector combustion process, Journal of the Royal Society of New Zealand 51 (2) (2021) 194–211

    2021

  5. [5]

    Welper, Interpolation of functions with parameter dependent jumps by transformed snapshots, SIAM Journal on Scientific Computing 39 (4) (2017) A1225–A1250

    G. Welper, Interpolation of functions with parameter dependent jumps by transformed snapshots, SIAM Journal on Scientific Computing 39 (4) (2017) A1225–A1250

    2017

  6. [6]

    Peherstorfer, Model reduction for transport-dominated problems via online adaptive bases and adaptive sam- pling, SIAM Journal on Scientific Computing 42 (5) (2020) A2803–A2836

    B. Peherstorfer, Model reduction for transport-dominated problems via online adaptive bases and adaptive sam- pling, SIAM Journal on Scientific Computing 42 (5) (2020) A2803–A2836

    2020

  7. [7]

    Peherstorfer, Breaking the Kolmogorov barrier with nonlinear model reduction, Notices of the American Mathematical Society 69 (5) (2022) 725–733

    B. Peherstorfer, Breaking the Kolmogorov barrier with nonlinear model reduction, Notices of the American Mathematical Society 69 (5) (2022) 725–733

    2022

  8. [8]

    Barnett, C

    J. Barnett, C. Farhat, Quadratic approximation manifold for mitigating the Kolmogorov barrier in nonlinear projection-based model order reduction, Journal of Computational Physics (2022) 111348

    2022

  9. [9]

    Geelen, S

    R. Geelen, S. Wright, K. Willcox, Operator inference for non-intrusive model reduction with quadratic mani- folds, Computer Methods in Applied Mechanics and Engineering 403 (2023) 115717

    2023

  10. [10]

    Barnett, C

    J. Barnett, C. Farhat, Y . Maday, Neural-network-augmented projection-based model order reduction for mitigat- ing the Kolmogorov barrier to reducibility, Journal of Computational Physics 492 (2023) 112420

    2023

  11. [11]

    Ares de Parga, R

    S. Ares de Parga, R. Tezaur, C. G. Hernández, C. Farhat, Nonlinear projection-based model order reduction with machine learning regression for closure error modeling in the latent space, Computer Methods in Applied Mechanics and Engineering 448 (2026) 118443. 24

    2026

  12. [12]

    J. Guo, D. Xiao, Nonlinear model reduction by probabilistic manifold decomposition, SIAM Journal on Scien- tific Computing 48 (1) (2026) A209–A235

    2026

  13. [13]

    K. Lee, K. T. Carlberg, Model reduction of dynamical systems on nonlinear manifolds using deep convolutional autoencoders, Journal of Computational Physics 404 (2020) 108973

    2020

  14. [14]

    K. Lee, K. T. Carlberg, Deep conservation: A latent-dynamics model for exact satisfaction of physical conser- vation laws, in: Proceedings of the AAAI Conference on Artificial Intelligence, V ol. 35, 2021, pp. 277–285

    2021

  15. [15]

    Magargal, P

    L. Magargal, P. Khodabakhshi, S. Rodriguez, J. Jaworski, J. Michopoulos, Projection-based model-order reduc- tion via graph autoencoders suited for unstructured meshes, Data-Centric Engineering 6 (2025) e52

    2025

  16. [16]

    Wiewel, M

    S. Wiewel, M. Becher, N. Thuerey, Latent space physics: Towards learning the temporal evolution of fluid flow, Computer Graphics Forum 38 (2) (2019) 71–82

    2019

  17. [17]

    Fresca, L

    S. Fresca, L. Dede, A. Manzoni, A comprehensive deep learning-based approach to reduced order modeling of nonlinear time-dependent parametrized PDEs, Journal of Scientific Computing 87 (2) (2021) 1–36

    2021

  18. [18]

    R. Fu, D. Xiao, I. Navon, F. Fang, L. Yang, C. Wang, S. Cheng, A non-linear non-intrusive reduced order model of fluid flow by auto-encoder and self-attention deep learning methods, International Journal for Numerical Methods in Engineering 124 (13) (2023) 3087–3111

    2023

  19. [19]

    R. Fu, D. Xiao, A. Buchan, X. Lin, Y . Feng, G. Dong, A parametric non-linear non-intrusive reduce-order model using deep transfer learning, Computer Methods in Applied Mechanics and Engineering 438 (2025) 117807

    2025

  20. [20]

    Bui-Thanh, K

    T. Bui-Thanh, K. Willcox, O. Ghattas, Model reduction for large-scale systems with high-dimensional paramet- ric input space, SIAM Journal on Scientific Computing 30 (6) (2008) 3270–3288

    2008

  21. [21]

    Carlberg, C

    K. Carlberg, C. Bou-Mosleh, C. Farhat, Efficient non-linear model reduction via a least-squares Petrov–Galerkin projection and compressive tensor approximations, International Journal for Numerical Methods in Engineering 86 (2) (2011) 155–181

    2011

  22. [22]

    Carlberg, M

    K. Carlberg, M. Barone, H. Antil, Galerkin v. least-squares Petrov-Galerkin projection in nonlinear model re- duction, Journal of Computational Physics 330 (2017) 693–734

    2017

  23. [23]

    L. K. Magargal, P. Khodabakhshi, S. N. Rodriguez, Hyper-reduction-free reduced-order newton solvers for projection-based model-order reduction of nonlinear dynamical systems, Engineering with Computers 42 (2026) 133

    2026

  24. [24]

    E. J. Parish, C. R. Wentland, K. Duraisamy, The adjoint Petrov–Galerkin method for non-linear model reduction, Computer Methods in Applied Mechanics and Engineering 365 (2020) 112991

    2020

  25. [25]

    Antoulas, Approximation of large-scale dynamical systems, Society for Industrial and Applied Mathematics, Philadelphia, PA, 2009

    A. Antoulas, Approximation of large-scale dynamical systems, Society for Industrial and Applied Mathematics, Philadelphia, PA, 2009

    2009

  26. [26]

    Barrault, Y

    M. Barrault, Y . Maday, N. C. Nguyen, A. T. Patera, An ‘empirical interpolation’ method: application to efficient reduced-basis discretization of partial differential equations, Comptes Rendus Mathematique 339 (9) (2004) 667–672

    2004

  27. [27]

    Chaturantabut, D

    S. Chaturantabut, D. C. Sorensen, Nonlinear model reduction via discrete empirical interpolation, SIAM Journal on Scientific Computing 32 (5) (2010) 2737–2764

    2010

  28. [28]

    Drmac, S

    Z. Drmac, S. Gugercin, A new selection operator for the discrete empirical interpolation method—improved a priori error bound and extensions, SIAM Journal on Scientific Computing 38 (2) (2016) A631–A648

    2016

  29. [29]

    Carlberg, C

    K. Carlberg, C. Farhat, J. Cortial, D. Amsallem, The GNAT method for nonlinear model reduction: effective implementation and application to computational fluid dynamics and turbulent flows, Journal of Computational Physics 242 (2013) 623–647. 25

    2013

  30. [30]

    Farhat, T

    C. Farhat, T. Chapman, P. Avery, Structure-preserving, stability, and accuracy properties of the energy- conserving sampling and weighting method for the hyper reduction of nonlinear finite element dynamic models, International Journal for Numerical Methods in Engineering 102 (5) (2015) 1077–1110

    2015

  31. [31]

    P. J. Schmid, Dynamic mode decomposition of numerical and experimental data, Journal of Fluid Mechanics 656 (2010) 5–28

    2010

  32. [32]

    Peherstorfer, K

    B. Peherstorfer, K. Willcox, Data-driven operator inference for nonintrusive projection-based model reduction, Computer Methods in Applied Mechanics and Engineering 306 (2016) 196–215

    2016

  33. [33]

    S. A. McQuarrie, P. Khodabakhshi, K. E. Willcox, Nonintrusive reduced-order models for parametric partial differential equations via data-driven operator inference, SIAM Journal on Scientific Computing 45 (4) (2023) A1917–A1946

    2023

  34. [34]

    C. Gu, Qlmor: A projection-based nonlinear model order reduction approach using quadratic-linear represen- tation of nonlinear systems, IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems 30 (9) (2011) 1307–1320

    2011

  35. [35]

    Kramer, K

    B. Kramer, K. E. Willcox, Nonlinear model order reduction via lifting transformations and proper orthogonal decomposition, AIAA journal 57 (6) (2019) 2297–2307

    2019

  36. [36]

    E. Qian, B. Kramer, B. Peherstorfer, K. Willcox, Lift & Learn: physics-informed machine learning for large- scale nonlinear dynamical systems, Physica D: Nonlinear Phenomena 406 (2020) 132401

    2020

  37. [37]

    Swischuk, B

    R. Swischuk, B. Kramer, C. Huang, K. Willcox, Learning physics-based reduced-order models for a single- injector combustion process, AIAA Journal 58 (6) (2020) 2658–2672

    2020

  38. [38]

    Khodabakhshi, K

    P. Khodabakhshi, K. E. Willcox, Non-intrusive data-driven model reduction for differential–algebraic equations derived from lifting transformations, Computer Methods in Applied Mechanics and Engineering 389 (2022) 114296

    2022

  39. [39]

    W. D. Fries, X. He, Y . Choi, Lasdi: Parametric latent space dynamics identification, Computer Methods in Applied Mechanics and Engineering 399 (2022) 115436

    2022

  40. [40]

    X. He, Y . Choi, W. D. Fries, J. L. Belof, J.-S. Chen, gLaSDI: Parametric physics-informed greedy latent space dynamics identification, Journal of Computational Physics 489 (2023) 112267

    2023

  41. [41]

    Qian, I.-G

    E. Qian, I.-G. Farcas, K. Willcox, Reduced operator inference for nonlinear partial differential equations, SIAM Journal on Scientific Computing 44 (4) (2022) A1934–A1959

    2022

  42. [42]

    Karypis, V

    G. Karypis, V . Kumar, Multilevel k-way partitioning scheme for irregular graphs, Journal of Parallel and Dis- tributed Computing (1998)

    1998

  43. [43]

    Barwey, V

    S. Barwey, V . Shankar, V . Viswanathan, R. Maulik, Multiscale graph neural network autoencoders for inter- pretable scientific machine learning, Journal of Computational Physics 495 (2023) 112537

    2023

  44. [44]

    T. Tran, L. Magargal, P. Khodabakhshi, Scalable node clustering for graph autoencoders used in model-order reduction, in: AIAA SCITECH 2026 Forum, 2026

    2026

  45. [45]

    Magargal, S

    L. Magargal, S. Moreno-Rivera, P. Khodabakhshi, Topology-inspired clustering hierarchy for graph autoen- coders via spectral graph theory, in: AIAA SCITECH 2026 Forum, 2026

    2026

  46. [46]

    Hamilton, Z

    W. Hamilton, Z. Ying, J. Leskovec, Inductive representation learning on large graphs, Advances in Neural Infor- mation Processing Systems 30 (2017)

    2017

  47. [47]

    Y . Kim, Y . Choi, D. Widemann, T. Zohdi, A fast and accurate physics-informed neural network reduced order model with shallow masked autoencoder, Journal of Computational Physics 451 (2022) 110841. 26

    2022

  48. [48]

    S. L. Brunton, J. L. Proctor, J. N. Kutz, Discovering governing equations from data by sparse identification of nonlinear dynamical systems, Proceedings of the national academy of sciences 113 (15) (2016) 3932–3937

    2016

  49. [49]

    M. R. Chmiel, J. Barnett, C. Farhat, Unified LSPG model reduction framework and assessment for hypersonic computational fluid dynamics, AIAA Journal 63 (1) (2025) 72–90

    2025

  50. [50]

    Geelen, K

    R. Geelen, K. Willcox, Localized non-intrusive reduced-order modelling in the operator inference framework, Philosophical Transactions of the Royal Society A 380 (2229) (2022) 20210206

    2022

  51. [51]

    Cucchiara, A

    S. Cucchiara, A. Iollo, T. Taddei, H. Telib, Model order reduction by convex displacement interpolation, Journal of Computational Physics 514 (2024) 113230

    2024

  52. [52]

    Canny, A computational approach to edge detection, Pattern Analysis and Machine Intelligence, IEEE Trans- actions on PAMI-8 (1986) 679 – 698

    J. Canny, A computational approach to edge detection, Pattern Analysis and Machine Intelligence, IEEE Trans- actions on PAMI-8 (1986) 679 – 698

    1986

  53. [53]

    Ducros, V

    F. Ducros, V . Ferrand, F. Nicoud, C. Weber, D. Darracq, C. Gacherieu, T. Poinsot, Large-eddy simulation of the shock/turbulence interaction, Journal of Computational Physics 152 (2) (1999) 517–549

    1999

  54. [54]

    R. T. Rockafellar, R. J. Wets, Variational analysis, Springer

  55. [55]

    A. A. Taha, A. Hanbury, An efficient algorithm for calculating the exact Hausdorffdistance, IEEE Transactions on Pattern Analysis and Machine Intelligence 37 (11) (2015) 2153–2163

    2015

  56. [56]

    Kobayashi, Modeling and numerical simulations of dendritic crystal growth, Physica D: Nonlinear Phenom- ena 63 (3) (1993) 410–423

    R. Kobayashi, Modeling and numerical simulations of dendritic crystal growth, Physica D: Nonlinear Phenom- ena 63 (3) (1993) 410–423

    1993

  57. [57]

    Nishikawa, K

    H. Nishikawa, K. Kitamura, Very simple, carbuncle-free, boundary-layer-resolving, rotated-hybrid Riemann solvers, Journal of Computational Physics 227 (4) (2008) 2560–2581

    2008

  58. [58]

    Modesti, S

    D. Modesti, S. Pirozzoli, A low-dissipative solver for turbulent compressible flows on unstructured meshes, with OpenFOAM implementation, Computers & Fluids 152 (2017) 14–23

    2017

  59. [59]

    Tibshirani, Regression shrinkage and selection via the lasso, Journal of the Royal Statistical Society Series B: Statistical Methodology 58 (1) (1996) 267–288

    R. Tibshirani, Regression shrinkage and selection via the lasso, Journal of the Royal Statistical Society Series B: Statistical Methodology 58 (1) (1996) 267–288

    1996

  60. [60]

    Maulik, B

    R. Maulik, B. Lusch, P. Balaprakash, Reduced-order modeling of advection-dominated systems with recurrent neural networks and convolutional autoencoders, Physics of Fluids 33 (3) (2021) 037106

    2021

  61. [61]

    Paszke, S

    A. Paszke, S. Gross, F. Massa, A. Lerer, J. Bradbury, G. Chanan, T. Killeen, Z. Lin, N. Gimelshein, L. Antiga, et al., Pytorch: an imperative style, high-performance deep learning library, Advances in Neural Information Processing Systems 32 (2019)

    2019

  62. [62]

    M. Fey, J. E. Lenssen, Fast graph representation learning with PyTorch Geometric, arXiv preprint arXiv:1903.02428 (2019)

    work page internal anchor Pith review Pith/arXiv arXiv 1903

  63. [63]

    Glorot, Y

    X. Glorot, Y . Bengio, Understanding the difficulty of training deep feedforward neural networks, in: Proceed- ings of the Thirteenth International Conference on Artificial Intelligence and Statistics, JMLR Workshop and Conference Proceedings, 2010, pp. 249–256

    2010

  64. [64]

    D. P. Kingma, J. Ba, Adam: A method for stochastic optimization, arXiv preprint arXiv:1412.6980 (2014)

    work page internal anchor Pith review Pith/arXiv arXiv 2014

  65. [65]

    Clevert, T

    D.-A. Clevert, T. Unterthiner, S. Hochreiter, Fast and accurate deep network learning by exponential linear units (ELUs), in: International Conference on Learning Representations (ICLR), 2016

    2016

  66. [66]

    C. R. Harris, K. J. Millman, S. J. van der Walt, R. Gommers, P. Virtanen, D. Cournapeau, E. Wieser, J. Taylor, S. Berg, N. J. Smith, R. Kern, M. Picus, S. Hoyer, M. H. van Kerkwijk, M. Brett, A. Haldane, J. F. del Río, M. Wiebe, P. Peterson, P. Gérard-Marchant, K. Sheppard, T. Reddy, W. Weckesser, H. Abbasi, C. Gohlke, T. E. Oliphant, Array programming wi...

    2020

  67. [67]

    Virtanen, R

    P. Virtanen, R. Gommers, T. E. Oliphant, M. Haberland, T. Reddy, D. Cournapeau, E. Burovski, P. Peterson, W. Weckesser, J. Bright, S. J. van der Walt, M. Brett, J. Wilson, K. J. Millman, N. Mayorov, A. R. J. Nelson, E. Jones, R. Kern, E. Larson, C. J. Carey,˙I. Polat, Y . Feng, E. W. Moore, J. VanderPlas, D. Laxalde, J. Perktold, R. Cimrman, I. Henriksen,...

    2020

  68. [68]

    # of MP operations

    S. K. Lam, A. Pitrou, S. Seibert, Numba: A LLVM-based Python JIT compiler, in: Proceedings of the Second Workshop on the LLVM Compiler Infrastructure in HPC, 2015, pp. 1–6. Appendix A. Fundamentals of Geometric deep least-squares Petrov-Galerkin The deep LSPG framework [13] employed a nonlinear manifold LSPG projection using a CNN-based autoen- coder. Whi...

    2015