pith. machine review for the scientific record. sign in

arxiv: 2604.16621 · v1 · submitted 2026-04-17 · 💻 cs.CE

Recognition: unknown

Physics-informed, Generative Adversarial Design of Funicular Shells

R\'uben Louren\c{c}o , Ic\'iar Alfaro , Beatriz Moya , Elias Cueto
Authors on Pith no claims yet

Pith reviewed 2026-05-10 06:51 UTC · model grok-4.3

classification 💻 cs.CE
keywords funicular shellsgenerative adversarial networksmembrane factorphysics-informed design3D concrete printingshell structuresstructural efficiencyunreinforced concrete
0
0 comments X

The pith

A physics-informed GAN generates previously unseen three-dimensional funicular shell geometries that favor pure compression.

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

The paper sets out to generalize the ancient idea of funicular forms—structures that carry load through compression alone—from two-dimensional arches to arbitrary three-dimensional shells. This matters for 3D-printed concrete, which cannot contain steel bars and therefore collapses if bending stresses appear. The authors embed a physical constraint called the membrane factor inside a generative adversarial network so that an auxiliary discriminator penalizes any generated shape whose stresses would be dominated by bending. Training produces stable output of smooth, efficient shells that do not appear in the original data set. If the approach holds, designers gain an automated route to complex compression-only geometries without hand-crafted rules or exhaustive search.

Core claim

The central claim is that a modified deep convolutional generative adversarial network, equipped with an auxiliary discriminator driven by the membrane factor, learns to produce realistic and structurally efficient funicular shell geometries in three dimensions. The membrane factor serves as the physics-informed constraint that discourages designs dominated by bending moments, allowing the generator to explore forms that carry loads primarily through in-plane compression.

What carries the argument

The auxiliary discriminator guided by the membrane factor, a scalar that quantifies how completely a shell's stresses act as membrane forces rather than bending moments.

If this is right

  • The model consistently outputs shells with elevated membrane-factor values, indicating load paths that stay close to pure compression.
  • Generated forms are smooth and topologically varied while remaining previously unseen in the training set.
  • Training remains stable even with the added physics penalty, avoiding common GAN failure modes.
  • The framework removes the need for explicit geometric rules to extend the funicular concept into three dimensions.

Where Pith is reading between the lines

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

  • The same constraint-driven GAN approach could be applied to other unreinforced or minimally reinforced structural elements such as vaults or domes.
  • Physical prototypes printed from the generated designs and tested to failure would provide direct evidence of whether high membrane factor predicts real compressive behavior.
  • The method opens a route to fully automated exploration of shell topologies that satisfy equilibrium under gravity without user-specified boundary conditions.

Load-bearing premise

The membrane factor measured on the generated digital geometry serves as a reliable proxy for true funicular behavior once the shell is built and loaded.

What would settle it

Run finite-element analysis on several generated shells under self-weight and measure whether the bending-moment contribution remains below a few percent of total strain energy across the entire surface.

Figures

Figures reproduced from arXiv: 2604.16621 by Beatriz Moya, Elias Cueto, Ic\'iar Alfaro, R\'uben Louren\c{c}o.

Figure 1
Figure 1. Figure 1: Deep convolutional GAN architecture. 3.3. Addressing mode collapse Due to the adversarial training dynamics of GANs and the non-convex nature of the loss landscape, reaching the Nash equilibrium is particularly challenging [13]. This often leads to mode collapse, a common failure mode of GANs in which the generator collapses to a single mode of the data distribution, while the discriminator fails to distin… view at source ↗
Figure 2
Figure 2. Figure 2: Procedurally generated funicular shell geometries. [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Mesh densities obtained with the described mesh properties for surfaces (a) without central hole and (b) with a [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: The 3D surface Γ is discretized into a 2D grid of pixels with coordinates xij and value f(xi, yj , z), which is projected onto the 12 plane. For this work, the geometries were embedded in a square grid; a mask was used to define the holes. In particular, the variables were interpolated from the dense FEA mesh of points to a regular grid of 64 × 64 pixels. The full dataset was split into training (80%), val… view at source ↗
Figure 5
Figure 5. Figure 5: PB-PUNet architecture to predict the behaviour of shell structures. [PITH_FULL_IMAGE:figures/full_fig_p009_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: AD-DCGAN architecture. G converts a PCA-guided latent vector z into synthetic images and binary masks. Both Dshell and Dmask classify the generated images and binary masks, respectively, as real or fake. Daux is a pre-trained PINN that evaluates the physical compliance of the generated shells through the membrane factor distribution mf . Following the detailed representation from [PITH_FULL_IMAGE:figures/… view at source ↗
Figure 7
Figure 7. Figure 7: AD-DCGAN generator architecture for shell design. A [PITH_FULL_IMAGE:figures/full_fig_p011_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: AD-DCGAN discriminator architecture for shell design. A series of four convolutions progressively downsample the [PITH_FULL_IMAGE:figures/full_fig_p011_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Learning curves for the optimized PB-PUNet model trained on the full dataset: (a) training and validation losses, [PITH_FULL_IMAGE:figures/full_fig_p016_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Pixel-wise RRMSE∞ prediction error of the PB-PUNet on the membrane factor distribution mf for the first sample of the test set. Analysing the results for the membrane factor mf , the PB-PUNet model approximates the global FEA solution quite well, exhibiting overall low RRMSE∞ across the shell domain, achieving mean and median errors as low as approximately 2%. The error map shows the higher magnitude erro… view at source ↗
Figure 11
Figure 11. Figure 11: Pixel-wise RRMSE∞ prediction error of the PB-PUNet on the vertical displacement uz for the first sample of the test set. displacement areas across the lateral boundaries, resulting on a maximum error of 17.4%. For the remaining state variables the PB-PUNet model provides excellent results on approximating the global FEA solution with minimally noticeable local differences. For the stress resultants F and … view at source ↗
Figure 12
Figure 12. Figure 12: Features of the penultimate layer of the AD-DCGAN discriminator, corresponding to ground-truth and generated [PITH_FULL_IMAGE:figures/full_fig_p017_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Post-processed surfaces generated by the AD-DCGAN and corresponding membrane factor distribution [PITH_FULL_IMAGE:figures/full_fig_p019_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Pixel-wise RRMSE∞ prediction error for the membrane factor distributions mf of three surfaces generated by the AD-DCGAN model. Contour maps presented column-wise for each surface with corresponding 3D representation at the top. 19 [PITH_FULL_IMAGE:figures/full_fig_p020_14.png] view at source ↗
read the original abstract

Shell structures are pivotal in the fields of architecture and engineering, due to their aesthetic appeal and structural efficiency. Recently, 3D concrete printing has reignited the interest in these structures. But, as printed concrete cannot be reinforced with steel, structures built in this way must be designed to withstand primarily pure compression: they must be funicular shells. Nevertheless, a fundamental challenge remains unsolved since Robert Hooke's discovered the catenary arch in 1675: it is not known whether the concept of a funicular polygon can be generalised to three-dimensional structures. Generative Adversarial Networks (GANs), have shown remarkable success in generating realistic data samples matching the distribution of the training data and have been shown to produce highly convincing synthetic images. This work proposes a physics-informed generative adversarial framework for the design of funicular shell structures. The approach employs a modified Deep Convolutional Generative Adversarial architecture physically guided by an auxiliary discriminator to generate realistic and structurally efficient shell geometries. Specifically, the model is constrained by the membrane factor to penalize geometries dominated by bending. An additional discriminator is also employed allowing the model to deal with more complex structures. Results show that the developed model is stable and capable of generating physically optimal, previously unseen, funicular shells with smooth forms and high membrane factor distributions.

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 / 1 minor

Summary. The manuscript proposes a physics-informed generative adversarial network (GAN) for designing funicular shell structures for 3D concrete printing. It modifies a deep convolutional GAN architecture with an auxiliary discriminator that penalizes low membrane-factor geometries to enforce pure compression, plus an additional discriminator for complex forms. The central claim is that the resulting model stably generates smooth, previously unseen, physically optimal funicular shells exhibiting high membrane-factor distributions.

Significance. If the central claim holds under rigorous validation, the work would offer a novel data-driven route to generalizing the funicular concept from 2D catenaries to arbitrary 3D shells, potentially enabling more efficient unreinforced printed structures. The integration of an independent structural-mechanics constraint (membrane factor) into the GAN training loop is a constructive step, but the absence of quantitative benchmarks, baselines, or finite-element corroboration currently limits the assessed significance.

major comments (2)
  1. [Abstract] Abstract: the assertion that the model generates 'physically optimal, previously unseen, funicular shells with ... high membrane factor distributions' is unsupported by any reported quantitative metrics, baseline comparisons, error analysis, or validation examples against full shell theory.
  2. [Results] Results (implied by abstract claims): the membrane factor is used as a scalar proxy to penalize bending, yet no verification is described that high values guarantee local equilibrium, zero transverse shear, or negligible bending moments on non-developable 3D surfaces generated by the GAN; the mapping can break for rapidly varying curvatures or non-thin shells.
minor comments (1)
  1. [Abstract] Abstract: the role and architecture of the 'additional discriminator' for complex structures is mentioned but not specified, leaving the training procedure unclear.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive and detailed feedback. The comments highlight important aspects of validation and theoretical grounding that we address point-by-point below. We have revised the manuscript to incorporate additional quantitative support, baseline comparisons, and explicit discussion of assumptions and limitations while preserving the core contribution of the physics-informed GAN framework.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the assertion that the model generates 'physically optimal, previously unseen, funicular shells with ... high membrane factor distributions' is unsupported by any reported quantitative metrics, baseline comparisons, error analysis, or validation examples against full shell theory.

    Authors: We agree that the abstract claims benefit from more explicit quantitative backing. In the revised version we have added mean and standard-deviation membrane-factor statistics across large batches of generated samples, direct side-by-side histograms comparing the physics-informed model to a standard DCGAN baseline (showing consistently higher MF values), and additional visual examples with overlaid membrane-factor color maps. We have also clarified that “physically optimal” is used in the specific sense of the membrane-theory proxy employed during training. Full finite-element verification against complete shell theory for every sample was not performed in the original study; we therefore qualify the claim accordingly and note this as a direction for future work. revision: partial

  2. Referee: [Results] Results (implied by abstract claims): the membrane factor is used as a scalar proxy to penalize bending, yet no verification is described that high values guarantee local equilibrium, zero transverse shear, or negligible bending moments on non-developable 3D surfaces generated by the GAN; the mapping can break for rapidly varying curvatures or non-thin shells.

    Authors: The membrane factor is introduced as a scalar derived from the ratio of membrane to total strain energy under the assumptions of classical thin-shell (Kirchhoff–Love) theory. For the smooth, thin geometries targeted by 3D concrete printing, elevated values indicate that in-plane forces dominate and bending moments remain small. We acknowledge that the proxy is not universally exact: transverse shear, local equilibrium violations, or significant bending can appear on non-developable surfaces with rapid curvature changes or when the thin-shell hypothesis is violated. The revised manuscript now contains an expanded discussion of these limitations, explicit statement of the modeling assumptions, and references to the relevant shell-theory literature. We have also added a forward-looking statement that comprehensive finite-element corroboration lies beyond the present generative-modeling scope. revision: yes

Circularity Check

0 steps flagged

No significant circularity; physics-informed constraint drawn from independent mechanics

full rationale

The paper's central derivation employs a standard DCGAN architecture augmented by an auxiliary discriminator that applies the membrane factor as an external penalty term drawn from classical shell theory. This factor is not fitted to or defined by the GAN outputs, nor does any equation reduce the optimality claim to a self-referential fit, renaming, or self-citation chain. The generation of new shell geometries remains an open mapping from noise to shapes, with the physical guidance supplied by an independent structural proxy rather than by construction from the model's own predictions. No load-bearing step collapses to tautology.

Axiom & Free-Parameter Ledger

1 free parameters · 2 axioms · 0 invented entities

The central claim depends on standard GAN training assumptions plus the structural mechanics premise that membrane factor reliably indicates funicular performance; no new entities are postulated.

free parameters (1)
  • membrane factor penalty weight
    Hyperparameter balancing the physics loss term against adversarial losses, selected during training to enforce compression dominance.
axioms (2)
  • domain assumption Membrane theory provides a valid scalar proxy for funicular (bending-free) behavior in thin 3D shells
    Invoked to define the auxiliary discriminator objective and to interpret high membrane factor as structural optimality.
  • domain assumption Constrained GANs converge to distributions of physically valid shell geometries
    Underlying the expectation that the generator will produce stable, smooth, high-membrane-factor outputs.

pith-pipeline@v0.9.0 · 5542 in / 1383 out tokens · 67564 ms · 2026-05-10T06:51:54.581788+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

46 extracted references · 35 canonical work pages · 6 internal anchors

  1. [1]

    Hooke, A description of helioscopes, and some other instruments, printed by T.R

    R. Hooke, A description of helioscopes, and some other instruments, printed by T.R. for John Martyn, 1676.doi:10.3931/e-rara-2171

    work page doi:10.3931/e-rara-2171

  2. [2]

    Hill, The rational mechanics of flexible or elastic bodies 1638–1788: C

    R. Hill, The rational mechanics of flexible or elastic bodies 1638–1788: C. truesdell: Introduction to leonhardi euleri opera omnia, vol. x et xi. seriei seeundae. orell füssli turiei, switzerland, 1960. 435 pp, Journal of the Mechanics and Physics of Solids 10 (2) (1962) 184.doi:10.1016/0022-5096(62) 90022-4

    work page doi:10.1016/0022-5096(62 1960

  3. [3]

    Block, M

    P. Block, M. DeJong, J. Ochsendorf, As hangs the flexible line: Equilibrium of masonry arches, Nexus Network Journal 8 (2006) 13–24.doi:10.1007/s00004-006-0015-9

    work page doi:10.1007/s00004-006-0015-9 2006

  4. [4]

    Block, M

    P. Block, M. DeJong, J. Ochsendorf, As hangs the flexible line: Equilibrium of masonry arches, Nexus Network Journal 8 (2) (2006) 13–24

    2006

  5. [5]

    Adriaenssens, P

    S. Adriaenssens, P. Block, D. Veenendaal, C. Williams, Shell structures for architecture: form finding and optimization, Routledge, 2014

    2014

  6. [6]

    F. Bos, R. Wolfs, Z. Ahmed, T. Salet, Additive manufacturing of concrete in construction: potentials and challenges of 3d concrete printing, Virtual and physical prototyping 11 (3) (2016) 209–225.doi: 10.1080/17452759.2016.1209867

    work page doi:10.1080/17452759.2016.1209867 2016

  7. [7]

    Mader, M

    T. Mader, M. Schreter-Fleischhacker, O. Shkundalova, M. Neuner, G. Hofstetter, Constitutive modeling of orthotropic nonlinear mechanical behavior of hardened 3d printed concrete, Acta Mechanica 234 (2023) 5893–5918.doi:10.1007/s00707-023-03706-z

    work page doi:10.1007/s00707-023-03706-z 2023

  8. [8]

    Bhooshan, V

    S. Bhooshan, V. Bhooshan, A. Dell’Endice, J. Chu, P. Singer, J. Megens, T. Van Mele, P. Block, The striatus bridge: Computational design and robotic fabrication of an unreinforced, 3d-concrete-printed, masonry arch bridge, Architecture, structures and construction 2 (4) (2022) 521–543

    2022

  9. [9]

    Chilton, H

    J. Chilton, H. Isler, The engineer’s contribution to contemporary architecture - Heinz Isler, Thomas Telford, 2000

    2000

  10. [10]

    P.Block, J.Ochsendorf, Thrustnetworkanalysis: anewmethodologyforthree-dimensionalequilibrium, Journal of the International Association for shell and spatial structures 48 (3) (2007) 167–173

    2007

  11. [11]

    Olivieri, L

    C. Olivieri, L. Sibille, S. Cocking, R. N. Charandabi, F. Fabbrocino, S. Adriaenssens, Design of purely compressive shells under vertical and horizontal loads through machine learning-driven form-finding, Mechanics Research Communications 148 (2025) 104491.doi:10.1016/j.mechrescom.2025.104491

    work page doi:10.1016/j.mechrescom.2025.104491 2025

  12. [12]

    Pastrana, E

    R. Pastrana, E. Medina, I. M. de Oliveira, S. Adriaenssens, R. P. Adams, Real-time design of ar- chitectural structures with differentiable mechanics and neural networks (2025).arXiv:2409.02606, doi:10.48550/arXiv.2409.02606

    work page doi:10.48550/arxiv.2409.02606 2025

  13. [13]

    I.J.Goodfellow, J.Pouget-Abadie, M.Mirza, B.Xu, D.Warde-Farley, S.Ozair, A.Courville, Y.Bengio, Generative adversarial networks (2014).arXiv:1406.2661,doi:10.48550/arXiv.1406.2661

    work page internal anchor Pith review doi:10.48550/arxiv.1406.2661 2014

  14. [14]

    J. Wu, C. Zhang, T. Xue, W. T. Freeman, J. B. Tenenbaum, Learning a probabilistic latent space of object shapes via 3d generative-adversarial modeling (2017).arXiv:1610.07584,doi:10.48550/ arXiv.1610.07584

    work page arXiv 2017

  15. [15]

    W. Li, C. Wang, H. Guan, J. Wang, J. Yang, C. Zhang, D. Tao, Generative adversarial physics-informed neural networks for solving forward and inverse problem with small labeled samples, Computers & Mathematics with Applications 183 (2025) 98–120.doi:10.1016/j.camwa.2025.01.025

    work page doi:10.1016/j.camwa.2025.01.025 2025

  16. [16]

    Smith, D

    E. Smith, D. Meger, Improved adversarial systems for 3d object generation and reconstruction (2017). arXiv:1707.09557,doi:10.48550/arXiv.1707.09557. 26

    work page doi:10.48550/arxiv.1707.09557 2017

  17. [17]

    Chen, Guanjie, Z

    C. Chen, Guanjie, Z. H. Wei, Z. Li, Physics-informed generative adversarial networks for sequence generation with limited data, in: Interpretable Inductive Biases and Physically Structured Learning, 2020. URLhttps://inductive-biases.github.io/papers/7.pdf

    2020

  18. [18]

    A. Daw, M. Maruf, A. Karpatne, Pid-gan: A gan framework based on a physics-informed discriminator for uncertainty quantification with physics, in: Proceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery & Data Mining, KDD ’21, Association for Computing Machinery, 2021, p. 237–247.doi:10.1145/3447548.3467449

    work page doi:10.1145/3447548.3467449 2021

  19. [19]

    Ciftci, K

    K. Ciftci, K. Hackl, A physics-informed gan framework based on model-free data-driven computational mechanics, Computer Methods in Applied Mechanics and Engineering 424 (2024) 116907.doi:10. 1016/j.cma.2024.116907

    work page arXiv 2024

  20. [20]

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

    M. Raissi, P. Perdikaris, G. Karniadakis, Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations, Journal of Computational Physics 378 (2019) 686–707.doi:10.1016/j.jcp.2018.10.045

    work page doi:10.1016/j.jcp.2018.10.045 2019

  21. [21]

    J.-H.Bastek, D.M.Kochmann, Physics-informedneuralnetworksforshellstructures, EuropeanJournal of Mechanics - A/Solids 97 (2023) 104849.doi:10.1016/j.euromechsol.2022.104849

    work page doi:10.1016/j.euromechsol.2022.104849 2023

  22. [22]

    T. D. Nguyen, T. Le, H. Vu, D. Phung, Dual discriminator generative adversarial nets (2017).arXiv: 1709.03831,doi:10.48550/arXiv.1709.03831

    work page doi:10.48550/arxiv.1709.03831 2017

  23. [23]

    Theory and analysis of elastic plates and shells, CRC press, 1999

    1999

  24. [24]

    A.C.Ugural, StressesinBeams, PlatesandShells, 3rdEdition, CRCPress, 2009.doi:10.1201/b17516

    work page doi:10.1201/b17516 2009

  25. [25]

    Ventsel, T

    E. Ventsel, T. Krauthammer, Thin plates and shells: Theory, analysis, and applications, Applied Mechanics Reviews 55 (4) (2002) B72–B73.doi:10.1115/1.1483356

    work page doi:10.1115/1.1483356 2002

  26. [26]

    J. N. Reddy, Theory and Analysis of Elastic Plates and Shells, 2nd Edition, CRC Press, 2006.doi: 10.1201/9780849384165

    work page doi:10.1201/9780849384165 2006

  27. [27]

    Improved Techniques for Training GANs

    T. Salimans, I. Goodfellow, W. Zaremba, V. Cheung, A. Radford, X. Chen, Improved techniques for training gans (2016).arXiv:1606.03498,doi:10.48550/arXiv.1606.03498

    work page Pith review doi:10.48550/arxiv.1606.03498 2016

  28. [28]

    Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks

    A. Radford, L. Metz, S. Chintala, Unsupervised representation learning with deep convolutional gen- erative adversarial networks (2016).arXiv:1511.06434,doi:10.48550/arXiv.1511.06434

    work page internal anchor Pith review doi:10.48550/arxiv.1511.06434 2016

  29. [29]

    Spectral Normalization for Generative Adversarial Networks

    T. Miyato, T. Kataoka, M. Koyama, Y. Yoshida, Spectral normalization for generative adversarial networks (2018).arXiv:1802.05957,doi:/10.48550/arXiv.1802.05957

    work page Pith review doi:10.48550/arxiv.1802.05957 2018

  30. [30]

    Wasserstein GAN

    M. Arjovsky, S. Chintala, L. Bottou, Wasserstein gan (2017).arXiv:1701.07875,doi:10.48550/ arXiv.1701.07875

    work page Pith review arXiv 2017

  31. [31]

    Z. Lin, V. Sekar, G. Fanti, Why spectral normalization stabilizes gans: Analysis and improvements (2021).arXiv:2009.02773,doi:10.48550/arXiv.2009.02773

    work page doi:10.48550/arxiv.2009.02773 2021

  32. [32]

    URLhttp://www.blender.org

    Blender Online Community, Blender - a 3D modelling and rendering package, Blender Foundation, Blender Institute, Amsterdam (2026). URLhttp://www.blender.org

    2026

  33. [33]

    Smith, ABAQUS/Standard User’s Manual, Version 6.11, United States (2009)

    M. Smith, ABAQUS/Standard User’s Manual, Version 6.11, United States (2009)

    2009

  34. [34]

    Osher, R

    S. Osher, R. Fedkiw, K. Piechor, Level set methods and dynamic implicit surfaces, Appl. Mech. Rev. 57 (3) (2004) B15–B15.doi:10.1115/1.1760520. 27

    work page doi:10.1115/1.1760520 2004

  35. [35]

    U-Net: Convolutional Networks for Biomedical Image Segmentation

    O. Ronneberger, P. Fischer, T. Brox, U-net: Convolutional networks for biomedical image segmentation (2015).arXiv:1505.04597,doi:10.48550/arXiv.1505.04597

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.1505.04597 2015

  36. [36]

    L. Sun, H. Gao, S. Pan, J.-X. Wang, Surrogate modeling for fluid flows based on physics-constrained deep learning without simulation data, Computer Methods in Applied Mechanics and Engineering 361 (2020) 112732.doi:10.1016/j.cma.2019.112732

    work page doi:10.1016/j.cma.2019.112732 2020

  37. [37]

    L. Guo, S. Ye, J. Han, H. Zheng, H. Gao, D. Z. Chen, J.-X. Wang, C. Wang, Ssr-vfd: Spatial super- resolution for vector field data analysis and visualization, in: 2020 IEEE Pacific Visualization Sympo- sium (PacificVis), 2020, pp. 71–80.doi:10.1109/PacificVis48177.2020.8737

    work page doi:10.1109/pacificvis48177.2020.8737 2020

  38. [38]

    Z. Gao, L. Zhao, J. Zheng, T. Yi, C. Zhang, Q. Wang, Principal component guided dcgan for mitigating mode collapse in electromagnetic pulse signal synthesis, Scientific Reports 15 (2025) 36858.doi: 10.1038/s41598-025-20962-2

    work page doi:10.1038/s41598-025-20962-2 2025

  39. [39]

    Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift

    S. Ioffe, C. Szegedy, Batch normalization: Accelerating deep network training by reducing internal covariate shift (2015).arXiv:1502.03167,doi:10.48550/arXiv.1502.03167

    work page internal anchor Pith review doi:10.48550/arxiv.1502.03167 2015

  40. [40]

    Despotovic, V

    M. Despotovic, V. Nedic, D. Despotovic, S. Cvetanovic, Evaluation of empirical models for predicting monthly mean horizontal diffuse solar radiation, Renewable and Sustainable Energy Reviews 56 (2016) 246–260.doi:10.1016/j.rser.2015.11.058

    work page doi:10.1016/j.rser.2015.11.058 2016

  41. [41]

    Smith, Edoardo M

    M. Heusel, H. Ramsauer, T. Unterthiner, B. Nessler, S. Hochreiter, Gans trained by a two time-scale update rule converge to a local nash equilibrium (2018).arXiv:1706.08500,doi:10.48550/arXiv. 1706.08500

    work page internal anchor Pith review doi:10.48550/arxiv 2018

  42. [42]

    Rethinking the Inception Architecture for Computer Vision

    C. Szegedy, V. Vanhoucke, S. Ioffe, J. Shlens, Z. Wojna, Rethinking the inception architecture for computer vision (2015).arXiv:1512.00567,doi:10.48550/arXiv.1512.00567

    work page Pith review doi:10.48550/arxiv.1512.00567 2015

  43. [43]

    J. Deng, W. Dong, R. Socher, L.-J. Li, K. Li, L. Fei-Fei, Imagenet: A large-scale hierarchical image database, in: 2009 IEEE Conference on Computer Vision and Pattern Recognition, 2009, pp. 248–255. doi:10.1109/CVPR.2009.5206848

    work page doi:10.1109/cvpr.2009.5206848 2009

  44. [44]

    van der Maaten, G

    L. van der Maaten, G. Hinton, Visualizing data using t-sne, Journal of Machine Learning Research 9 (86) (2008) 2579–2605. URLhttp://jmlr.org/papers/v9/vandermaaten08a.html

    2008

  45. [45]

    UMAP: Uniform Manifold Approximation and Projection for Dimension Reduction

    L. McInnes, J. Healy, J. Melville, Umap: Uniform manifold approximation and projection for dimension reduction (2020).arXiv:1802.03426,doi:10.48550/arXiv.1802.03426

    work page internal anchor Pith review doi:10.48550/arxiv.1802.03426 2020

  46. [46]

    Scholkopf, C

    B. Scholkopf, C. Burges, A. Smola, Kernel principal component analysis, in: Advances in Kernel Methods – Support Vector Learning, MIT Press, 1999, pp. 327–352. 28

    1999