pith. sign in

arxiv: 2605.00732 · v1 · submitted 2026-05-01 · ❄️ cond-mat.stat-mech

Reconstruction of spin structures from topological charge distributions via generative neural network systems

Kyra H. M. Klos , Jan Disselhoff , Michael Wand , Karin Everschor-Sitte , Friederike Schmid This is my paper

Pith reviewed 2026-05-09 18:24 UTC · model grok-4.3

classification ❄️ cond-mat.stat-mech
keywords generative adversarial networksXY modeltopological defectsKosterlitz-Thouless transitionspin configurationsWasserstein GANtopological data analysisvortex pairs
0
0 comments X

The pith

An extended Wasserstein GAN generates microscopic spin configurations from prescribed topological charge patterns that match key thermodynamic observables in the 2D XY model below the Kosterlitz-Thouless transition.

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

The paper develops a method to generate detailed microscopic spin arrangements starting from higher-level patterns of topological defects. It augments a Wasserstein generative adversarial network with physical constraints and Fourier-space information so that the output spin fields remain consistent with given vortex distributions and temperature. The resulting configurations reproduce magnetization, susceptibility, helicity modulus, and spin-spin correlations over a wide temperature range below the transition point. Discrepancies appear in the specific heat, showing that energy fluctuations are not fully captured. Topological data analysis further detects subtle global differences in spin correlations near criticality that standard measures miss.

Core claim

The paper claims that extending a Wasserstein generative adversarial network with physical constraints and Fourier-space information produces microscopic spin configurations consistent with prescribed macroscopic topological charge distributions and thermodynamic parameters in the two-dimensional XY model. These generated configurations accurately reproduce magnetization, susceptibility, helicity modulus, and spin-spin correlations over a wide range of temperatures below the Kosterlitz-Thouless transition. Deviations in specific heat indicate limits in reproducing higher-order energy fluctuations, while topological data analysis reveals subtle differences in global spin-correlation structure

What carries the argument

The physically augmented Wasserstein GAN that incorporates physical constraints and Fourier-space information to map macroscopic topological charge distributions onto consistent microscopic spin fields.

If this is right

  • The generated configurations yield reliable values for magnetization, susceptibility, helicity modulus, and spin-spin correlations across temperatures below the transition.
  • The approach enables studies of defect patterns without running full microscopic simulations for each case.
  • Topological data analysis identifies global structural differences near criticality that conventional correlation functions overlook.
  • Deviations in specific heat show that higher-order energy fluctuations remain incompletely reproduced.

Where Pith is reading between the lines

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

  • The method could be applied to other systems dominated by topological defects, such as magnetic films or liquid crystals, to link coarse defect maps to microscopic detail.
  • Additional constraints on energy moments or fluctuation spectra may be needed to close the gap in specific heat.
  • Hybrid workflows that combine defect-based coarse descriptions with generative refinement could reduce the cost of exploring large-scale critical behavior.
  • Validation with topological data analysis may become a standard check when generative models are used near phase transitions.

Load-bearing premise

That adding physical constraints and Fourier-space information to the generative model is enough to make the output spin configurations reproduce the full statistical mechanics of the underlying system.

What would settle it

Compute the specific heat from the generated configurations and compare it directly to exact Monte Carlo results for the XY model at temperatures approaching the Kosterlitz-Thouless transition from below.

Figures

Figures reproduced from arXiv: 2605.00732 by Friederike Schmid, Jan Disselhoff, Karin Everschor-Sitte, Kyra H. M. Klos, Michael Wand.

Figure 1
Figure 1. Figure 1: FIG. 1. Example of a vortex-antivortex pair (blue and red view at source ↗
Figure 2
Figure 2. Figure 2: FIG. 2. Construction of cubical complexes view at source ↗
Figure 3
Figure 3. Figure 3: FIG. 3. Construction of persistance diagrams and persistent view at source ↗
Figure 5
Figure 5. Figure 5: FIG. 5. Architecture of the generator network view at source ↗
Figure 6
Figure 6. Figure 6: FIG. 6. Architecture of the critic network view at source ↗
Figure 7
Figure 7. Figure 7: FIG. 7. Training progress of generator (blue) compared to view at source ↗
Figure 9
Figure 9. Figure 9: FIG. 9. Physical observables versus defect pair distance view at source ↗
Figure 10
Figure 10. Figure 10: FIG. 10. Spin-spin correlation function view at source ↗
Figure 12
Figure 12. Figure 12: FIG. 12. Geometrical, topological, and graph-based descriptors calculated from generated (blue) and simulated (red) data: view at source ↗
Figure 13
Figure 13. Figure 13: FIG. 13. Mean difference ∆ view at source ↗
Figure 14
Figure 14. Figure 14: FIG. 14. Pearson correlation coefficient (PCC) matrix capturing the linear correlation between the means and variances of the view at source ↗
Figure 15
Figure 15. Figure 15: FIG. 15. Histograms of energy view at source ↗
Figure 16
Figure 16. Figure 16: FIG. 16. Impact of Fourier critic on the quality of generated spin configurations. The graphs compare physical observables view at source ↗
Figure 17
Figure 17. Figure 17: FIG. 17. Impact of Fourier critic on the quality of generated spin configurations. The graphs compare spin pair correlation view at source ↗
Figure 18
Figure 18. Figure 18: FIG. 18. Spin-spin correlation functions in the full temperature range for defect field distances view at source ↗
Figure 19
Figure 19. Figure 19: FIG. 19. Geometrical and topological measures as described in Section view at source ↗
Figure 20
Figure 20. Figure 20: FIG. 20. Scalability of the network architecture with respect to system size. The graphs show physical observables versus mean view at source ↗
Figure 21
Figure 21. Figure 21: FIG. 21. Scalability of the network architecture with respect to system size. The graphs show spin-spin correlation functions view at source ↗
Figure 22
Figure 22. Figure 22: FIG. 22. Scalability of the network architecture with respect to defect number. The graphs show physical observables versus view at source ↗
Figure 23
Figure 23. Figure 23: FIG. 23. Transferability of the network architecture to higher defect numbers. The graphs show physical observables versus view at source ↗
Figure 24
Figure 24. Figure 24: FIG. 24. Exemplary spin configurations generated through generalization of the WGAN network presented in the main article, view at source ↗
read the original abstract

Localized topological defects inherently possess a multiscale character. While their microstructure configuration depends on the specific physical system, their topological features and mutual interactions can be described on the macroscale in terms of a particle representation. However, determining the physical properties associated with a given defect pattern often requires knowledge of the underlying microscopic structure. In this work, we extend a Wasserstein generative adversarial neural network by incorporating physical constraints and Fourier-space information to generate microscopic spin configurations consistent with prescribed macroscopic patterns and thermodynamic parameters. Using the two-dimensional XY model as a test case, where vortex-antivortex pairs act as long-range interacting defects, we show that the model generates spin configurations that accurately reproduce magnetization, susceptibility, helicity modulus, and spin-spin correlations over a wide range of temperatures below the Kosterlitz-Thouless transition. At the same time, deviations in the specific heat reveal limitations in reproducing higher order energy fluctuations. A complementary analysis based on topological data analysis uncovers subtle differences in global spin-correlation structures at near critical temperatures that are not apparent from conventional correlation functions alone. These results demonstrate both the promise and current limitations of generative approaches for multiscale studies of defect-dominated spin systems and at the same time highlight topological methods as valuable tools for characterizing critical behavior.

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 manuscript extends a Wasserstein GAN by adding physical constraints and Fourier-space information to generate microscopic spin configurations from prescribed topological charge distributions. Using the 2D XY model as a testbed, it claims that the generated configurations accurately reproduce magnetization, susceptibility, helicity modulus, and spin-spin correlations over a wide temperature range below the Kosterlitz-Thouless transition, while deviations appear in the specific heat and topological data analysis detects subtle global correlation differences near criticality that standard functions miss.

Significance. If the empirical results hold under rigorous validation, the work demonstrates a viable route for multiscale reconstruction in defect-dominated spin systems by linking macroscopic topological patterns to microscopic degrees of freedom. The explicit incorporation of physical constraints and the complementary TDA analysis are strengths that could generalize to other systems with long-range interacting defects. The acknowledged limitations in higher-order fluctuations also usefully bound the current applicability of such generative approaches.

major comments (2)
  1. [Abstract] Abstract: the central claim of accurate reproduction is scoped to magnetization, susceptibility, helicity modulus, and spin-spin correlations, yet the explicit deviations in specific heat (a direct probe of energy fluctuations) indicate that the added constraints and Fourier information do not fully enforce the underlying statistical mechanics. This is load-bearing because the paper positions the method as generating configurations 'consistent with prescribed macroscopic patterns and thermodynamic parameters'; a quantitative assessment (e.g., relative errors or distribution overlaps) of how well the matched quantities agree is needed to substantiate the claim.
  2. [Results] Results section (TDA analysis): the abstract states that TDA uncovers subtle differences in global spin-correlation structures near criticality not apparent from conventional correlations. Without reported TDA metrics, persistence diagrams, or statistical significance tests comparing generated versus reference configurations, it is difficult to judge whether these differences are physically meaningful or artifacts of the generative process.
minor comments (2)
  1. Methods: training details, network architecture, loss weighting for physical constraints, and the precise form of the Fourier-space information should be expanded with sufficient specificity for independent reproduction.
  2. Figures and tables: all comparisons of thermodynamic quantities should include error bars or bootstrap uncertainties derived from multiple generated ensembles to allow readers to assess the degree of agreement.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their careful reading of the manuscript, positive assessment of its significance, and constructive comments. We address each major comment below and indicate the revisions we will make.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the central claim of accurate reproduction is scoped to magnetization, susceptibility, helicity modulus, and spin-spin correlations, yet the explicit deviations in the specific heat (a direct probe of energy fluctuations) indicate that the added constraints and Fourier information do not fully enforce the underlying statistical mechanics. This is load-bearing because the paper positions the method as generating configurations 'consistent with prescribed macroscopic patterns and thermodynamic parameters'; a quantitative assessment (e.g., relative errors or distribution overlaps) of how well the matched quantities agree is needed to substantiate the claim.

    Authors: We agree that quantitative metrics would strengthen the presentation of the results for the matched observables. The abstract already scopes the accuracy claim to magnetization, susceptibility, helicity modulus, and spin-spin correlations while explicitly noting deviations in specific heat; this scoping is intentional to bound the method's applicability. In the revised manuscript we will add explicit quantitative assessments in the results section, including relative errors (as percentages) for the thermodynamic quantities across temperatures and distribution-overlap measures (such as Jensen-Shannon divergence) for the spin-spin correlation functions, thereby substantiating the agreement without overstating the enforcement of all statistical-mechanical relations. revision: yes

  2. Referee: [Results] Results section (TDA analysis): the abstract states that TDA uncovers subtle differences in global spin-correlation structures near criticality not apparent from conventional correlations. Without reported TDA metrics, persistence diagrams, or statistical significance tests comparing generated versus reference configurations, it is difficult to judge whether these differences are physically meaningful or artifacts of the generative process.

    Authors: We acknowledge that the TDA discussion would be more rigorous with quantitative support. The manuscript currently presents the TDA findings qualitatively to highlight that global structural differences near criticality are visible in persistence diagrams but not in conventional two-point correlations. In the revision we will include representative persistence diagrams for both generated and reference ensembles, report quantitative TDA metrics (e.g., bottleneck distances between diagrams), and add statistical significance tests (permutation or bootstrap procedures) to establish that the observed differences are not artifacts of the generative process. revision: yes

Circularity Check

0 steps flagged

No significant circularity

full rationale

The paper describes an empirical, data-driven workflow: a Wasserstein GAN augmented with physical constraints and Fourier information is trained on XY-model configurations to map topological charge patterns to microscopic spins. The reported results consist of numerical comparisons showing that generated samples reproduce magnetization, susceptibility, helicity modulus and two-point correlations below the KT transition, while explicitly noting deviations in specific heat and subtle TDA differences near criticality. No algebraic derivation, parameter fit, or uniqueness theorem is invoked whose output is then relabeled as a prediction; the performance metrics are obtained by direct evaluation on held-out Monte Carlo data and are therefore independent of the training procedure itself. The work contains no self-citation load-bearing steps or ansatz smuggling.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on the assumption that a generative network trained on XY-model data can learn a mapping from macroscopic topological charge distributions and thermodynamic parameters to microscopic spin fields whose statistics match the true Boltzmann distribution for the chosen observables.

axioms (1)
  • domain assumption Vortex-antivortex pairs act as long-range interacting defects whose macroscopic pattern determines the relevant thermodynamic behavior below the Kosterlitz-Thouless transition.
    Invoked as the test case for the generative model.

pith-pipeline@v0.9.0 · 5538 in / 1463 out tokens · 46901 ms · 2026-05-09T18:24:27.958474+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

64 extracted references · 1 canonical work pages

  1. [1]

    Effective field theory,

    H. Georgi, “Effective field theory,”Annual Reviews Nu- clear Particle Science., vol. 43, pp. 209–252, 1993

    1993

  2. [2]

    Exploring topological defects in epitaxial BiFeO3 thin films,

    R. K. Vasudevan, Y.-C. Chen, H.-H. Tai, N. Balke, P. Wu, S. Bhattacharya, L. Q. Chen, Y.-H. Chu, I.- N. Lin, S. V. Kalinin, and V. Nagarajan, “Exploring topological defects in epitaxial BiFeO3 thin films,”ACS Nano, vol. 5, no. 2, pp. 879–887, 2011. PMID: 21214267

    2011

  3. [3]

    Skyrmion lattice in a chiral magnet,

    S. M¨ uhlbauer, B. Binz, F. Jonietz, C. Pfleiderer, A. Rosch, A. Neubauer, R. Georgii, and P. B¨ oni, “Skyrmion lattice in a chiral magnet,”Science, vol. 323, no. 5916, pp. 915–919, 2009

    2009

  4. [4]

    Topological point defects in nematic liquid crystals,

    M. Kleman and O. D. Lavrentovich, “Topological point defects in nematic liquid crystals,”Philosophical Maga- zine, vol. 86, no. 25-26, pp. 4117–4137, 2006

    2006

  5. [5]

    Topological defects in cholesteric liquid crystal shells,

    A. Darmon, M. Benzaquen, S. ˇCopar, O. Dauchot, and T. Lopez-Leon, “Topological defects in cholesteric liquid crystal shells,”Soft Matter, vol. 12, pp. 9280–9288, 2016

    2016

  6. [6]

    Command of active matter by topological defects and patterns,

    C. Peng, T. Turiv, Y. Guo, Q.-H. Wei, and O. D. Lavren- tovich, “Command of active matter by topological defects and patterns,”Science, vol. 354, no. 6314, pp. 882–885, 2016

    2016

  7. [7]

    Topological active matter,

    S. Shankar, A. Souslov, M. J. Bowick, M. C. Marchetti, and V. Vitelli, “Topological active matter,”Nature Re- views Physics, vol. 4, pp. 380–398, 2022

    2022

  8. [8]

    Perspective on uncon- ventional computing using magnetic skyrmions,

    O. Lee, R. Msiska, M. A. Brems, M. Kl¨ aui, H. Kure- bayashi, and K. Everschor-Sitte, “Perspective on uncon- ventional computing using magnetic skyrmions,”Applied Physics Letters, vol. 122, p. 260501, 06 2023

    2023

  9. [9]

    Topological defects and phase transitions,

    J.Kosterlitz, “Topological defects and phase transitions,” Review of Modern Physics, vol. 89, 2017

    2017

  10. [10]

    Topological point defects of liquid crystals in quasi-two-dimensional geometries,

    K. Harth and R. Stannarius, “Topological point defects of liquid crystals in quasi-two-dimensional geometries,” Frontiers in Physics, vol. 8, 2020

    2020

  11. [11]

    Theory of defect motion in 2d passive and active nematic liquid crystals,

    X. Tang and J. V. Selinger, “Theory of defect motion in 2d passive and active nematic liquid crystals,”Soft Matter, vol. 15, pp. 587–601, 2019

    2019

  12. [12]

    Realizing quan- titative quasiparticle modeling of skyrmion dynamics in arbitrary potentials,

    M. A. Brems, T. Sparmann, S. M. Fr¨ ohlich, L.-C. Dany, J. Roth¨ orl, F. Kammerbauer, E. M. Jefremovas, O. Farago, M. Kl¨ aui, and P. Virnau, “Realizing quan- titative quasiparticle modeling of skyrmion dynamics in arbitrary potentials,”Physical Review Letters, vol. 134, p. 046701, Jan 2025

    2025

  13. [13]

    Steady-state motion of magnetic do- mains,

    A. A. Thiele, “Steady-state motion of magnetic do- mains,”Physical Review Letters, vol. 30, pp. 230–233, Feb 1973

    1973

  14. [14]

    End-to-end machine learn- ing for experimental physics: using simulated data to train a neural network for object detection in video mi- croscopy,

    E. N. Minor, S. D. Howard, A. A. S. Green, M. A. Glaser, C. S. Park, and N. A. Clark, “End-to-end machine learn- ing for experimental physics: using simulated data to train a neural network for object detection in video mi- croscopy,”Soft Matter, vol. 16, pp. 1751–1759, 2020

    2020

  15. [15]

    Machine eye for defects: Machine learning-based solution to identify and characterize topological defects in textured images of nematic materials,

    H. Ren, W. Wang, W. Tang, and R. Zhang, “Machine eye for defects: Machine learning-based solution to identify and characterize topological defects in textured images of nematic materials,”Physical Review Research, vol. 6, p. 013259, Mar 2024

    2024

  16. [16]

    Machine learning vortices at the kosterlitz-thouless transition,

    M. J. Beach, A. Golubeva, and R. G. Melko, “Machine learning vortices at the kosterlitz-thouless transition,” Physical Review B, vol. 97, 2018

    2018

  17. [17]

    Machine learn- ing topological defects of confined liquid crystals in two dimensions,

    M. Walters, Q. Wei, and J. Z. Y. Chen, “Machine learn- ing topological defects of confined liquid crystals in two dimensions,”Physical Review E, vol. 99, p. 062701, Jun 2019

    2019

  18. [18]

    A machine learn- ing approach to robustly determine director fields and analyze defects in active nematics,

    Y. Li, Z. Zarei, P. N. Tran, Y. Wang, A. Baskaran, S. Fraden, M. F. Hagan, and P. Hong, “A machine learn- ing approach to robustly determine director fields and analyze defects in active nematics,”Soft Matter, vol. 20, pp. 1869–1883, 2024

    2024

  19. [19]

    A deeper look into natural sciences with physics-based and data-driven measures,

    D. R. Rodrigues, K. Everschor-Sitte, S. Gerber, and I. Horenko, “A deeper look into natural sciences with physics-based and data-driven measures,”IScience, vol. 24, p. 102171, MAR 19 2021

    2021

  20. [20]

    Scalable computational measures for entropic de- tection of latent relations and their applications to mag- netic imaging,

    I. Horenko, D. Rodrigues, T. O’Kane, and K. Everschore- Sitte, “Scalable computational measures for entropic de- tection of latent relations and their applications to mag- netic imaging,”Communications in Applied Mathematics and Computational Science, vol. 16, no. 2, pp. 266–297, 2021

    2021

  21. [22]

    Semi-supervised learn- ing of order parameter in 2D Ising and XY mod- els using conditional variational autoencoders,

    A. Naravane and N. Mathur, “Semi-supervised learn- ing of order parameter in 2D Ising and XY mod- els using conditional variational autoencoders,”ArXiv, vol. abs/1705.09524, 6 2023

    work page arXiv 2023

  22. [23]

    Learning thermodynamics and topological order of the two-dimensional XY model with generative real-valued restricted boltzmann machines,

    K. Zhang, “Learning thermodynamics and topological order of the two-dimensional XY model with generative real-valued restricted boltzmann machines,”Physical Re- view B, vol. 111, p. 024112, Jan 2025

    2025

  23. [24]

    Topological magnetic structure generation using VAE- GAN hybrid model and discriminator-driven latent sam- pling,

    S. Park, H. Yoon, D. Lee, J. Choi, H. Kwon, and C. Won, “Topological magnetic structure generation using VAE- GAN hybrid model and discriminator-driven latent sam- pling,”Scientific Reports, vol. 13, Dec. 2023

    2023

  24. [25]

    Ordering, metastability and phase transitions in two-dimensional systems,

    J. Kosterlitz and D. Thouless, “Ordering, metastability and phase transitions in two-dimensional systems,”Jour- 16 nal of Physics C: Solid State Physics, vol. 6, 1973

    1973

  25. [26]

    Destruction of long-range order in one-dimensional and twodimensional systems having a continuous symmetry group i. classical systems.,

    V. L. Berezinskii, “Destruction of long-range order in one-dimensional and twodimensional systems having a continuous symmetry group i. classical systems.,”Jour- nal of Experimental and Theortical Physics, vol. 32, pp. 493–500, 1970

    1970

  26. [27]

    Superfluidity and the two dimensional XY model,

    D. R. Nelson, “Superfluidity and the two dimensional XY model,”Physics Reports, vol. 49, no. 2, pp. 255–259, 1979

    1979

  27. [28]

    P. M. Chaikin and T. C. Lubensky,Principles of Con- densed Matter Physics. Cambridge University Press, 1995

    1995

  28. [29]

    Absence of ferromag- netism or antiferromagnetism in one- or two-dimensional isotropic heisenberg models,

    N. D. Mermin and H. Wagner, “Absence of ferromag- netism or antiferromagnetism in one- or two-dimensional isotropic heisenberg models,”Physical Review Letters, vol. 17, pp. 1133–1136, Nov 1966

    1966

  29. [30]

    On the theory of phase transitions,

    L. D. Landau, “On the theory of phase transitions,” Journal of Experimental and Theoretical Physics, vol. 7, pp. 19–32, 1937

    1937

  30. [31]

    The two-dimensional XY model at the Berezinskii-Kosterlitz-Thouless transition,

    M. Hasenbusch, “The two-dimensional XY model at the Berezinskii-Kosterlitz-Thouless transition,”Journal of Physics A: Mathematical and General, vol. 38, no. 26, pp. 5869–5883, 2005

    2005

  31. [32]

    Helicity modulus, superfluidity, and scaling in isotropic systems,

    M. E. Fisher, M. N. Barber, and D. Jasnow, “Helicity modulus, superfluidity, and scaling in isotropic systems,” Physical Review. A, vol. 8, pp. 1111–1124, Aug 1973

    1973

  32. [33]

    Critical properties from monte carlo coarse graining and renormalization,

    K. Binder, “Critical properties from monte carlo coarse graining and renormalization,”Phys. Rev. Lett., vol. 47, pp. 693–696, Aug 1981

    1981

  33. [34]

    Nakahara,Geometry, topology and physics

    M. Nakahara,Geometry, topology and physics. 2003. Bristol, UK: Hilger (1990) 505 p. (Graduate student se- ries in physics)

    2003

  34. [35]

    Persistent homology—a survey,

    H. Edelsbrunner and J. Harer, “Persistent homology—a survey,”Discrete & Computational Geometry - DCG, vol. 453, 01 2008

    2008

  35. [36]

    Quantita- tive analysis of phase transitions in two-dimensional XY models using persistent homology,

    N. Sale, J. Giansiracusa, and B. Lucini, “Quantita- tive analysis of phase transitions in two-dimensional XY models using persistent homology,”Physical Review. E, vol. 105, p. 024121, Feb 2022

    2022

  36. [37]

    Topological persistence and simplification,

    Edelsbrunner, Letscher, and Zomorodian, “Topological persistence and simplification,”Discrete Computational Geometry, vol. 28, p. 511–533, Nov. 2002

    2002

  37. [38]

    A roadmap for the computation of per- sistent homology,

    N. Otter, M. Porter, U. Tillmann, P. Grindrod, and H. Harrington, “A roadmap for the computation of per- sistent homology,”EPJ Data Science, vol. 6, 06 2015

    2015

  38. [39]

    Persistence images: A sta- ble vector representation of persistent homology,

    H. Adams, T. Emerson, M. Kirby, R. Neville, C. Pe- terson, P. Shipman, S. Chepushtanova, E. Hanson, F. Motta, and L. Ziegelmeier, “Persistence images: A sta- ble vector representation of persistent homology,”Jour- nal of Machine Learning Research, vol. 18, no. 8, pp. 1– 35, 2017

    2017

  39. [40]

    Scikit-tda: Topological data anal- ysis for python,

    C. T. Nathaniel Saul, “Scikit-tda: Topological data anal- ysis for python,” 2019

    2019

  40. [41]

    Robust mor- phological measures for large scale structure in the uni- verse,

    K. R. Mecke, T. Buchert, and H. Wagner, “Robust mor- phological measures for large scale structure in the uni- verse,”Astronomy and Astrophysics, vol. 288, pp. 697– 704, 1994

    1994

  41. [42]

    Neural networks fail to learn periodic functions and how to fix it,

    L. Ziyin, T. Hartwig, and M. Ueda, “Neural networks fail to learn periodic functions and how to fix it,” inAdvances in Neural Information Processing Systems, vol. 33, pp. 1583–1594, Curran Associates, Inc., 2020

    2020

  42. [43]

    Pytorch: An imperative style, high- performance deep learning library,

    A. Paszke, S. Gross, F. Massa, A. Lerer, J. Bradbury, G. Chanan, T. Killeen, Z. Lin, N. Gimelshein, L. Antiga, A. Desmaison, A. Kopf, E. Yang, Z. DeVito, M. Raison, A. Tejani, S. Chilamkurthy, B. Steiner, L. Fang, J. Bai, and S. Chintala, “Pytorch: An imperative style, high- performance deep learning library,” inAdvances in Neu- ral Information Processing...

    2019

  43. [44]

    Wasserstein Gen- erative Adversarial Networks,

    A. Martin, C. Soumith, and B. Leon, “Wasserstein Gen- erative Adversarial Networks,”Proceedings of the 34th International Conference on Machine Learning, PMLR, vol. 70, pp. 214–223, 2017

    2017

  44. [45]

    Generative adversarial nets,

    I. Goodfellow, J. Pouget-Abadie, M. Mirza, B. Xu, D. Warde-Farley, S. Ozair, A. Courville, and Y. Ben- gio, “Generative adversarial nets,” inAdvances in Neu- ral Information Processing Systems(Z. Ghahramani, M. Welling, C. Cortes, N. Lawrence, and K. Weinberger, eds.), vol. 27, Curran Associates, Inc., 2014

    2014

  45. [46]

    Mode collapse in generative adversarial networks: An overview,

    Y. Kossale, M. Airaj, and A. Darouichi, “Mode collapse in generative adversarial networks: An overview,” in 2022 8th International Conference on Optimization and Applications (ICOA), pp. 1–6, 2022

    2022

  46. [47]

    Does diffusion beat GAN in image super reso- lution?,

    D. Kuznedelev, V. Startsev, D. Shlenskii, and S. Kas- tryulin, “Does diffusion beat GAN in image super reso- lution?,” 2024

    2024

  47. [48]

    U-net: Con- volutional networks for biomedical image segmentation,

    O. Ronneberger, P.Fischer, and T. Brox, “U-net: Con- volutional networks for biomedical image segmentation,” inMedical Image Computing and Computer-Assisted In- tervention (MICCAI), vol. 9351 ofLNCS, pp. 234–241, Springer, 2015

    2015

  48. [49]

    Arbitrary style transfer in real-time with adaptive instance normalization,

    X. Huang and S. Belongie, “Arbitrary style transfer in real-time with adaptive instance normalization,” in 2017 IEEE International Conference on Computer Vi- sion (ICCV), pp. 1510–1519, 2017

    2017

  49. [50]

    A style-based genera- tor architecture for generative adversarial networks,

    T. Karras, S. Laine, and T. Aila, “A style-based genera- tor architecture for generative adversarial networks,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp. 4401–4410, 2019

    2019

  50. [51]

    Rectified linear units im- prove restricted boltzmann machines,

    V. Nair and G. E. Hinton, “Rectified linear units im- prove restricted boltzmann machines,” inInternational Conference on Machine Learning, 2010

    2010

  51. [52]

    Rectifier nonlinearities improve neural net- work acoustic models,

    A. L. Maas, “Rectifier nonlinearities improve neural net- work acoustic models,” inProceedings of the 30th Inter- national Conference on Machine Learning, Vol. 28, 3., 2013

    2013

  52. [53]

    Deconvolution and checkerboard artifacts,

    A. Odena, V. Dumoulin, and C. Olah, “Deconvolution and checkerboard artifacts,”Distill, 2016

    2016

  53. [54]

    Generative adversarial networks,

    I. Goodfellow, J. Pouget-Abadie, M. Mirza, B. Xu, D. Ward-Farley, S. Ozair, A. Courville, and Y. Bengio, “Generative adversarial networks,”Advances in neural information processing systems, pp. 2672–2680, 2014

    2014

  54. [55]

    On a space of completely additive functions,

    L. W. Kantorovich and G. S. Rubinstein, “On a space of completely additive functions,”Vestnik Leningradskogo Universiteta, vol. 13, pp. 52–59, 1958

    1958

  55. [56]

    On information and sufficiency,

    S. Kullback and R. A. Leibler, “On information and sufficiency,”Annals of Mathematical Statistics, vol. 22, pp. 79–86, 1951

    1951

  56. [57]

    Villani,Topics in Optimal Transportation

    C. Villani,Topics in Optimal Transportation. American Mathematical Society, 2003

    2003

  57. [58]

    Improved training of Wasserstein GANs,

    G. Ishaan, A. Faruk, A. Martin, D. Vincent, and C. Aaron, “Improved training of Wasserstein GANs,” NIPS’17: Proceedings of the 31st International Confer- ence on Neural Information Processing Syste, vol. 30, 2017

    2017

  58. [59]

    Evaluation of mode collapse in generative 1 adversarial networks,

    L. Sayeri, S. Maha, and L. Anastasiya, Belyaeva and Molei, “Evaluation of mode collapse in generative 1 adversarial networks,” 2018/19

    2018

  59. [60]

    Inception-v4, inception-resnet and the impact of resid- ual connections on learning,

    C. Szegedy, S. Ioffe, V. Vanhoucke, and A. A. Alemi, “Inception-v4, inception-resnet and the impact of resid- ual connections on learning,” inProceedings of the Thirty-First AAAI Conference on Artificial Intelligence, p. 4278–4284, 2017

    2017

  60. [61]

    Deep resid- ual learning for image recognition,

    K. He, X. Zhang, S. Ren, and J. Sun, “Deep resid- ual learning for image recognition,” in2016 IEEE Con- ference on Computer Vision and Pattern Recognition, pp. 770–778, 2016

    2016

  61. [62]

    This deliberate choice allows the network to robustly learn and resolve physically relevant but sta- tistically uncommon configurations

    We note that this construction does not correspond to direct ensemble sampling of the XY model, as configura- tions with rare defect pair separations are intentionally oversampled. This deliberate choice allows the network to robustly learn and resolve physically relevant but sta- tistically uncommon configurations

  62. [63]

    Adam: A method for stochas- tic optimization,

    D. Kingma and J. Ba, “Adam: A method for stochas- tic optimization,”International Conference on Learning Representations, 2014

    2014

  63. [64]

    Watch your up- convolution: CNN based generative deep neural networks are failing to reproduce spectral distributions,

    R. Durall, M. Keuper, and J. Keuper, “Watch your up- convolution: CNN based generative deep neural networks are failing to reproduce spectral distributions,” in2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp. 7887–7896, 2020

    2020

  64. [65]

    Spectral distribution aware im- age generation,

    S. Jung and M. Keuper, “Spectral distribution aware im- age generation,” inAAAI Conference on Artificial Intel- ligence, 2020. Supplementary Material: Reconstruction of spin structures from topological charge distributions via generative neural network systems Kyra H. M. Klos, 4 Jan Disselhoff,5 Michael Wand,5 Karin Everschor-Sitte,6 and Friederike Schmid...

    2020