Pith. sign in

REVIEW 3 major objections 6 minor 58 references

Minimal surfaces can be built exactly from two holomorphic neural nets via a spinor formula, then fit to a boundary curve.

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

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

T0 review · grok-4.5

2026-07-30 22:04 UTC pith:CKJYCSYK

load-bearing objection Solid methods paper: spinor neuralization gives minimality by construction; Plateau claims rest on softer empirical ground than the math. the 3 major comments →

arxiv 2607.23437 v1 pith:CKJYCSYK submitted 2026-07-26 cs.GR cs.LG

Neural Representation of Minimal Surfaces

Jiayin Sun , Albert Chern This is my paper
classification cs.GR cs.LG MSC 53A1049Q0568T0730F15
keywords minimal surfacesPlateau problemspinor representationcomplex-valued neural networksWeierstrass–Enneperholomorphic curvesnull curves
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

This paper shows how to represent minimal surfaces—soap-film shapes with zero mean curvature—so that every surface produced by the network is already minimal, up to a small numerical integration error. Instead of training a network to approximately satisfy the minimal-surface PDE, the authors encode the classical spinor parameterization of null holomorphic curves into complex-valued neural nets for two holomorphic functions p and q. Integrating the resulting triple automatically yields a surface that is regular or has only removable branch points. They then train those nets, together with a learned reparameterization of the boundary circle, so the surface’s edge matches a prescribed closed space curve (the Plateau problem). A sympathetic reader cares because the construction turns an analytic identity into a trainable architecture: minimality is free, and the only remaining task is boundary fitting.

Core claim

Any pair of holomorphic functions p and q on the disk, realized by complex-valued neural networks and inserted into Euclid’s spinor formulas, produces via real-part integration a generalized minimal surface that is either regular or has only removable branch points; optimizing those networks plus a boundary homeomorphism solves the Plateau problem for a given rectifiable Jordan curve up to quadrature and fitting error.

What carries the argument

Euclid’s spinor representation: the null triple (φ₁,φ₂,φ₃)=(p²-q²,-i(p²+q²),2pq). It converts two holomorphic functions into a null holomorphic velocity whose real-part integral is automatically a (generalized) minimal surface, avoiding poles of the classical Weierstrass–Enneper Gauss map.

Load-bearing premise

That alternating gradient steps on the surface nets and a ReLU-exponential density network for circle homeomorphisms will reliably drive the unparameterized boundary curves into close agreement, even though only density of the homeomorphism class is proved and no convergence guarantee is given.

What would settle it

Train the architecture on a closed curve whose unique minimal disk is known analytically (e.g., a planar circle or a catenoid boundary); if the recovered surface’s maximum mean curvature does not fall as O(1/N²) with quadrature steps N, or if the boundary L² mismatch stays large after long training, the claim fails.

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

If this is right

  • Every random initialization of the p,q nets already yields an approximately minimal surface; training only has to match the boundary.
  • The method supplies the analytic Weierstrass data (φ₁,φ₂,φ₃) as a byproduct, not merely a mesh or implicit field.
  • Branch points that appear are automatically removable, so the Plateau solutions stay in the regular class after infinitesimal perturbation.
  • Mean-curvature error is controlled solely by the quadrature rule and converges at second order, independent of the training loss.

Where Pith is reading between the lines

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

  • The same spinor nets could be dropped into other geometric variational problems whose Euler–Lagrange equations admit holomorphic first integrals.
  • Extending the domain from the disk to higher-genus Riemann surfaces would require holomorphic sections of spinor bundles rather than ordinary CVNNs.
  • Because minimality is hard-wired, the architecture is a natural differentiable layer inside larger graphics or fabrication pipelines that need soap-film-like surfaces as primitives.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 6 minor

Summary. The paper introduces a neural representation of minimal surfaces based on the spinor (Euclid) parameterization of null holomorphic curves: two complex-valued neural networks p, q with holomorphic activations are composed via Eq. (8) into a null triple (φ₁, φ₂, φ₃), whose real part integrated once yields a conformal harmonic — hence minimal — surface by construction, with no PDE residual in the loss. Theorem 2.2 shows every regular minimal surface on the disk admits such (p, q); Corollary 2.1 shows that arbitrary holomorphic (p, q) yield surfaces that are regular or have only removable branch points (via the parity result of Theorem 2.1, proved in Appendix B). For the Plateau problem, the authors optimize p, q jointly with a reparameterization Φ̃ ∈ Homeo⁺(S¹) represented by a normalized integral of a positive ReLU-exp MLP (Theorem 3.1, density proved in Appendix D), alternating RMSprop updates (Algorithm 1) on a boundary L²-plus-tangent loss. Experiments compare against a neural Weierstrass–Enneper baseline on two curves (Figs. 3–4), recover prescribed surfaces (Fig. 5), and verify that mean curvature vanishes at the O(1/N²) trapezoidal quadrature rate (Figs. 8–11). I checked the mathematical core independently: the even-order-zero arguments in Theorem 2.2 via the null relation (φ₁−iφ₂)(φ₁+iφ₂) = −φ₃² are correct, the global sign ambiguity in φ₃ = ±2pq is legitimately absorbed by the sign freedom of Lemma 2.1 (Appendix C's monodromy argument is standard and correct on the disk

Significance. If the empirical claims are substantiated, this is a useful contribution to geometry processing: an architecture-level exactness guarantee for a nonlinear geometric PDE, with all branch points provably removable, no pole management, and publicly included code. The minimality guarantee is parameter-free — mean curvature is never part of the loss, so its measured vanishing is a genuine, falsifiable check, and it passes at the predicted quadrature order. The spinor formulation also yields the Weierstrass data as an analytic fingerprint of each surface, a byproduct other neural methods do not provide. The main limitation is that the evidence for actually solving the Plateau problem is currently thinner than the representation theory behind it.

major comments (3)
  1. [§4, Figs. 3–5] §4.1–4.2, Figures 3–5: the quantitative evidence that the alternating optimization 'solves the Plateau problem' is not yet convincing. For the two real input curves (perturbed circle, Eq. (15); tennis seam), only visual agreement is reported — no boundary L² or Hausdorff error, no area value, and no comparison against an independent Plateau solver (e.g., Surface Evolver [Brakke 1992] or the currents-based method of [Wang and Chern 2021], both already cited). The only quantitative recovery tests (Fig. 5) synthesize ground truths from randomized neural (p,q) of the same architecture as the model — i.e., from the hypothesis class itself — so success there demonstrates self-consistency of the pipeline rather than the ability to span an arbitrary rectifiable Jordan curve. The Taylor-perturbed variant (Fig. 5 right) departs from the class only mildly. At minimum, please report boundary-error m
  2. [§2.4, Problem 3.1] §2.4 and Problem 3.1: there is an approximation-theoretic gap between what is proved and the claim that 'any solution to the Plateau problem for rectifiable Jordan curves can be parameterized' and neurally approximated. Theorem 2.2 gives holomorphic (p,q) on the open disk D for a regular minimal surface, and the cited universal approximation results ([Calafà et al. 2024]) hold uniformly on compact subsets of D. The Plateau objective (13), however, is a boundary trace condition on ∂D; uniform approximability of the target (p,q) up to the boundary does not follow from compact-subset density, and for general rectifiable Γ the Weierstrass data need not extend nicely to ∂D. A short remark clarifying the precise sense (topology, domain) in which the neural class is dense for boundary-fitting purposes — or a targeted result, e.g., for Γ with some boundary regularity — would close this gap.
  3. [§3.2, Alg. 1] §3.2, Algorithm 1: the manuscript does not discuss the known failure modes of the alternating (surface, reparameterization) scheme, and the loss (13) alone does not prevent several of them: (i) degenerate density ρ that concentrates Φ̃⁻¹ and stalls progress; (ii) local minima in the joint (p,q,Φ̃) landscape; (iii) the three-dimensional Möbius null direction of the boundary correspondence, which Fig. 6/7 implicitly confirm is present in the recovered data — it is neither quotiented out nor regularized, and its effect on conditioning is not discussed. Additionally, nothing in the objective enforces that the recovered surface is the Douglas area-minimizing solution rather than another (possibly unstable) minimal surface spanning Γ, so the phrase 'solves the Plateau problem up to quadrature and fitting error' should be tempered accordingly. An ablation on initialization sensitivity and a dis
minor comments (6)
  1. [Throughout] Typos: 'nueral' (§1, related-work paragraph on UAT); 'holomoprhic' (proof of Thm 2.2); 'A class fundamental questions' (§2.1.1); 'taking Γ as the boundary of a prescribing a minimal surface' and 'recover the the same minimal surface' (§4.2); 'Hass and Hughies' should be 'Hughes' (§2.1.3 and references); 'whereas the ability to remove a non-immersed point... is not captured' — footnote 2 is generally helpful but would benefit from copyediting.
  2. [Problem 3.1] The measure μ in (13) is unspecified ('any measure on S¹'); since the implemented loss is the uniform L² measure plus the tangent term of §3.2.1, please state this at (13).
  3. [§2.4, §4] N = 100 trapezoidal steps is fixed for all experiments with no sensitivity study on the boundary-fit quality as a function of N; since the boundary map itself is evaluated through the quadrature (12), a brief table of boundary error vs. N would complement the curvature study of Fig. 11.
  4. [§4.1] Figures 3 and 4: the Weierstrass–Enneper baseline failure is attributed to pole singularities in g; a plot of the learned f, g (analogous to Figs. 6–7) for the failing baseline would make this concrete and rule out under-training as the cause.
  5. [§2.1.3, Eq. (4)] The 𝔮 convention in (4) (𝔮 = 0 for removable, 1 for genuine) is opposite to some conventions in the cited immersion literature; a one-line note would prevent confusion.
  6. [Fig. 8] Fig. 8 uses the catenoid parameterized on [−1,1]×[0,2π] rather than the disk; please clarify that this quadrature study is independent of the disk-based pipeline (12) and explain briefly how it transfers.

Circularity Check

0 steps flagged

No significant circularity: minimality is inherited from classical null-curve/spinor theory by construction, and training only targets boundary fidelity.

full rationale

The paper’s central representation (Euclid spinor formulas (8)/(11), Corollary 2.1) is an exact classical parameterization of generalized minimal surfaces; any (p, q) holomorphic pair yields a null holomorphic triple, hence a conformal harmonic surface up to quadrature. That is intentional architecture, not a fitted quantity renamed as a discovery. Theorems 2.1–2.2 and Lemma 2.1 are proved in-appendices from standard complex analysis (even-order zeros, monodromy on the disk) without load-bearing self-citation chains that smuggle the conclusion. The Plateau objective (Problem 3.1, Algorithm 1) optimizes only boundary homeomorphism fidelity; vanishing mean curvature is never in the loss and is reported post hoc as tracking the known O(1/N²) trapezoidal error (Figs. 9, 11). Self-citations (Chern et al. 2018 on branch-point quadratic forms; Wang & Chern 2021 as related work) supply background geometry, not uniqueness theorems that force the present claims. Recovery experiments (Fig. 5) that regenerate surfaces from the same neural (p, q) class are weak external validation, not algebraic circularity: no predicted observable is identical to a fitted constant by definition. Derivation chain is self-contained against classical benchmarks.

Axiom & Free-Parameter Ledger

4 free parameters · 5 axioms · 0 invented entities

Load-bearing content is classical minimal-surface complex analysis plus standard MLP universal-approximation facts; the paper adds architectural choices and a training schedule, not new physical entities. Free parameters are ordinary neural hyperparameters and the tangent-loss weight.

free parameters (4)
  • tangent loss weight λ = 0.5
    Hand-chosen coefficient balancing L_distance and L_tangent; paper sets λ=0.5 and shows λ=0 is worse, but value is not derived.
  • trapezoidal partitions N = 100
    Controls quadrature error that is the only source of non-exact minimality; fixed at 100 for all examples.
  • CVNN depth/width for p,q = 10×20
    10 layers, width 20; architectural capacity choice that determines which holomorphic maps are representable in practice.
  • reparameterization MLP widths and learning rate = lr=1e-5, batch=128
    Architecture [1,64,64,32,1], lr 1e-5, batch 128; standard training knobs not fixed by theory.
axioms (5)
  • standard math A surface is generalized minimal iff it is conformal and harmonic; equivalently the real part of a null holomorphic curve in C³.
    Definition 2.1 and §2.1.2; classical Osserman theory.
  • standard math Douglas–Osserman: every rectifiable Jordan curve bounds a regular minimal disk.
    Cited to justify that Plateau solutions exist and are representable; Problem 2.1.
  • standard math Holomorphic square roots exist on the disk precisely when zeros have even order (Lemma 2.1).
    Used to prove spinor functions p,q are holomorphic (Thm 2.2).
  • domain assumption Complex MLPs with holomorphic activations are holomorphic and dense in holomorphic functions on compacta (Calafà et al. UAT).
    Invoked to claim the neural spinor class can approximate any simply-connected minimal surface on compact subsets of D.
  • ad hoc to paper Orientation-preserving homeomorphisms of S¹ can be approximated in sup norm by normalized integrals of positive ReLU-exp MLPs (Thm 3.1).
    Proved in Appendix D for this architecture; needed so the boundary loss is not over-constrained by parameterization.

pith-pipeline@v1.2.0-grok45-kimik3 · 23318 in / 3048 out tokens · 63395 ms · 2026-07-30T22:04:43.128898+00:00 · methodology

0 comments
read the original abstract

We propose a neural representation for minimal surfaces. Unlike prior approaches based on discretization or Physics-Informed Neural Networks (PINNs), where meshes or neural fields are optimized to approximate the governing equations, our method builds on an exact representation, similar to the classical Weierstrass--Enneper parameterization, yielding minimal surfaces up to negligible quadrature error in evaluation. We formulate a training objective for the Plateau problem that optimizes over this representation.

Figures

Figures reproduced from arXiv: 2607.23437 by Albert Chern, Jiayin Sun.

Figure 1
Figure 1. Figure 1: Overview of the Neural Architecture. The holomorphic functions [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Given the curve Γ : S 1 → R 3 , we seek a minimal surface r : D → R 3 together with a homeomorphism Φ˜ between 𝜕D and S 1 that minimizes the distance between the boundary map r|𝜕D and Γ ◦ Φ˜ . some 𝑏 ∈ R. Every such function can be written as Ψ(𝑡) := ( ∫ 𝑡 0 𝜌 (𝑠) 𝑑𝑠)/(∫ 1 0 𝜌 (𝑠) 𝑑𝑠) + 𝑏 (14) for some positive-valued function 𝜌 : [0, 1] → R>0. We let 𝜌 be a real-valued single-input-single-output MLP with … view at source ↗
Figure 3
Figure 3. Figure 3: Comparison of Plateau problem optimization using (a) our method [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Comparison of Plateau problem optimization using (a) our method [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Two comparisons of the trained surface (blue interior, black bound [PITH_FULL_IMAGE:figures/full_fig_p007_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Corresponding functions p, q, f, and g for (i) the trained surface, and [PITH_FULL_IMAGE:figures/full_fig_p007_6.png] view at source ↗
Figure 8
Figure 8. Figure 8: Comparison of trapezoidal integration with varying numbers of quad [PITH_FULL_IMAGE:figures/full_fig_p009_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Visualization of the absolute mean curvature [PITH_FULL_IMAGE:figures/full_fig_p009_9.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

58 extracted references · 4 linked inside Pith

  1. [1]

    ACM Trans

    Wang, Stephanie and Chern, Albert , title =. ACM Trans. Graph. , issue_date =. 2021 , pages =. doi:10.1145/3450626.3459781 , publisher =

    arXiv 2021

  2. [2]

    ACADIA , volume=

    Minimal surfaces as self-organizing systems , author=. ACADIA , volume=

  3. [3]

    Facta universitatis-series: Architecture and Civil Engineering , volume=

    Minimal surfaces for architectural constructions , author=. Facta universitatis-series: Architecture and Civil Engineering , volume=

  4. [4]

    Advanced Engineering Materials , volume=

    Multifunctional mechanical metamaterials based on triply periodic minimal surface lattices , author=. Advanced Engineering Materials , volume=. 2019 , publisher=

    2019

  5. [5]

    International Journal of Heat and Mass Transfer , volume=

    Characterisation of heat transfer within 3D printed TPMS heat exchangers , author=. International Journal of Heat and Mass Transfer , volume=. 2023 , publisher=

    2023

  6. [6]

    The Mathematica Journal , volume=

    Visualizing Minimal Surfaces , author=. The Mathematica Journal , volume=

  7. [7]

    Notices of the AMS , volume=

    The visualization of mathematics: Towards a mathematical exploratorium , author=. Notices of the AMS , volume=

  8. [8]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =

    Palmer, David and Smirnov, Dmitriy and Wang, Stephanie and Chern, Albert and Solomon, Justin , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =. 2022 , pages =

    2022

  9. [9]

    Annals of Mathematics , volume=

    Minimal surfaces from circle patterns: Geometry from combinatorics , author=. Annals of Mathematics , volume=. 2006 , publisher=

    2006

  10. [10]

    Annals of Mathematics , volume=

    A method of numerical solution of the problem of Plateau , author=. Annals of Mathematics , volume=. 1927 , publisher=

    1927

  11. [11]

    Experimental mathematics , volume=

    Computing discrete minimal surfaces and their conjugates , author=. Experimental mathematics , volume=. 1993 , publisher=

    1993

  12. [12]

    Computing , volume=

    A contribution to the numerical approximation of minimal surfaces , author=. Computing , volume=. 1977 , publisher=

    1977

  13. [13]

    Numerische Mathematik , volume=

    An algorithm for evolutionary surfaces , author=. Numerische Mathematik , volume=. 1990 , publisher=

    1990

  14. [14]

    Experimental mathematics , volume=

    The surface evolver , author=. Experimental mathematics , volume=. 1992 , publisher=

    1992

  15. [15]

    1991 , publisher=

    Computing minimal surfaces via level set curvature flow , author=. 1991 , publisher=

    1991

  16. [16]

    2008 , publisher=

    Pythagorean-hodograph curves: algebra and geometry inseparable , author=. 2008 , publisher=

    2008

  17. [17]

    , author=

    Shape from metric. , author=. ACM Trans. Graph. , volume=

  18. [18]

    The American Mathematical Monthly , volume=

    Immersions and mod-2 quadratic forms , author=. The American Mathematical Monthly , volume=. 1977 , publisher=

    1977

  19. [19]

    Topology , volume=

    Immersions of surfaces in 3-manifolds , author=. Topology , volume=. 1985 , publisher=

    1985

  20. [20]

    1984 , publisher=

    Regular homotopy classes of immersed surfaces , author=. 1984 , publisher=

    1984

  21. [21]

    arXiv preprint arXiv:2309.02589 , year=

    Approximating high-dimensional minimal surfaces with physics-informed neural networks , author=. arXiv preprint arXiv:2309.02589 , year=

    Pith/arXiv arXiv

  22. [22]

    Thin-Walled Structures , volume=

    Physics-informed neural networks for the form-finding of tensile membranes by solving the Euler--Lagrange equation of minimal surfaces , author=. Thin-Walled Structures , volume=. 2023 , publisher=

    2023

  23. [23]

    Communications in mathematical physics , volume=

    Null vectors, spinors, and strings , author=. Communications in mathematical physics , volume=. 1986 , publisher=

    1986

  24. [24]

    arXiv preprint dg-ga/9512003 , year=

    The spinor representation of minimal surfaces , author=. arXiv preprint dg-ga/9512003 , year=

    Pith/arXiv arXiv

  25. [25]

    1994 , institution=

    Discrete isothermic surfaces , author=. 1994 , institution=

    1994

  26. [26]

    International Mathematics Research Notices , volume=

    Discrete minimal surfaces: critical points of the area functional from integrable systems , author=. International Mathematics Research Notices , volume=. 2018 , publisher=

    2018

  27. [27]

    Computer Methods in Applied Mechanics and Engineering , volume=

    Physics-informed holomorphic neural networks (PIHNNs): Solving 2D linear elasticity problems , author=. Computer Methods in Applied Mechanics and Engineering , volume=. 2024 , publisher=

    2024

  28. [28]

    2013 , publisher=

    A survey of minimal surfaces , author=. 2013 , publisher=

    2013

  29. [29]

    Annals of Mathematics , volume=

    A proof of the regularity everywhere of the classical solution to Plateau's problem , author=. Annals of Mathematics , volume=. 1970 , publisher=

    1970

  30. [30]

    Transactions of the American Mathematical Society , volume=

    Solution of the problem of Plateau , author=. Transactions of the American Mathematical Society , volume=. 1931 , publisher=

    1931

  31. [31]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Neural surface maps , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  32. [32]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Deepsdf: Learning continuous signed distance functions for shape representation , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  33. [33]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Occupancy networks: Learning 3d reconstruction in function space , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  34. [34]

    A papier-m

    Groueix, Thibault and Fisher, Matthew and Kim, Vladimir G and Russell, Bryan C and Aubry, Mathieu , booktitle=. A papier-m

  35. [35]

    1761 , publisher=

    Essai d'une nouvelle methode pour de'terminer les maxima, et les minima des formules integrales indefinies , author=. 1761 , publisher=

  36. [36]

    Mathematics of Computation , volume=

    Numerical solution of the minimal surface equation , author=. Mathematics of Computation , volume=. 1967 , publisher=

    1967

  37. [37]

    Mathematics of Computation , volume=

    Numerical solution of Plateau’s problem by a finite element method , author=. Mathematics of Computation , volume=

  38. [38]

    Numerische Mathematik , volume=

    On discrete Dirichlet and Plateau problems , author=. Numerische Mathematik , volume=. 1961 , publisher=

    1961

  39. [39]

    ACM SIGGRAPH 2011 papers , pages=

    Spin transformations of discrete surfaces , author=. ACM SIGGRAPH 2011 papers , pages=

    2011

  40. [40]

    1990 , publisher=

    A crystalline approximation theorem for hypersurfaces , author=. 1990 , publisher=

    1990

  41. [41]

    Proceedings of the twenty-seventh annual symposium on Computational geometry , pages=

    The least spanning area of a knot and the optimal bounding chain problem , author=. Proceedings of the twenty-seventh annual symposium on Computational geometry , pages=

  42. [42]

    Proceedings of the forty-second ACM symposium on Theory of computing , pages=

    Optimal homologous cycles, total unimodularity, and linear programming , author=. Proceedings of the forty-second ACM symposium on Theory of computing , pages=

  43. [43]

    Discrete & computational geometry , volume=

    Lexicographic optimal homologous chains and applications to point cloud triangulations , author=. Discrete & computational geometry , volume=. 2022 , publisher=

    2022

  44. [44]

    arXiv preprint arXiv:2402.14009 , year=

    Geometry-informed neural networks , author=. arXiv preprint arXiv:2402.14009 , year=

    Pith/arXiv arXiv

  45. [45]

    Machine Learning: Science and Technology , volume=

    Physics-informed neural network solves minimal surfaces in curved spacetime , author=. Machine Learning: Science and Technology , volume=. 2026 , publisher=

    2026

  46. [46]

    SIAM Journal on Scientific Computing , volume=

    Computing least area hypersurfaces spanning arbitrary boundaries , author=. SIAM Journal on Scientific Computing , volume=. 1997 , publisher=

    1997

  47. [47]

    , author=

    Analytisch-geometrische Untersuchungen. , author=. Nachrichten von der K

  48. [48]

    Untersuchungen

    Weierstrass, Karl , year=. Untersuchungen

  49. [49]

    IEEE/CAA Journal of Automatica Sinica , volume=

    Complex-valued neural networks: A comprehensive survey , author=. IEEE/CAA Journal of Automatica Sinica , volume=. 2022 , publisher=

    2022

  50. [50]

    arXiv preprint arXiv:2407.19258 , year=

    Comprehensive survey of complex-valued neural networks: Insights into backpropagation and activation functions , author=. arXiv preprint arXiv:2407.19258 , year=

    Pith/arXiv arXiv

  51. [51]

    Annals of Mathematics , volume=

    Normal and integral currents , author=. Annals of Mathematics , volume=. 1960 , publisher=

    1960

  52. [52]

    Partial Differential Equations: Proceedings of a Symposium held in Tianjin, June 23--July 5, 1986 , pages=

    Co-area, liquid crystals, and minimal surfaces , author=. Partial Differential Equations: Proceedings of a Symposium held in Tianjin, June 23--July 5, 1986 , pages=. 2006 , organization=

    1986

  53. [53]

    Comptes Rendus

    The Plateau problem from the perspective of optimal transport , author=. Comptes Rendus. Math

  54. [54]

    Proceedings of the 26th annual conference on Computer graphics and interactive techniques , pages=

    Implicit fairing of irregular meshes using diffusion and curvature flow , author=. Proceedings of the 26th annual conference on Computer graphics and interactive techniques , pages=

  55. [55]

    Numerische Mathematik , volume=

    Variational convergence of discrete minimal surfaces , author=. Numerische Mathematik , volume=. 2019 , publisher=

    2019

  56. [56]

    Computers & Graphics , volume=

    Geometric implicit neural representations for signed distance functions , author=. Computers & Graphics , volume=. 2024 , publisher=

    2024

  57. [57]

    European conference on computer vision , pages=

    A level set theory for neural implicit evolution under explicit flows , author=. European conference on computer vision , pages=. 2022 , organization=

    2022

  58. [58]

    ACM SIGGRAPH 2024 Conference Papers , pages=

    Neural geometry fields for meshes , author=. ACM SIGGRAPH 2024 Conference Papers , pages=

    2024