pith. sign in

arxiv: 2504.20941 · v4 · submitted 2025-04-29 · 💻 cs.CR · math.DG· stat.OT

Conformal-DP: A Density-Aware Mechanism for Differential Privacy over Riemannian Manifolds via Conformal Transformation

Peilin He , Liou Tang , M. Amin Rahimian , James Joshi This is my paper

Pith reviewed 2026-05-22 18:28 UTC · model grok-4.3

classification 💻 cs.CR math.DGstat.OT
keywords differential privacyRiemannian manifoldsconformal transformationsdensity-aware mechanismsgeodesic error boundprivacy-utility trade-offmanifold data
0
0 comments X

The pith

Conformal transformations create density-balanced geometries on Riemannian manifolds for a differential privacy mechanism whose error depends only on local density ratios.

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

The paper proposes Conformal-DP, which applies conformal transformations to Riemannian manifold data so that perturbations can be calibrated according to local densities rather than assuming uniform distribution. It establishes that the resulting mechanism meets ε-differential privacy on any complete Riemannian manifold provided mild regularity conditions hold. A closed-form bound on expected geodesic error is also derived; the bound is controlled solely by the ratio of the underlying data densities and does not involve the manifold's global curvature. Existing manifold DP methods produce biased noise when densities vary across the space, and the new approach is shown to improve the privacy-utility trade-off on both synthetic and real heterogeneous datasets while matching worst-case performance of uniform-distribution baselines.

Core claim

Conformal-DP leverages conformal transformations to induce a density-balanced geometry on the manifold, allowing noise calibration that respects local data densities. The mechanism satisfies ε-differential privacy on arbitrary complete Riemannian manifolds under mild regularity assumptions. It further supplies a closed-form expression for expected geodesic error whose value depends only on the data density ratio and remains independent of global curvature.

What carries the argument

Conformal transformation that induces a density-balanced geometry for calibrating privacy perturbations on Riemannian manifolds

If this is right

  • The mechanism satisfies ε-differential privacy on any complete Riemannian manifold under the mild regularity assumptions.
  • Expected geodesic error after perturbation is bounded by a closed-form expression depending only on the data density ratio.
  • Privacy-utility performance improves over prior manifold mechanisms when data density is heterogeneous.
  • Worst-case utility matches that of state-of-the-art mechanisms designed for uniform data distributions.

Where Pith is reading between the lines

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

  • The curvature independence may allow the same mechanism to be deployed on manifolds whose curvature varies sharply without retuning the noise scale.
  • Conformal balancing could be combined with existing Euclidean DP primitives when data are embedded in a manifold that admits such transformations.
  • The density-ratio dependence suggests that preprocessing steps that estimate local density ratios become the main practical bottleneck for deployment.

Load-bearing premise

The ε-differential privacy guarantee and the curvature-independent error bound both rest on the existence of a conformal transformation that successfully balances the data density across the manifold together with the validity of the stated mild regularity assumptions.

What would settle it

An experiment that measures expected geodesic error on a complete manifold with known strong curvature variation and finds that the error changes with curvature, or that exhibits an ε-DP violation on some complete manifold satisfying the regularity conditions, would falsify the central claims.

Figures

Figures reproduced from arXiv: 2504.20941 by James Joshi, Liou Tang, M. Amin Rahimian, Peilin He.

Figure 1
Figure 1. Figure 1: An overview of Conformal-DP mechanism. Here, the privacy loss is quantified by the privacy budget ε: the smaller the value of ε, the better the privacy achieved. The probabilities are taken over the randomness of the algorithm A, which typically injects random noise or employs sampling to limit the impact of any single record on the output. In practice, analyses often involve multiple differentially privat… view at source ↗
Figure 5
Figure 5. Figure 5: Comparison of the utility for the three mechanisms [PITH_FULL_IMAGE:figures/full_fig_p013_5.png] view at source ↗
Figure 2
Figure 2. Figure 2: Comparison of the utility for the three mechanisms [PITH_FULL_IMAGE:figures/full_fig_p013_2.png] view at source ↗
Figure 6
Figure 6. Figure 6: Heatmap for data distortion on a 3-dimensional hyper [PITH_FULL_IMAGE:figures/full_fig_p013_6.png] view at source ↗
Figure 4
Figure 4. Figure 4: Comparison of the utility for the three mechanisms [PITH_FULL_IMAGE:figures/full_fig_p013_4.png] view at source ↗
Figure 7
Figure 7. Figure 7: Utilities of private Frechet means under varying privacy ´ budgets (ε) and different numbers of samples for CIFAR-10 dataset. 0.1 0.3 0.5 0.7 0.9 Privacy Budgets 2 4 6 8 10 12 14 g ( p, n p ) Distance vs Epsilon with ±0.5 Std Error Bands (Dataset=FashionMNIST, N=6000) Tangent Gaussian ( = 10 9 ) Riemannian Laplace Conformal Laplace (ours) (a) Class N = 6000 0.1 0.3 0.5 0.7 0.9 Privacy Budgets 0 2 4 6 8 10 … view at source ↗
Figure 8
Figure 8. Figure 8: Utilities of private Frechet means under varying privacy ´ budgets (ε) and different numbers of samples for Fashion￾MNIST dataset. more visible when ε is small, where privacy constraints are tighter and noise effects become stronger. In [PITH_FULL_IMAGE:figures/full_fig_p015_8.png] view at source ↗
read the original abstract

Differential Privacy (DP) is being increasingly adopted for non-Euclidean data that lie on complex, high-dimensional manifolds. Existing DP mechanisms for manifold data consider geometric properties when calibrating privacy perturbations, but they largely fail to capture variations in data density within datasets, leading to biased perturbations and suboptimal privacy-utility trade-offs due to heterogeneous data distributions. In this paper, we propose a novel density-aware differential privacy mechanism on Riemannian manifolds, referred to as Conformal-DP, that leverages conformal transformations to calibrate perturbations based on local densities and to induce a density-balanced geometry. We prove that our mechanism satisfies $\epsilon$-differential privacy on any complete Riemannian manifold under mild regularity assumptions. In addition, we derive a closed-form expected geodesic error bound that depends only on the underlying data density ratio and is independent of global curvature. Our empirical results on synthetic and real-world datasets demonstrate that the proposed Conformal-DP mechanism substantially improves the privacy-utility trade-off in heterogeneous data distribution settings, with worst-case performance comparable to state-of-the-art manifold DP mechanisms that assume uniformly distributed data.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 2 minor

Summary. The paper proposes Conformal-DP, a density-aware differential privacy mechanism for Riemannian manifold data. It employs conformal transformations to induce a density-balanced geometry and calibrate perturbations according to local densities. The central claims are a proof that the mechanism satisfies ε-differential privacy on any complete Riemannian manifold under mild regularity assumptions, together with a closed-form expected geodesic error bound that depends only on the underlying data density ratio and is independent of global curvature. Experiments on synthetic and real-world datasets are reported to demonstrate improved privacy-utility trade-offs under heterogeneous distributions.

Significance. If the stated proofs and curvature-independent bound hold, the work would meaningfully advance manifold differential privacy by addressing density heterogeneity, a limitation of prior geometric mechanisms. The closed-form error bound and the use of conformal geometry to absorb density variation into the metric could simplify analysis and deployment on complex manifolds. The empirical results provide supporting evidence for practical gains, though the primary value lies in the theoretical contributions if they can be verified.

major comments (2)
  1. [§4, Theorem 1] §4, Theorem 1 (ε-DP proof): The argument that the mechanism satisfies ε-DP on arbitrary complete manifolds rests on the existence of a globally smooth conformal factor that balances arbitrary positive densities while preserving completeness and the bounded-gradient properties needed for the sensitivity calculation. Standard existence results for prescribed curvature or density equations (e.g., Yamabe-type problems) do not guarantee such a factor without singularities or loss of completeness on non-compact or high-curvature manifolds; the paper provides no constructive existence proof or additional regularity conditions that close this gap.
  2. [§5.1, Eq. (12)] §5.1, Eq. (12) (error bound derivation): The closed-form expected geodesic error is asserted to depend only on the data density ratio and to be independent of global curvature. The derivation absorbs curvature effects into the transformed density ratio, but the steps do not explicitly show that the conformal factor eliminates all residual curvature dependence in the geodesic distance or noise calibration; if the transformed metric retains curvature terms not captured by the local ratio, the independence claim fails.
minor comments (2)
  1. [§3] Notation for the conformal factor λ and the pulled-back metric g' is introduced without an explicit definition of the transformation map in §3; adding a short display equation would improve readability.
  2. [§6] The experimental figures (e.g., Figure 4) report average utility but omit error bars or statistical tests across random seeds; this makes it harder to assess the significance of the reported improvements over baselines.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their careful reading of the manuscript and for providing detailed comments that help improve the clarity of our theoretical contributions. We respond to each major comment below.

read point-by-point responses

  1. Referee: [§4, Theorem 1] §4, Theorem 1 (ε-DP proof): The argument that the mechanism satisfies ε-DP on arbitrary complete manifolds rests on the existence of a globally smooth conformal factor that balances arbitrary positive densities while preserving completeness and the bounded-gradient properties needed for the sensitivity calculation. Standard existence results for prescribed curvature or density equations (e.g., Yamabe-type problems) do not guarantee such a factor without singularities or loss of completeness on non-compact or high-curvature manifolds; the paper provides no constructive existence proof or additional regularity conditions that close this gap.

    Authors: We thank the referee for highlighting this important aspect of the proof. The manuscript states that the ε-DP guarantee holds under mild regularity assumptions on the data density. These assumptions are intended to ensure the existence of a suitable conformal factor that is globally smooth, preserves completeness, and maintains the necessary bounded gradient properties. To make this more explicit, we will revise the statement of Theorem 1 and the surrounding discussion in §4 to specify the precise regularity conditions (e.g., the density being positive, smooth, and satisfying integrability conditions that guarantee solvability of the conformal transformation equation). We will also include a brief reference to relevant results from conformal geometry that support existence under these conditions, without claiming a new constructive proof. This addresses the gap by clarifying the scope of applicability. revision: partial

  2. Referee: [§5.1, Eq. (12)] §5.1, Eq. (12) (error bound derivation): The closed-form expected geodesic error is asserted to depend only on the data density ratio and to be independent of global curvature. The derivation absorbs curvature effects into the transformed density ratio, but the steps do not explicitly show that the conformal factor eliminates all residual curvature dependence in the geodesic distance or noise calibration; if the transformed metric retains curvature terms not captured by the local ratio, the independence claim fails.

    Authors: We appreciate the referee's careful scrutiny of the error bound derivation. In §5.1, the conformal transformation is designed such that the density variation is absorbed into the metric scaling, and the noise is calibrated in the transformed space where the effective density is uniform. The expected geodesic error is then computed using the properties of the transformed manifold, leading to an expression that depends solely on the original density ratio. To strengthen this, we will expand the derivation steps around Eq. (12) to explicitly trace how any curvature-dependent terms in the geodesic distance are neutralized by the conformal factor's scaling, confirming the independence from global curvature. This elaboration will make the cancellation explicit without altering the bound itself. revision: partial

Circularity Check

0 steps flagged

No significant circularity; derivation relies on external geometric assumptions and standard DP calibration

full rationale

The paper claims an ε-DP proof on complete Riemannian manifolds under mild regularity assumptions and a closed-form geodesic error bound depending only on local density ratio. No equations, definitions, or steps in the abstract or described claims reduce these results to self-definitional inputs, fitted parameters renamed as predictions, or load-bearing self-citations. The conformal transformation is presented as a mechanism to induce density-balanced geometry, with the error bound derived from that geometry rather than presupposing the target result. The derivation chain appears self-contained against external benchmarks in differential privacy and Riemannian geometry.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Only the abstract is available, so the ledger is necessarily incomplete. The central claims rest on unspecified mild regularity assumptions and the technical feasibility of density-calibrated conformal maps.

axioms (1)
  • domain assumption Mild regularity assumptions on the manifold and data distribution
    Invoked in the abstract to support the ε-DP proof on any complete Riemannian manifold.

pith-pipeline@v0.9.0 · 5736 in / 1229 out tokens · 37029 ms · 2026-05-22T18:28:22.316273+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

78 extracted references · 78 canonical work pages · 3 internal anchors

  1. [1]

    Pennec, S

    X. Pennec, S. Sommer, and T. Fletcher,Riemannian Geometric Statistics in Medical Image Analysis. Academic Press, 2019

    work page 2019

  2. [2]

    I. L. Dryden,Statistical Analysis on High-Dimensional Spheres and Shape Spaces. Chichester, UK: John Wiley & Sons, 2005

    work page 2005

  3. [3]

    Non-euclidean statistics for covariance matrices, with applications to diffusion tensor imaging,

    I. L. Dryden, A. Koloydenko, and D. Zhou, “Non-euclidean statistics for covariance matrices, with applications to diffusion tensor imaging,” The Annals of Applied Statistics, vol. 3, no. 3, pp. 1102–1123, 2009

    work page 2009

  4. [4]

    Manifold regularization: A geometric framework for learning from labeled and unlabeled examples,

    M. Belkin, P. Niyogi, and V . Sindhwani, “Manifold regularization: A geometric framework for learning from labeled and unlabeled examples,” The Journal of Machine Learning Research, vol. 7, pp. 2399–2434, 2006

    work page 2006

  5. [5]

    Manifold regularization and semi-supervised learning: Some theoretical analyses,

    P. Niyogi, “Manifold regularization and semi-supervised learning: Some theoretical analyses,”The Journal of Machine Learning Research, vol. 14, no. 1, pp. 1229–1250, 2013

    work page 2013

  6. [6]

    A novel dynamic system in the space of spd matrices with applications to appearance tracking,

    G. Cheng and B. C. Vemuri, “A novel dynamic system in the space of spd matrices with applications to appearance tracking,”SIAM Journal on Imaging Sciences, vol. 6, no. 1, pp. 592–615, 2013

    work page 2013

  7. [7]

    Statistical analysis on stiefel and grassmann manifolds with applications in computer vision,

    P. Turaga, A. Veeraraghavan, and R. Chellappa, “Statistical analysis on stiefel and grassmann manifolds with applications in computer vision,” in2008 IEEE Conference on Computer Vision and Pattern Recognition. IEEE, 2008, pp. 1–8

    work page 2008

  8. [8]

    P. K. Turaga and A. Srivastava,Riemannian Computing in Computer Vision. Springer, 2016, vol. 1

    work page 2016

  9. [9]

    Differential privacy,

    C. Dwork, “Differential privacy,” inICALP 2006: Automata, Languages and Programming, Part II, ser. Lecture Notes in Computer Science, vol

    work page 2006

  10. [10]

    Berlin, Heidelberg: Springer, 2006, pp. 1–12

    work page 2006

  11. [11]

    Boosting and differential privacy,

    C. Dwork, G. N. Rothblum, and S. Vadhan, “Boosting and differential privacy,” in2010 IEEE 51st Annual Symposium on Foundations of Computer Science (FOCS), IEEE. IEEE, 2010, pp. 51–60

    work page 2010

  12. [12]

    Differential privacy: A survey of results,

    C. Dwork, “Differential privacy: A survey of results,” inTheory and Applications of Models of Computation, M. Agrawal, D. Du, Z. Duan, and A. Li, Eds. Berlin, Heidelberg: Springer Berlin Heidelberg, 2008, pp. 1–19

    work page 2008

  13. [13]

    R ´enyi differential privacy,

    I. Mironov, “R ´enyi differential privacy,” in2017 IEEE 30th Computer Security Foundations Symposium (CSF). IEEE, 2017, pp. 263–275

    work page 2017

  14. [14]

    Concentrated Differential Privacy

    C. Dwork and G. N. Rothblum, “Concentrated differential privacy,” 2016, arXiv:1603.01887

    work page internal anchor Pith review Pith/arXiv arXiv 2016

  15. [15]

    Random Differential Privacy

    R. Hall, A. Rinaldo, and L. Wasserman, “Random differential privacy,” 2011, arXiv:1112.2680

    work page internal anchor Pith review Pith/arXiv arXiv 2011

  16. [16]

    A global geometric framework for nonlinear dimensionality reduction,

    J. B. Tenenbaum, V . d. Silva, and J. C. Langford, “A global geometric framework for nonlinear dimensionality reduction,”Science, vol. 290, no. 5500, pp. 2319–2323, 2000

    work page 2000

  17. [17]

    Nonlinear dimensionality reduction by locally linear embedding,

    S. T. Roweis and L. K. Saul, “Nonlinear dimensionality reduction by locally linear embedding,”Science, vol. 290, no. 5500, pp. 2323–2326, 2000

    work page 2000

  18. [18]

    Deepshape: Deep- learned shape descriptor for 3d shape retrieval,

    J. Xie, G. Dai, F. Zhu, E. K. Wong, and Y . Fang, “Deepshape: Deep- learned shape descriptor for 3d shape retrieval,”IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 39, no. 7, pp. 1335– 1345, 2016

    work page 2016

  19. [19]

    Differential privacy over riemannian manifolds,

    M. Reimherr, K. Bharath, and C. Soto, “Differential privacy over riemannian manifolds,”Advances in Neural Information Processing Systems, vol. 34, pp. 12 292–12 303, 2021

    work page 2021

  20. [20]

    Differentially private fr´echet mean on the manifold of symmetric positive definite (spd) matrices with log-euclidean metric,

    S. Utpala, P. Vepakomma, and N. Miolane, “Differentially private fr´echet mean on the manifold of symmetric positive definite (spd) matrices with log-euclidean metric,” 2022, arXiv:2208.04245

    work page arXiv 2022

  21. [21]

    Dependent differential privacy for correlated data,

    J. Zhao, J. Zhang, and H. V . Poor, “Dependent differential privacy for correlated data,” in2017 IEEE Globecom Workshops. IEEE, 2017, pp. 1–7

    work page 2017

  22. [22]

    Scenario-based adaptations of differ- ential privacy: A technical survey,

    Y . Zhao, J. T. Du, and J. Chen, “Scenario-based adaptations of differ- ential privacy: A technical survey,”ACM Computing Surveys, vol. 56, no. 8, pp. 1–39, 2024

    work page 2024

  23. [23]

    Privacy preserving data sharing with anonymous id assignment,

    L. A. Dunning and R. Kresman, “Privacy preserving data sharing with anonymous id assignment,”IEEE Transactions on Information Forensics and Security, vol. 8, no. 2, pp. 402–413, 2012

    work page 2012

  24. [24]

    Smooth sensitivity for geo-privacy,

    Y . Liang and K. Yi, “Smooth sensitivity for geo-privacy,” inProceedings of the 2024 on ACM SIGSAC Conference on Computer and Communi- cations Security, ser. CCS ’24. New York, NY , USA: Association for Computing Machinery, 2024, p. 333–347

    work page 2024

  25. [25]

    Conformal transformations of riemannian manifolds,

    M. Obata, “Conformal transformations of riemannian manifolds,”Jour- nal of Differential Geometry, vol. 4, no. 3, pp. 311–333, 1970

    work page 1970

  26. [26]

    Conformal deformation of a riemannian metric to constant scalar curvature,

    R. Schoen, “Conformal deformation of a riemannian metric to constant scalar curvature,”Journal of Differential Geometry, vol. 20, no. 2, pp. 479–495, 1984

    work page 1984

  27. [27]

    Conformal deformation to constant negative scalar curvature on noncompact riemannian manifolds,

    P. Aviles and R. C. McOwen, “Conformal deformation to constant negative scalar curvature on noncompact riemannian manifolds,”Journal of Differential Geometry, vol. 27, no. 2, pp. 225–239, 1988

    work page 1988

  28. [28]

    Francesco, P

    P. Francesco, P. Mathieu, and D. S ´en´echal,Conformal Field Theory. Springer Science & Business Media, 2012

    work page 2012

  29. [29]

    The algorithmic foundations of differential privacy,

    C. Dwork, A. Rothet al., “The algorithmic foundations of differential privacy,”Foundations and Trends® in Theoretical Computer Science, vol. 9, no. 3–4, pp. 211–407, 2014

    work page 2014

  30. [30]

    Gaussian differential privacy on riemannian manifolds,

    Y . Jiang, X. Chang, Y . Liu, L. Ding, L. Kong, and B. Jiang, “Gaussian differential privacy on riemannian manifolds,”Advances in Neural Information Processing Systems, vol. 36, pp. 14 665–14 684, 2023

    work page 2023

  31. [31]

    Applied conformal field theory,

    P. Ginsparg, “Applied conformal field theory,” arXiv preprint hep- th/9108028, 1990

    work page arXiv 1990

  32. [32]

    Riemannian residual neural networks,

    I. Katsman, E. Chen, S. Holalkere, A. Asch, A. Lou, S. N. Lim, and C. M. De Sa, “Riemannian residual neural networks,” inAdvances in Neural Information Processing Systems, A. Oh, T. Naumann, A. Glober- son, K. Saenko, M. Hardt, and S. Levine, Eds., vol. 36. Curran Associates, Inc., 2023, pp. 63 502–63 514

    work page 2023

  33. [33]

    Shape and struc- ture preserving differential privacy,

    C. Soto, K. Bharath, M. Reimherr, and A. Slavkovi ´c, “Shape and struc- ture preserving differential privacy,”Advances in Neural Information Processing Systems, vol. 35, pp. 24 693–24 705, 2022

    work page 2022

  34. [34]

    Gaussian differential privacy,

    J. Dong, A. Roth, and W. J. Su, “Gaussian differential privacy,”Journal of the Royal Statistical Society: Series B (Statistical Methodology), vol. 84, no. 1, pp. 3–37, 2022

    work page 2022

  35. [35]

    Review of

    R. E. Greene, “Review of ”riemannian geometry” by s. gallot, d. hulin, and j. lafontaine,” 1989, book review

    work page 1989

  36. [36]

    J. M. Lee,Introduction to Riemannian Manifolds. Springer, 2018, vol. 2

    work page 2018

  37. [37]

    Lang,Introduction to Differentiable Manifolds

    S. Lang,Introduction to Differentiable Manifolds. Springer Science & Business Media, 2006

    work page 2006

  38. [38]

    Srivastava and E

    A. Srivastava and E. P. Klassen,Functional and Shape Data Analysis. Springer, 2016, vol. 1

    work page 2016

  39. [39]

    del Castillo and Tominich,Differentiable Manifolds

    T. del Castillo and Tominich,Differentiable Manifolds. Springer, 2020

    work page 2020

  40. [40]

    The liouville theorem for conformal maps: old and new,

    A. Ottazzi, “The liouville theorem for conformal maps: old and new,” Seminario Dottorato 2010/11, University of Padua, 2010, available via Citeseer

    work page 2010

  41. [41]

    Weinberg,Gravitation and Cosmology: Principles and Applications of the General Theory of Relativity

    S. Weinberg,Gravitation and Cosmology: Principles and Applications of the General Theory of Relativity. John Wiley & Sons, 2013

    work page 2013

  42. [42]

    Formal privacy for functional data with Gaussian perturbations,

    A. Mirshani, M. Reimherr, and A. Slavkovi ´c, “Formal privacy for functional data with Gaussian perturbations,” inProceedings of the 36th International Conference on Machine Learning, ser. Proceedings of Machine Learning Research, K. Chaudhuri and R. Salakhutdinov, Eds., vol. 97. PMLR, 09–15 Jun 2019, pp. 4595–4604

    work page 2019

  43. [43]

    A statistical framework for differential privacy,

    L. Wasserman and S. Zhou, “A statistical framework for differential privacy,”Journal of the American Statistical Association, vol. 105, no. 489, pp. 375–389, 2010

    work page 2010

  44. [44]

    Benefits and pitfalls of the exponential mechanism with applications to Hilbert spaces and functional PCA,

    J. Awan, A. Kenney, M. Reimherr, and A. Slavkovi ´c, “Benefits and pitfalls of the exponential mechanism with applications to Hilbert spaces and functional PCA,” inProceedings of the 36th International Conference on Machine Learning, ser. Proceedings of Machine Learning Research, K. Chaudhuri and R. Salakhutdinov, Eds., vol. 97. PMLR, 09–15 Jun 2019, pp. 374–384

    work page 2019

  45. [45]

    Geometry of laplace-beltrami operator on a complete riemannian manifold,

    H. Urakawa, “Geometry of laplace-beltrami operator on a complete riemannian manifold,”Progress in Differential Geometry, vol. 22, pp. 347–406, 1993

    work page 1993

  46. [46]

    On estimation of a probability density function and mode,

    E. Parzen, “On estimation of a probability density function and mode,” The Annals of Mathematical Statistics, vol. 33, no. 3, pp. 1065–1076,

    work page

  47. [47]

    Available: http://www.jstor.org/stable/2237880

    [Online]. Available: http://www.jstor.org/stable/2237880

    work page arXiv

  48. [48]

    Beck,Elliptic Regularity Theory, ser

    L. Beck,Elliptic Regularity Theory, ser. Lecture Notes of the Unione Matematica Italiana. Cham: Springer, 2016, vol. 19

    work page 2016

  49. [49]

    Fern ´andez-Real and X

    X. Fern ´andez-Real and X. Ros-Oton,Regularity Theory for Elliptic PDE. EMS Press, 2023

    work page 2023

  50. [50]

    Hilbert,Methods of mathematical physics

    D. Hilbert,Methods of mathematical physics. CUP Archive, 1985

    work page 1985

  51. [51]

    Gilbarg and N

    D. Gilbarg and N. S. Trudinger,Elliptic Partial Differential Equations of Second Order, ser. Classics in Mathematics. Berlin, Heidelberg: Springer Berlin Heidelberg, jan 1997

    work page 1997

  52. [52]

    Jost and J

    J. Jost and J. Jost,Riemannian Geometry and Geometric Analysis. Springer, 2008, vol. 42005

    work page 2008

  53. [53]

    Schwarz,Hodge Decomposition-A Method for Solving Boundary Value Problems

    G. Schwarz,Hodge Decomposition-A Method for Solving Boundary Value Problems. Springer, 2006

    work page 2006

  54. [54]

    An extension of e. Hopf’s maximum principle with an ap- plication to riemannian geometry,

    E. Calabi, “An extension of e. Hopf’s maximum principle with an ap- plication to riemannian geometry,”Duke Mathematical Journal, vol. 25, no. 1, pp. 45–56, March 1958

    work page 1958

  55. [55]

    R. A. Adams and J. J. Fournier,Sobolev Spaces. Elsevier, 2003, vol. 140

    work page 2003

  56. [56]

    Sur le principe de la moyenne et la limite de la moyenne d’une fonction,

    S. Banach and H. Steinhaus, “Sur le principe de la moyenne et la limite de la moyenne d’une fonction,”Studia Mathematica, vol. 2, pp. 50–68, 1930

    work page 1930

  57. [57]

    Bi-lipschitz embeddings into low-dimensional eu- clidean spaces,

    J. Matou ˇsek, “Bi-lipschitz embeddings into low-dimensional eu- clidean spaces,”Commentationes Mathematicae Universitatis Carolinae, vol. 31, no. 3, pp. 589–600, 1990

    work page 1990

  58. [58]

    Chavel,Riemannian Geometry: A Modern Introduction, ser

    I. Chavel,Riemannian Geometry: A Modern Introduction, ser. Cam- bridge Tracts in Mathematics. Cambridge: Cambridge University Press, 1995, vol. 108

    work page 1995

  59. [59]

    Topping,Lectures on the Ricci Flow, ser

    P. Topping,Lectures on the Ricci Flow, ser. London Mathematical Society Lecture Note Series. Cambridge: Cambridge University Press, 2006, vol. 325

    work page 2006

  60. [60]

    Rosenberg,The Laplacian on a Riemannian Manifold: An Introduc- tion to Analysis on Manifolds, ser

    S. Rosenberg,The Laplacian on a Riemannian Manifold: An Introduc- tion to Analysis on Manifolds, ser. Cambridge Tracts in Mathematics. Cambridge: Cambridge University Press, 1997, vol. 31

    work page 1997

  61. [61]

    Billingsley,Probability and measure

    P. Billingsley,Probability and measure. John Wiley & Sons, 2017

    work page 2017

  62. [62]

    Improving the Gaussian mechanism for differential privacy: Analytical calibration and optimal denoising,

    B. Balle and Y .-X. Wang, “Improving the Gaussian mechanism for differential privacy: Analytical calibration and optimal denoising,” in Proceedings of the 35th International Conference on Machine Learning, ser. Proceedings of Machine Learning Research, J. Dy and A. Krause, Eds., vol. 80. PMLR, 10–15 Jul 2018, pp. 394–403

    work page 2018

  63. [63]

    Petersen,Riemannian Geometry

    P. Petersen,Riemannian Geometry. Springer, 2006, vol. 171

    work page 2006

  64. [64]

    Riemannian center of mass and mollifier smoothing,

    H. Karcher, “Riemannian center of mass and mollifier smoothing,” Communications on pure and applied mathematics, vol. 30, no. 5, pp. 509–541, 1977

    work page 1977

  65. [65]

    Probability, convexity, and harmonic maps with small image i: uniqueness and fine existence,

    W. S. Kendall, “Probability, convexity, and harmonic maps with small image i: uniqueness and fine existence,”Proceedings of the London Mathematical Society, vol. 3, no. 2, pp. 371–406, 1990

    work page 1990

  66. [66]

    Intrinsic statistics on riemannian manifolds: Basic tools for geometric measurements,

    X. Pennec, “Intrinsic statistics on riemannian manifolds: Basic tools for geometric measurements,”Journal of Mathematical Imaging and Vision, vol. 25, pp. 127–154, 2006

    work page 2006

  67. [67]

    The composition theorem for dif- ferential privacy,

    P. Kairouz, S. Oh, and P. Viswanath, “The composition theorem for dif- ferential privacy,” inProceedings of the 32nd International Conference on Machine Learning, ser. Proceedings of Machine Learning Research, F. Bach and D. Blei, Eds., vol. 37. Lille, France: PMLR, 07–09 Jul 2015, pp. 1376–1385

    work page 2015

  68. [68]

    Empirical graph laplacian approximation of laplace-beltrami operators: Large sample results,

    E. Gin ´e and V . Koltchinskii, “Empirical graph laplacian approximation of laplace-beltrami operators: Large sample results,” inHigh Dimen- sional Probability, ser. Lecture Notes–Monograph Series. Institute of Mathematical Statistics, 2006, vol. 49, pp. 238–259

    work page 2006

  69. [69]

    Moments of von mises and fisher distributions and applications,

    T. Hillen, K. J. Painter, A. C. Swan, and A. D. Murtha, “Moments of von mises and fisher distributions and applications,”Mathematical biosciences and engineering, vol. 14, no. 3, pp. 673–694, 2017

    work page 2017

  70. [70]

    Simulation of the von mises fisher distribution,

    A. T. Wood, “Simulation of the von mises fisher distribution,”Commu- nications in Statistics - Simulation and Computation, vol. 23, no. 1, pp. 157–164, 1994

    work page 1994

  71. [71]

    Learning multiple layers of features from tiny images,

    A. Krizhevsky, “Learning multiple layers of features from tiny images,” University of Toronto, Technical Report TR-2009, 2009

    work page 2009

  72. [72]

    Fashion-MNIST: a Novel Image Dataset for Benchmarking Machine Learning Algorithms

    H. Xiao, K. Rasul, and R. V ollgraf, “Fashion-mnist: A novel im- age dataset for benchmarking machine learning algorithms,” 2017, arXiv:1708.07747

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  73. [73]

    Principal geodesic analysis for the study of nonlinear statistics of shape,

    P. T. Fletcher, C. Lu, S. M. Pizer, and S. Joshi, “Principal geodesic analysis for the study of nonlinear statistics of shape,”IEEE Transactions on Medical Imaging, vol. 23, no. 8, pp. 995–1005, 2004

    work page 2004

  74. [74]

    N. I. Fisher,Statistical Analysis of Circular Data. Cambridge University Press, 1995

    work page 1995

  75. [75]

    Concentrated differential privacy: Simplifica- tions, extensions, and lower bounds,

    M. Bun and T. Steinke, “Concentrated differential privacy: Simplifica- tions, extensions, and lower bounds,” inProceedings, Part I, of the 14th International Conference on Theory of Cryptography - Volume 9985. Berlin, Heidelberg: Springer-Verlag, 2016, p. 635–658

    work page 2016

  76. [76]

    Schauder estimates for elliptic and parabolic equa- tions,

    X.-J. Wanget al., “Schauder estimates for elliptic and parabolic equa- tions,”Chinese Annals of Mathematics-Series B, vol. 27, no. 6, p. 637, 2006

    work page 2006

  77. [77]

    A generalization of the gamma distribution,

    E. W. Stacy, “A generalization of the gamma distribution,”The Annals of Mathematical Statistics, vol. 33, no. 3, pp. 1187–1192, 1962. [Online]. Available: http://www.jstor.org/stable/2237889

    work page arXiv 1962

  78. [78]

    Geomstats: A python package for riemannian geometry in machine learning,

    N. Miolane, A. L. Brigant, J. Mathe, B. Hou, N. Guigui, Y . Thanwerdas, S. Heyder, O. Peltre, N. Koep, H. Zaatiti, H. Hajri, Y . Cabanes, T. Gerald, P. Chauchat, C. Shewmake, B. Kainz, C. Donnat, S. P. Holmes, and X. Pennec, “Geomstats: A python package for riemannian geometry in machine learning,” 2020, arXiv:2004.04667. APPENDIX A. Proof for Theorem III...

    work page arXiv 2020