Pith. sign in

REVIEW 3 major objections 6 minor 61 references

A structural condition on the empirical cost function yields uniform convergence of generalized conditional Fréchet means without checking equicontinuity of the estimator.

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 19:37 UTC pith:O3CZT4MJ

load-bearing objection Clean alternative route to uniform Fréchet-mean-set convergence that actually works for the usual geometries, plus usable MoM and exceedance extensions. the 3 major comments →

arxiv 2607.26837 v1 pith:O3CZT4MJ submitted 2026-07-29 stat.ME math.STstat.TH

Uniform Convergence of Generalized Conditional Fr\'echet Means with Applications to Weighted Fr\'echet Aggregation and Exceedance Set Estimation

Houren Hong , Jiazhen Xu , Andrew T.A. Wood This is my paper
classification stat.ME math.STstat.TH MSC 62R2062G2062G05
keywords Fréchet meanFréchet regressionuniform consistencymetric spacesmedian of meansdistributed regressionexceedance setobject-oriented data
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.

Object-valued data live in metric spaces that lack linear structure, so standard proofs that Fréchet-regression estimators converge uniformly over covariates become intractable: one cannot easily verify asymptotic uniform equicontinuity of the estimator itself. This paper replaces that check with a verifiable condition on the empirical cost function (local convexity along a geodesic in an ambient convex set). Once the condition holds, uniform consistency of the estimator set follows from ordinary uniform laws of large numbers on the cost. The same foundation delivers two practical tools for non-Euclidean data: a weighted aggregation scheme that supports both distributed computing and median-of-means robustness, and a consistent exceedance-set procedure that locates covariate regions where the conditional Fréchet mean exceeds a threshold and quantifies the total magnitude of the exceedance. Simulations on graph Laplacians and an analysis of New York City bike-share networks illustrate the methods.

Core claim

Under a novel structural condition (R0) on the empirical cost—existence of a convex ambient subset containing both a population and an empirical minimizer, together with at least one intermediate geodesic point at which the cost satisfies a convex-combination inequality—and a standard identifiability condition, the empirical generalized conditional Fréchet mean set converges uniformly to its population counterpart whenever the empirical cost converges uniformly. The argument never requires asymptotic uniform equicontinuity of the estimator itself.

What carries the argument

Condition (R0) together with Proposition 1: the condition converts a strictly positive lower bound on the empirical cost gap outside a neighborhood of the population mean set into a Hausdorff-distance bound between the empirical and population mean sets, thereby reducing uniform estimator consistency to a uniform law of large numbers on the cost.

Load-bearing premise

With high probability there must exist a convex piece of an ambient space that contains both a true mean and a sample mean, plus at least one intermediate point on the geodesic between them where the sample cost obeys a simple weighted-average inequality.

What would settle it

Construct a compact metric space and a sequence of empirical costs for which (R0) fails (no usable convex ambient subset or the convex-combination inequality is violated at every intermediate point) while the cost still converges uniformly; if the estimator then fails to converge uniformly, the central claim is refuted.

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

Share X Bluesky LinkedIn Reddit HN

If this is right

  • Global, local, kernel and k-NN Fréchet regression all inherit uniform consistency once (R0) is verified for the underlying metric space.
  • Block-wise Fréchet estimators can be aggregated by weighted Fréchet means or medians; the aggregate remains uniformly consistent even under heterogeneous block distributions.
  • Median-of-means Fréchet regression admits an explicit finite-sample deviation bound and a concrete rule for choosing the number of blocks at a prescribed confidence level.
  • Exceedance sets of a continuous functional of the conditional Fréchet mean, together with the normalized Lebesgue measure of those sets, are consistently estimable from any uniformly consistent Fréchet regression procedure.

Where Pith is reading between the lines

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

  • The same cost-side argument should extend immediately to other M-estimators on metric spaces (e.g., Fréchet quantiles or robust location functionals) whenever a local geodesic convexity inequality can be checked.
  • Because the condition only needs one intermediate geodesic point rather than full local convexity, it may remain valid on spaces with mild positive curvature that still admit short unique geodesics.
  • The explicit block-selection formula for MoM Fréchet regression supplies a practical default that existing geometric-median literature has lacked, and could be stress-tested on high-dimensional covariance or tree-valued responses.

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 develops a framework for uniform convergence of generalized conditional Fréchet mean sets over a compact predictor space, based on a structural condition (R0) on the empirical cost that links cost margins to one-sided Hausdorff distance (Proposition 1) and thereby yields Theorems 1–2 without verifying asymptotic uniform equicontinuity of the estimator itself. Condition (R0) is checked for Euclidean convex sets, compact Riemannian manifolds (with injectivity-radius control when weights may be negative), extrinsic embeddings, and Hadamard spaces/manifolds for global, local, kernel, k-NN, and deep Fréchet regression weights (Propositions 2–6). These guarantees are used to obtain uniform consistency for the listed Fréchet regression estimators (Theorems 3–6), a weighted aggregation scheme covering distributed (ℓ=2) and median-of-means (ℓ=1) Fréchet regression (Theorems 7–8, Corollary 1), and consistency of exceedance sets and aggregate exceedance measures (Propositions 7–8). Monte Carlo experiments on graph Laplacians and a Citi Bike network application illustrate the methods.

Significance. Uniform consistency of conditional Fréchet means is a recurring bottleneck in object-oriented data analysis because metric-space estimators typically lack closed forms, making equicontinuity arguments intractable. The reduction via (R0) plus a uniform LLN on the cost is a clean and reusable device; the verifications for standard spaces and weight schemes are concrete. The MoM concentration bound with an explicit block-count rule (Corollary 1) and the exceedance-set extension beyond Euclidean/Hilbert responses are useful downstream contributions that rest directly on the uniform-convergence foundation. Supplement proofs follow a transparent contrapositive-plus-empirical-process pattern, and the simulations align with the stated rates and robustness claims. If the arguments hold as written, the paper supplies a standard reference tool for Fréchet regression theory and two immediately applicable methodologies.

major comments (3)
  1. [Section 6, Propositions 2–6] Section 6 (and S6): all Monte Carlo designs and the real-data example use graph Laplacians under the Frobenius metric, which reduce to Euclidean structure via half-vectorization (as noted after Proposition 2). The theory is advertised for compact Riemannian manifolds and Hadamard spaces (Propositions 3–6), yet no finite-sample check is given on a genuinely curved or non-positively curved space (e.g., SPD matrices with affine-invariant metric, spheres, or a simple tree/CAT(0) example). A single such experiment would substantially strengthen the claim that the framework is useful beyond the Euclidean-reducible case.
  2. [Corollary 1, Section 4.2] Corollary 1: the explicit block choice B(δ,τ) and deviation bound U(n,δ,τ) inherit unknown constants D and rate exponents υ_j from Petersen & Müller (2019) and Chen & Müller (2022). The corollary is presented as giving a practical, tuning-free rule, but without guidance on calibrating D or verifying the block-wise rate conditions under contamination, the finite-sample guarantee is not operational. A short remark on how a user should choose or bound these quantities (or a sensitivity check in the simulations) is needed for the claim to be usable as stated.
  3. [Section 2, Condition (R0)] Condition (R0)(ii)–(iii) is the load-bearing structural premise (Definition/Condition (R0), Proposition 1). Propositions 2–6 verify it for listed model classes, but the paper gives little discussion of when it fails (highly nonconvex empirical costs, discrete Ω without a usable ambient convex structure, or negative-weight regimes outside the injectivity-radius bound in Proposition 3). A brief limitations paragraph or a concrete counter-example space would clarify the scope of Theorems 1–2 and all downstream results that invoke them.
minor comments (6)
  1. [Supplementary Material title] Supplement title uses “Fréchet Moment-of-Mean” while the main text consistently uses median-of-means (MoM); align terminology.
  2. [Section 4.2] Typo “collorary” for “corollary” in the paragraph preceding Corollary 1.
  3. [Section 3, Model 3.5] Model 3.5 (deep Fréchet regression) is included in the (R0) verifications but receives no separate uniform-convergence theorem analogous to Theorems 3–6; a sentence clarifying that the general Theorems 1–2 apply once the network weights satisfy the stated conditions would help.
  4. [Theorem 3, Remark 1] In Theorem 3 / Remark 1 the residual bias term sup d(er(x), r(x)) is correctly flagged; consider stating more explicitly in the main text when global Fréchet is consistent for the true conditional Fréchet mean (e.g., under correct linear specification).
  5. [Sections 6–7] Figure 1–3 captions and Table 1–3 are clear; ensure axis labels and the definition of network size (sum of adjacency entries) appear in the main text near first use.
  6. [Throughout] Minor notation drift: bm(x) vs br(x), cM vs bR, and κκκ vs κ; a single pass for consistency would help the reader.

Circularity Check

1 steps flagged

No significant circularity: uniform convergence is reduced to a stated structural condition (R0) plus a uniform LLN on costs, not defined from the target.

specific steps
  1. self citation load bearing [§2.1 comparison with Xu et al. (2025); Corollary 1 rate exponents]
    "Importantly, Condition (R0), under the additional uniqueness assumption, is still weaker than Assumption (A3) in Xu et al. (2025). We do not require the entire space Ω to be a convex metric space, nor do we restrict d and d_C to be identical metrics."

    Minor only: Xu–Wood overlap appears in a comparison claiming (R0) is weaker than their prior (A3), and Corollary 1 imports υ_j from Petersen–Müller/Chen–Müller. Neither step defines the uniform-convergence claim; the main theorems stand on (R0)+(R1)+cost LLN without needing those citations to close the argument. Not load-bearing circularity.

full rationale

The paper’s central chain is a legitimate sufficient-condition argument, not a closed loop. Proposition 1 links positive empirical-cost margins outside a neighborhood of m(x) to one-sided Hausdorff closeness of bm(x) to m(x) under (R0); Theorems 1–2 then obtain sup_x d(bm(x), m(x)) = o_p(1) once that margin condition (or (R1) plus uniform cost convergence) holds. (R0) is an assumption verified on model classes (Propositions 2–6), not defined in terms of the uniform-convergence conclusion. Downstream aggregation (Theorem 7), MoM bounds (Theorem 8, Corollary 1), and exceedance consistency (Propositions 7–8) inherit uniform consistency in the standard way. Self-citations (e.g. comparison to Xu et al. 2025 on a weaker convexity assumption; rate exponents from Petersen–Müller / Chen–Müller in Corollary 1) are background or external rate plugs, not load-bearing uniqueness theorems that force the result. No fitted quantity is relabeled as a first-principles prediction. Score 1 only for minor overlapping-author comparison that is not self-sealing.

Axiom & Free-Parameter Ledger

4 free parameters · 7 axioms · 2 invented entities

The central uniform-convergence claim rests on the new structural Condition (R0), standard M-estimation identifiability (R1), and uniform convergence of the empirical cost—plus model-specific regularity (kernels, densities, VC balls, injectivity-radius bounds). Downstream MoM and exceedance results add homogeneity of clean blocks, a contamination fraction τ<1/2, Lipschitz-type control on the scalar functional Λ, and a margin condition on Δ. Tuning choices (h, k, B) are free parameters in applications but not fitted to prove the theorems.

free parameters (4)
  • Bandwidth h (local/kernel Fréchet)
    Must satisfy h→0 and nh/log n→∞; practical choice not specified by a data-driven rule in the theory sections.
  • Number of neighbors k (kNN Fréchet)
    Must satisfy k→∞, k/n→0, k/log n→∞; finite-sample choice left open.
  • Number of blocks B and weights κ in aggregation/MoM = Simulations: B up to 100, κ_b=1/B; real data: B=80 weekday, B=40 weekend
    Corollary 1 gives B:=⌊log(1/δ)/((1−τ)ι(τ))⌋+1 for prescribed confidence; simulations fix B∈{60,80,100} or B=100 and equal weights by hand.
  • Exceedance threshold ρ (and percentile in real data) = 70th percentile in §7; ρ=8 in simulation §6.2
    User-chosen; real-data analysis uses the 70th percentile of estimated weekday–weekend Fréchet distance.
axioms (7)
  • ad hoc to paper Condition (R0): existence of empirical means; totally bounded Ω embeddable in a Polish space with convex subsets C_m meeting bm(x); existence of an intermediate point where empirical cost obeys a convex-combination inequality.
    Novel structural condition that replaces equicontinuity; verified for specific spaces in Props. 2–6 but postulated as the general sufficient condition in §2.
  • domain assumption Condition (R1): population cost separates points outside any ζ-neighborhood of m(x) uniformly in x.
    Standard M-estimation identifiability; routinely used in Fréchet regression literature.
  • domain assumption i.i.d. observations and standard kernel/density/VC regularity (U1)–(U5) for local, kernel, and kNN models.
    Classic nonparametric assumptions (Silverman/Mack–Silverman type; small-ball and VC for balls).
  • domain assumption For Riemannian cases: responses supported in a ball of radius < half injectivity radius; extra bound on negative-weight mass for global/local Fréchet.
    Ensures differentiability / avoidance of cut locus in Proposition 3.
  • domain assumption Block-wise clean estimators satisfy a uniform deviation bound with probability ≤p; contamination total weight τ < α−p/(1−p).
    Theorem 8 MoM robustness setup; extends Minsker-type contamination models.
  • domain assumption (L1): functional Λ is Hölder/Lipschitz in the product metric so level sets inherit uniform consistency.
    Standard level-set regularity (Willett–Nowak type), used in Propositions 7–8.
  • standard math Metric-space and geodesic/Hadamard/manifold background (convex metric spaces, Hausdorff distance on sets, Berge maximum theorem for continuity of rb).
    Used throughout §§2–4; cited via Khamsi–Kirk, Bridson–Haefliger, Boumal, etc.
invented entities (2)
  • Condition (R0) structural framework for uniform convergence of Fréchet mean sets independent evidence
    purpose: Bypass verification of asymptotic uniform equicontinuity of bm by linking empirical cost margins to one-sided Hausdorff distance.
    The paper's main theoretical device; not an external physical entity but a new sufficient condition package. Independent checks are the verifications in Props. 2–6 on known spaces.
  • Weighted Fréchet aggregation of order ℓ (distributed ℓ=2 and MoM ℓ=1) independent evidence
    purpose: Unify distributed and median-of-means Fréchet regression on metrically convex spaces.
    Methodological construction; special cases recover known Euclidean ideas but the metric-space theory and explicit B rule are new here.

pith-pipeline@v1.2.0-daily-grok45 · 48703 in / 4334 out tokens · 78538 ms · 2026-07-30T19:37:06.798593+00:00 · methodology

0 comments
Cite this review

Pith. "Pith review of Uniform Convergence of Generalized Conditional Fr\'echet Means with Applications to Weighted Fr\'echet Aggregation and Exceedance Set Estimation." pith.science (2026). https://pith.science/paper/O3CZT4MJ

@misc{pith2026260726837,
  author       = {Pith},
  title        = {Pith review of: Uniform Convergence of Generalized Conditional Fr\'echet Means with Applications to Weighted Fr\'echet Aggregation and Exceedance Set Estimation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/O3CZT4MJ}},
  note         = {Machine review of arXiv:2607.26837}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

The statistical analysis of object oriented data in non-Euclidean spaces heavily relies on generalized conditional Fr\'echet means, notably in the context of Fr\'echet regression. However, establishing the uniform convergence of these estimators presents several theoretical challenges. The difficulties are caused primarily by the absence of linear structures in general metric spaces, rendering standard techniques for verifying the asymptotic uniform equicontinuity of the estimator largely intractable. To overcome this limitation, this paper introduces an alternative theoretical framework for establishing uniform convergence that bypasses the need to verify uniform equicontinuity, under a novel structural condition on the empirical cost function of the generalized conditional Fr\'echet means. We demonstrate that this analytical condition is satisfied by various prominent Fr\'echet regression models across broad classes of metric spaces. Leveraging these foundational uniform convergence guarantees, we subsequently extend two widely used frameworks from Euclidean to non-Euclidean spaces: (i) a weighted Fr\'echet aggregation framework that facilitates both distributed regression and robust median-of-means regression; and (ii) an exceedance set estimation framework to identify critical covariate regions where the conditional generalized Fr\'echet mean surpasses a prescribed threshold, alongside a metric to quantify the aggregate magnitude of the exceedance. The theoretical properties of these proposed methods are empirically validated through Monte Carlo simulations and an application to dynamic transportation networks in New York City.

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

61 extracted references · 3 linked inside Pith

  1. [1]

    Abdelhakim, A. A. (2016). A convexity of functions on convex metric spaces of Takahashi and applications. Journal of the Egyptian Mathematical Society , 24(3), 348--354

    2016

  2. [2]

    Arnaudon, M., Barbaresco, F., and Yang, L. (2012). Medians and means in Riemannian geometry: existence, uniqueness and computation. In Matrix Information Geometry , pages 169--197. Springer

    2012

  3. [3]

    Bhattacharya, Rabindra N and Ellingson, L and Liu, X and Patrangenaru, V and Crane, M. (2012). Extrinsic analysis on manifolds is computationally faster than intrinsic analysis with applications to quality control by machine vision . Applied Stochastic Models in Business and Industry, 28(3), 222--235

    2012

  4. [4]

    Bortolussi, N., Durand, E., Blum, M., and Fran c ois, O. (2006). apTreeshape: statistical analysis of phylogenetic tree shape. Bioinformatics, 22(3), 363--364

    2006

  5. [5]

    Boumal, N. (2023). An Introduction to Optimization on Smooth Manifolds. Cambridge University Press

    2023

  6. [6]

    Bridson, M. R. and Haefliger, A. (2011). Metric Spaces of Non-Position Curvature. Springer

    2011

  7. [7]

    Capitaine, L., Bigot, J., Thi \'e baut, R., and Genuer, R. (2024). Fr \'e chet random forests for metric space valued regression with non Euclidean predictors. Journal of Machine Learning Research, 25(355), 1--41

    2024

  8. [8]

    and M \"u ller, H.-G

    Chen, Y. and M \"u ller, H.-G. (2022). Uniform convergence of local Fr \'e chet regression with applications to locating extrema and time warping for metric space valued trajectories. The Annals of Statistics , 50(3), 1573--1592

    2022

  9. [9]

    X., and Peng, L

    Chen, S. X., and Peng, L. (2021). Distributed statistical inference for massive data. The Annals of Statistics , 49(5), 2851--2869

    2021

  10. [10]

    C., Fletcher, P

    Davis, B. C., Fletcher, P. T., Bullitt, E. and Joshi, S. (2010). Population shape regression from random design data. International Journal of Computer Vision , 90(2), 255--266

    2010

  11. [11]

    Devroye, L., Lerasle, M., Lugosi, G., and Oliveira, R. I. (2016). Sub-Gaussian mean estimators. The Annals of Statistics , 44, 2695--2725

    2016

  12. [12]

    Dobriban, E., and Sheng, Y. (2021). Distributed linear regression by averaging. The Annals of Statistics , 49(2), 918--943

    2021

  13. [13]

    Dong, Y., Lu, Y., Wu, Y., and M \"u ller, H.-G. (2026). Variable selection for Nadaraya--Watson--Fr \'e chet regression. Journal of Nonparametric Statistics, 1--26

    2026

  14. [14]

    Dryden, I. L. and Mardia, K. V. (2016). Statistical Shape Analysis: With Applications in R. John Wiley & Sons

    2016

  15. [15]

    Evans, S. N. and Jaffe, A. Q. (2024). Limit theorems for Fr \'e chet mean sets. Bernoulli, 30(1), 419--447

    2024

  16. [16]

    T., Venkatasubramanian, S., and Joshi, S

    Fletcher, P. T., Venkatasubramanian, S., and Joshi, S. (2009). The geometric median on Riemannian manifolds with application to robust atlas estimation. NeuroImage , 45(1), S143--S152

    2009

  17. [17]

    Fr \'e chet, M. (1948). Les \'e l \'e ments al \'e atoires de nature quelconque dans un espace distanci \'e . Annales de l'institut Henri Poincar \'e , 10(4), 215--310

    1948

  18. [18]

    Gy \"o rfi, L., Humbert, P., and Le Bars, B. (2026). Metric space valued Fr \'e chet regression. arXiv preprint arXiv:2602.05225

    arXiv 2026

  19. [19]

    C., and Song, P

    Hector, E. C., and Song, P. X. (2020). Doubly distributed supervised learning and inference with high-dimensional correlated outcomes. Journal of Machine Learning Research , 21(1), 6983--7017

    2020

  20. [20]

    C., and Song, P

    Hector, E. C., and Song, P. X. (2021). A distributed and integrated method of moments for high-dimensional correlated data analysis. Journal of American Statistical Association , 116(534), 805--818

    2021

  21. [21]

    Huckemann, S. F. (2011). Intrinsic inference on the mean geodesic of planar shapes and tree discrimination by leaf growth. The Annals of Statistics, 39(2), 1098--1124

    2011

  22. [22]

    I., Zhou, Y., and M \"u ller, H.-G

    Iao, S. I., Zhou, Y., and M \"u ller, H.-G. (2025). Deep Fr \'e chet regression. Journal of the American Statistical Association, 120(551), 1437--1448

    2025

  23. [23]

    Khamsi, M. A. and Kirk, W. A. (2011). An Introduction to Metric Spaces and Fixed Point Theory. John Wiley & Sons

    2011

  24. [24]

    Kim, J., Park, J., and Bhattacharya, A. (2025). Robust estimation in metric spaces: achieving exponential concentration with a Fr \'e chet median. In International Conference on Artificial Intelligence and Statistics , pages 4222--4230. PMLR

    2025

  25. [25]

    Kolaczyk, E. D. and Cs \'a rdi, G. (2014). Statistical Analysis of Network Data with R. Springer

    2014

  26. [26]

    and M \"u ller, H.-G

    Kundu, P., Zhou, H. and M \"u ller, H.-G. (2025). Exceedance and force of centrality for functional data. arXiv preprint arXiv:2504.02296

    Pith/arXiv arXiv 2025

  27. [27]

    and Otsu, T

    Kurisu, D. and Otsu, T. (2025). Model averaging for global Fr\' e chet regression. Journal of Multivariate Analysis , 207, 105416

    2025

  28. [28]

    and Jung, S

    Lee, J. and Jung, S. (2026). General M-estimators of location on Riemannian manifolds: existence and uniqueness. Statistics & Probability Letters , 110670

    2026

  29. [29]

    Lee, J., and Jung, S. (2026). Huber means on Riemannian manifolds. Journal of the Royal Statistical Society Series B: Statistical Methodology , 88(2), 444--463

    2026

  30. [30]

    Li, H., Sugasawa, S., and Katayama, S. (2025). Robust global Fr \'e chet regression via weight regularization. arXiv preprint arXiv:2511.03694

    arXiv 2025

  31. [31]

    A selective review on statistical methods for massive data computation: distributed computing, subsampling, and minibatch techniques

    Li, X., Gao, Y., Chang, H., Huang, D., Ma, Y., Pan, R., Qi, H., Wang, F., Wu, S., Xu, K., Zhou, J., Zhu, X., Zhu, Y., and Wang, H.(2024). A selective review on statistical methods for massive data computation: distributed computing, subsampling, and minibatch techniques. Statistical Theory and Related Fields , 8(3), 163--185

    2024

  32. [32]

    Thomas, B., Zhu, H., and Dunson, D

    Lin, L., St. Thomas, B., Zhu, H., and Dunson, D. B. (2017). Extrinsic local regression on manifold-valued data. Journal of the American Statistical Association , 112(519), 1261--1273

    2017

  33. [33]

    and Dunson, D

    Lin, L., Lazar, D., Saparbayeva, B. and Dunson, D. (2024). Robust optimisation and inference on manifolds. Statistica Sinica , 34(3), 1299--1323

    2024

  34. [34]

    Lugosi, G., and Mendelson, S. (2019). Mean estimation and regression under heavy-tailed distributions: A survey. Foundations of Computational Mathematics , 19(5), 1145-1190

    2019

  35. [35]

    Mack, Y. P. and Silverman, B. W. (1982). Weak and strong uniform consistency of kernel regression estimates. Zeitschrift f \"u r Wahrscheinlichkeitstheorie und verwandte Gebiete, 61(3), 405--415

    1982

  36. [36]

    and Polonik, W

    Mammen, E. and Polonik, W. (2013). Confidence regions for level sets. Journal of Multivariate Analysis, 122, 202--214

    2013

  37. [37]

    A., Martin, R

    Maronna, R. A., Martin, R. D., Yohai, V. J., and Salibi \'a n-Barrera, M. (2019). Robust Statistics: Theory and Methods (with R). John Wiley & Sons

    2019

  38. [38]

    Marron, J. S. and Alonso, A. M. (2014). Overview of object oriented data analysis. Biometrical Journal , 56(5), 732--753

    2014

  39. [39]

    Marron, J. S. and Dryden, I.L. (2021). Object Oriented Data Analysis. CRC Press, Taylor and Francis Group

    2021

  40. [40]

    Minsker, S. (2015). Geometric median and robust estimation in Banach spaces. Bernoulli , 21(4), 2308-2335

    2015

  41. [41]

    and Jung, S

    Park, J. and Jung, S. (2026). Generalized Fr \'e chet Means with random minimizing domains and its strong consistency. Biometrika , asag002

    2026

  42. [42]

    and M \"u ller, H.-G

    Petersen, A. and M \"u ller, H.-G. (2019). Fr \'e chet regression for random objects with Euclidean predictors. The Annals of Statistics , 47(2), 691--719

    2019

  43. [43]

    and Lin, Z

    Qiu, R., Yu, Z. and Lin, Z. (2024). Semi-supervised Fr \'e chet regression. arXiv preprint arXiv:2404.10444

    Pith/arXiv arXiv 2024

  44. [44]

    Qiu, R., Yu, Z., and Zhu, R. (2024). Random forest weighted local Fr \'e chet regression with random objects. Journal of Machine Learning Research, 25(107), 1--69

    2024

  45. [45]

    and Vert, R

    Rigollet, P. and Vert, R. (2009). Optimal rates for plug-in estimators of density level sets. Bernoulli , 15(4), 1154--1178

    2009

  46. [46]

    D., and Nadler, B

    Rosenblatt, J. D., and Nadler, B. (2016). On the optimality of averaging in distributed statistical learning. Information and Inference: A Journal of the IMA , 5(4), 379--404

    2016

  47. [47]

    Sch \"o tz, C. (2022). Strong laws of large numbers for generalizations of Fr \'e chet mean sets. Statistics , 56(1), 34--52

    2022

  48. [48]

    Silverman, B. W. (1978). Weak and strong uniform consistency of the kernel estimate of a density and its derivatives. The Annals of Statistics , 6(1), 177--184

    1978

  49. [49]

    Song, W., Dubey, P., M \"u ller, H.-G., and Petersen, A. (2026). Inference for Fr \'e chet regression. arXiv preprint arXiv:2605.19519

    Pith/arXiv arXiv 2026

  50. [50]

    van der Vaart, A. (1998). Asymptotic Statistics. Cambridge University Press

    1998

  51. [51]

    and Wellner, J

    van der Vaart, A. and Wellner, J. (1996). Weak Convergence of Empirical Processes: With Applications to Statistics. New York: Springer

    1996

  52. [52]

    Wainwright, M. J. (2019). High-Dimensional Statistics: A Non-Asymptotic Viewpoint. Cambridge University Press

    2019

  53. [53]

    Willett, R. M. and Nowak, R. D. (2007). Minimax optimal level-set estimation. IEEE Transactions on Image Processing, 16(12), 2965--2979

    2007

  54. [54]

    Wu, S., Zhu, X., and Wang, H. (2023). Subsampling and jackknifing: a practically convenient solution for large data analysis with limited computational resources Statistica Sinica, 33(3), 2041--2064

    2023

  55. [55]

    Xu, J., Wood, A. T. A., and Zou, T. (2025). Robust functional principal component analysis for non-Euclidean random objects. Biometrika , asaf040

    2025

  56. [56]

    Xu, J., Wood, A. T. A., and Zou, T. (2025). Quantifying Periodicity in Non-Euclidean Random Objects. arXiv preprint arXiv:2510.18247

    arXiv 2025

  57. [57]

    Yun, H., and Park, B. U. (2023). Exponential concentration for geometric-median-of-means in non-positive curvature spaces. Bernoulli , 29(4), 2927-2960

    2023

  58. [58]

    I., and M \"u ller, H.-G

    Zhou, Y., Iao, S. I., and M \"u ller, H.-G. (2026). End-to-end deep learning for predicting metric space-valued outputs. Journal of Machine Learning Research, 27(132), 1--38

    2026

  59. [59]

    and M \"u ller, H.-G

    Zhou, Y. and M \"u ller, H.-G. (2022). Network regression with graph Laplacians. Journal of Machine Learning Research , 23(320), 1--41

    2022

  60. [60]

    Ziezold, H. (1977). On expected figures and a strong law of large numbers for random elements in quasi-metric spaces. In Transactions of the Seventh Prague Conference on Information Theory, Statistical Decision Functions, Random Processes and of the 1974 European Meeting of Statisticians , pp. 591--602. Springer

    1977

  61. [61]

    Thomas, B., Zhu, H

    Lin, L., St. Thomas, B., Zhu, H. and Dunson, D. B. (2017). Extrinsic local regression on manifold-valued data. Journal of the American Statistical Association , 112(519), 1261--1273

    2017