Huber-Wasserstein barycenters for robust distribution-valued data

Alberto Gonz\'alez-Sanz; Carlos Cardoso-Perell\'o

REVIEW 2 major objections 3 minor 110 references

Huber-Wasserstein barycenters for robust distribution-valued data

T0 review · 2 major / 3 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read The paper claims that a robust barycenter for distribution-valued data can be built by applying the Huber loss to individual transport displacements inside the optimal transport cost, with existence, consistency, breakdown point ~1/2, and…

desk verdict New construction, solid core theory, but Proposition 3.16 has a formula/proof mismatch that needs fixing before the localized-contamination claim is trusted. read the letter →

arxiv 2608.13131 v1 pith:ZF3A3OPB submitted 2026-08-13 stat.ME math.PRstat.ML

Carlos Cardoso-Perelló , Alberto González-Sanz This is my paper · ORCID

classification stat.MEmath.PRstat.ML MSC 62G3562G30

keywords HuberlossWassersteinbarycenteroptimaltransportrobuststatisticsdistribution-valueddatabreakdownpointinfluencefunctionquantilefunctions

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper tries to establish that a robust center for a collection of probability distributions can be obtained by putting the Huber loss inside the optimal transport cost, one displacement at a time, rather than applying it to the Wasserstein distance after optimization. The proposed Huber–Wasserstein barycenter minimizes $\Gamma_{c,P}(\nu)=\int(T_{\rho_c}(\nu,\mu)-c\int\|x\|\,d\mu)\,dP(\mu)$, where $T_{\rho_c}$ is the optimal transport cost with ground cost $\rho_c$. The paper proves this minimizer exists, is bounded and convex, is consistent under both one-stage and two-stage sampling, and has a finite-sample breakdown point essentially $1/2$. As $c$ varies it interpolates between the classical Wasserstein mean and the $L^1$-type Wasserstein median, and in one dimension its quantile function solves a scalar Huber location problem at every level, giving a bounded influence function and an explicit efficiency trade-off. A reader should care because this yields a robust alternative to Wasserstein barycenters that keeps the convex landscape of transport-based objectives.

What carries the argument

The carrying object is the Huber ground cost $\rho_c(t)=\frac12 t^2$ for $|t|<c$ and $\rho_c(t)=c|t|-\frac{c^2}{2}$ for $|t|\ge c$, used inside the optimal transport problem $T_{\rho_c}(\mu,\nu)=\inf_\pi \int \rho_c(\|x-y\|)\,d\pi(x,y)$. The barycenter objective subtracts $c\int\|x\|\,d\mu$ so that $\Gamma_{c,P}$ is finite for every $P$ on $P_1(\mathbb{R}^d)$ without extra moment conditions. Three properties do the work: the dual potentials are $c$-Lipschitz, which yields Lipschitz stability of the cost in $W_1$; the quadratic-region part of an optimal Huber plan is cyclically monotone for the squared cost while the linear-region part reduces to an $L^1$ optimal transport ray decomposition, giving existence of a Monge map $H_c$; and in $d=1$ the quantile representation makes the barycenter a pointwise scalar Huber location problem with score $\psi_c$. The averaged-potential characterization of minimizers connects the population problem to the empirical one.

What would settle it

Take $\mu$ uniform on $[0,1]$ and $\nu$ the uniform mixture of $[0,1]$ and $[2,3]$; the quadratic transport map is a step function, hence not $\alpha$-Lipschitz. Compute the one-dimensional Huber-optimal maps $H_c$ for increasing $c$ from the quantile equation and check whether $\|H_0-H_c\|_{L^2(\mu)}$ tends to zero at the rate predicted by Theorem 2.12. If it does not, the Lipschitz assumption is essential.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central claim is that the functional $\Gamma_{c,P}$ is a well-posed robust barycenter: its minimizer set $m_c(P)$ is nonempty, bounded, convex, and closed in $W_1$; empirical plug-in barycenters converge almost surely under one- and two-stage sampling; and the finite-sample breakdown point is essentially $1/2$. In dimension one the barycenter is characterized pointwise in quantile space by $\int \psi_c(Q_\nu(u)-Q_\mu(u))\,dP(\mu)=0$, where $\psi_c(t)=\max\{-c,\min\{t,c\}\}$, so each quantile displacement is clipped separately. The paper also establishes that the Huber-optimal transport problem admits a deterministic transport map for absolutely continuous sources, that optimal Kantorovich potentials are unique up to constants and $c$-Lipschitz, that barycenter minimizers are characterized by averaged Huber potentials, and that $c\to\infty$ recovers the quadratic Wasserstein barycenter while $c\to 0$ recovers the $W_1$-type Wasserstein median after normalization. The distinction from metric-space Huber means is explicit: Huberizing inside the transport cost preserves convexity and clips individual displacements rather than whole distributions.

Load-bearing premise

The quantitative claim that the Huber transport map approaches the quadratic transport map as $c$ grows large depends on the quadratic transport map being Lipschitz with a fixed constant; the paper gives no example where this holds and no alternative bound when it fails.

Editorial extensions

If this is right

Varying $c$ traces a continuous path from the classical Wasserstein barycenter ($c\to\infty$) to the $W_1$-type Wasserstein median ($c\to 0$), so the method supplies a one-parameter family of robust centers rather than a single estimator.
In one dimension the estimator can be computed exactly by solving a scalar Huber location problem at each quantile level, with bounded influence function and asymptotic normality, making the robustness–efficiency trade-off explicit.
The finite-sample breakdown point is essentially $1/2$: replacing fewer than half of the observed distributions cannot drive the empirical barycenter arbitrarily far, while replacing slightly more than half can.
One-stage and two-stage plug-in barycenters are consistent, so the robust center remains valid when each observed distribution is itself estimated from its own sample.
When contamination changes only part of a distribution, clipping each transport displacement preserves information in unaffected regions, while random-translation contamination makes the inside- and outside-Huber constructions coincide.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

Beyond the paper, the same piecewise-quadratic/linear proof strategy should produce Monge solutions for any ground cost that is quadratic near zero and affine outside a cutoff, so the existence theory likely extends to smooth truncations of the quadratic cost.
The cellwise-versus-casewise analogy suggests a practical rule not tested here: when contamination is localized in a known region of the distributions, small $c$ should be chosen to preserve clean regions, whereas when entire observations are suspect, the outside-Huber construction may be preferable.
In $d>1$, the formal map-level comparison indicates that a measure with an anomalous region producing large displacements from the center should still contribute full weight in regions where its transport displacement is small; a direct numerical test on localized spatial contamination would show whether the one-dimensional gain persists.
Because the breakdown point reaches essentially $1/2$ while the influence function stays bounded, the estimator occupies a useful middle ground between the non-robust Wasserstein mean and the more aggressive Wasserstein median, which may make it attractive for adversarially contaminated distributional datasets.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note

read the letter

The construction is genuinely new: Huber loss placed inside the ground cost rather than outside the Wasserstein distance. The core theory is largely sound and the paper deserves a serious referee, but there is one concrete internal contradiction in Proposition 3.16 that must be resolved.

What the paper does well: it separates itself cleanly from Lee–Jung metric Huber means, ROBOT hard truncation, and Brizzi et al.’s strictly convex h-Wasserstein barycenters. Existence and consistency (Thms 3.4, 3.5), the essentially-1/2 breakdown point (Thm 3.11), and the interpolation limits as c→0 and c→∞ (Thm 3.6) are real contributions. The one-dimensional influence function and asymptotic-efficiency results are useful and clearly derived. Theorem 2.8, existence of an optimal map for the Huber cost, is a heavy piece of work; the proof sketch is plausible. The related-work discussion is honest and the citation pattern looks appropriate. No fitted quantities feed back into the theory: c is a user parameter, so circularity is not an issue.

Soft spots, in proportion. The biggest is Proposition 3.16. The stated influence function for the outside-Huber center is c√α, while the proof derives c/√α. The proof’s value is the one consistent with the derivation: Qη divided by its L2 norm has amplitude 1/√α on the upper block. The stated c√α would, for α<1, actually undercut the paper’s claim that inside Huberization is more robust under localized contamination; the proof’s c/√α supports that claim. As written, the proposition cannot be used. It is a motivating example rather than a load-bearing theorem, but it is advertised in the abstract and in Section 3.4.1, so it needs a direct re-derivation and corrected statement.

Theorem 2.12 assumes the quadratic OT map is α-Lipschitz without giving examples or a fallback bound; that limits the quantitative map-stability claim, though the qualitative convergence is fine. Minor: the numerical experiments rely on single seeds without repeated-seed error bars, and the data-driven choice of c is reasonable but not deeply analyzed. Nothing else struck me as unsound.

Bottom line: I would send this to a competent referee, not desk-reject. The core results are defensible and the construction is worth publishing. The referee should insist on the Proposition 3.16 fix and a caveat on Theorem 2.12’s condition.

Referee Report

2 major / 3 minor

Summary. The paper proposes a robust barycenter for distribution-valued data by placing the Huber loss inside the optimal transport ground cost rather than outside as in previously studied metric-space Huber means. The main object is m_c(P) = argmin_{ν∈P1(Rd)} Γ_{c,P}(ν), where Γ_{c,P}(ν) integrates the centered Huber transport cost T_{ρ_c}(ν,μ) − c∫||x||dμ(x). The authors establish existence, convexity, boundedness, and weak closedness of the barycenter set; consistency under one- and two-stage sampling; quantitative convergence to Wasserstein medians as c↓0 and to quadratic Wasserstein barycenters as c↑∞; a characterization via averaged Kantorovich potentials; a finite-sample breakdown point essentially 1/2; and, in dimension one, quantile characterizations, pointwise influence functions, asymptotic normality, and comparative efficiency with distance-based Huberization. Numerical experiments on MNIST images, London Underground flow profiles, and synthetic score distributions illustrate the interpolation between Wasserstein means and medians and the advantage of displacement-wise clipping under localized contamination. The supplementary appendix contains full proofs of the main theorems, including a decomposition-based proof of existence of Huber optimal transport maps, and a Sinkhorn-type algorithm with convergence guarantees for the regularized barycenter problem.

Significance. If the results are correct, the paper makes a useful contribution to robust distribution-valued data analysis. The construction is natural: it preserves the convex geometry of transport-based barycenters while clipping large individual displacements, and it provides an explicit interpolation between Wasserstein means and W1 Wasserstein medians. The theoretical package is substantial and mostly self-contained: existence, consistency, breakdown, and one-dimensional influence-function results are proved in detail, and the computational appendix supplies a convergent fixed-grid algorithm with reproducible code. The finite-sample breakdown-point theorem and the explicit scalar influence-function formulas are concrete, falsifiable statements. The central theoretical claims do not rely on fitted quantities and are not circular: the cutoff c is a user-specified parameter in the theory, and the data-driven c-selection in the experiments does not feed back into the theorem statements.

major comments (2)

[Section 3.4.1, Proposition 3.16] The stated influence function for the outside-Huber barycenter, IF_out = c√α 1_{(1−α,1)}, is inconsistent with the proof in Appendix A.2. The proof derives q_ε/ε → c Q_η/||Q_η||_{L2} = c/√α 1_{(1−α,1)}, since ||Q_η||_{L2} = M√α. Thus the pointwise outside influence is c/√α, not c√α. This is not a cosmetic typo: the two readings have opposite implications for α<1. The printed c√α would make the outside estimator more robust than the inside one and undercut the paper's localized-contamination claim, whereas the proof's c/√α supports the advertised conclusion. Proposition 3.16 is the quantitative support for the abstract's claim that displacement-wise Huberization retains first-order information lost by distance-based Huberization, so the discrepancy must be resolved by a direct re-derivation before that message is accepted. The core existence, consistency, breakdown, and interpolation theorems are not affected, but Proposition 3.16 cannot be cited as written.
[Section 2.4, Theorem 2.12] The quantitative map stability result assumes that the quadratic optimal transport map H0 is α-Lipschitz, equivalently that the conjugate of the quadratic potential is 1/α-strongly convex. The paper gives no example of a pair (μ,ν) satisfying this condition and no fallback bound when it fails. For rough or non-Lipschitz quadratic maps, the L2 convergence of Huber OT maps to H0 as c→∞ is therefore not established; only the cost-level convergence of Proposition 2.11 is unconditional. Since the abstract and Section 2 list 'stability as the Huber parameter varies' among the contributions, this missing support should be addressed, either by exhibiting a nontrivial class satisfying the condition or by providing a bound under weaker assumptions.

minor comments (3)

[Section 3.2, after Lemma 3.8] The sentence 'If μ≪L^d is such that μ(∂supp(μ)) = 0 and supp(μ), the directional derivative is linear...' is incomplete; it should presumably read 'and supp(μ) is connected'.
[Section 4.1] The Sinkhorn regularization is chosen as ε_c = η times the 0.90-quantile of positive entries of the Huber cost matrix, so ε depends on both c and the data. The statement that observed changes along the c-path are attributable only to the Huber loss would be more convincing with a sensitivity check over η.
[Throughout] The paper alternates between 'Huber–Wasserstein barycenter', 'Huber–Wasserstein mean', and 'Huber center' for the same object. Consistent terminology would help readers track which estimator is being discussed, especially in Tables 1 and 2.

Circularity Check

0 steps flagged · score 0.0 of 10

Derivation is self-contained; no fitted parameter is renamed as a prediction and no load-bearing step reduces to its own definition.

full rationale

The paper defines the Huber-Wasserstein barycenter m_c(P) as the exact argmin of the explicit objective Γ_{c,P}(ν) = ∫(T_{ρ_c}(ν,μ) - c∫||x||dμ)dP(μ), with c a user-specified parameter rather than a fitted constant. All central results — existence (Theorem 3.4), one- and two-stage consistency (Proposition 3.5), limits as c→0 and c→∞ (Theorem 3.6), the first-order characterization (Theorem 3.7), the finite-sample breakdown point (Theorem 3.11), and the one-dimensional quantile score characterization (Proposition 3.12) — are proved directly from this definition and standard optimal-transport facts, not by invoking a conclusion equivalent to the definition. The numerical experiments select c using validation or trimmed cross-validation data, but this selection is not used inside any theoretical proof and no empirical fit is relabeled as a theoretical prediction; the interpolation and robustness claims are proven independently of the chosen c. Self-citations appear only as technical support (e.g., a two-stage-sampling lemma from Bachoc et al., 2026, used inside the proof of Proposition 3.5), and these are not the load-bearing source of the paper's main existence, consistency, or breakdown results. Two concerns noted in the manuscript — the α-Lipschitz assumption in Theorem 2.12 with no supplied examples, and the apparent c√α versus c/√α inconsistency in Proposition 3.16 — are matters of correctness or internal consistency, not circularity: neither involves fitting a parameter to data and then predicting the same fitted quantity. The honest finding is therefore no significant circularity, with score 0.

Assumptions & free parameters 2 free parameters · 7 assumptions · 0 invented entities

The central claims rest mostly on standard optimal transport duality, convexity of the Huber loss, and clearly stated domain assumptions such as absolute continuity of source measures, finite first moments, and N(n) going to infinity. No invented entities are introduced. The only user-tuned quantities in experiments are the Huber cutoff and Sinkhorn smoothing, and neither enters the proofs as a fitted constant. The alpha-Lipschitz quadratic-map condition in Theorem 2.12 is an explicit regularity assumption worth checking in applications.

free parameters (2)

Huber cutoff c = User-specified; experiment values 0.48 to 0.03 (MNIST), 0.02 (London), 0.06 (bias-score)
Controls the transition from quadratic to linear transport cost. It is a tuning parameter selected by validation or cross-validation in experiments, not fitted to make any theorem true.
Sinkhorn regularization epsilon_c = eta=0.006 times 0.90-quantile of positive cost entries (MNIST only)
Computational smoothing used for image experiments so that the Huber path is not confounded with different effective entropic smoothing. It does not enter the theoretical claims.

assumptions (7)

standard math Kantorovich duality, existence of optimal plans, and cyclical monotonicity hold for the lower semicontinuous ground cost rho_c(||x-y||) on P1(Rd).
Invoked in Theorem 2.1 and throughout Section 2, following Villani (2009).
standard math The Huber loss rho_c is convex, continuously differentiable, c-Lipschitz, quadratic below c and linear above c.
Used in Lemma 2.2, the c-Lipschitz potential bounds, and the one-dimensional score equations.
domain assumption For the Monge-map theorem, the source measure mu is absolutely continuous with respect to Lebesgue measure; for potential uniqueness, mu(∂supp(mu))=0 and supp(mu) is connected.
Used in Theorems 2.7 and 2.8 to obtain differentiability of potentials and gradient maps.
domain assumption In Theorem 2.12, the quadratic optimal transport map H0 is assumed to be alpha-Lipschitz.
Needed for the strong convexity argument on the conjugate potential and for the stated L2 stability bound.
standard math The Ambrosio-Pratelli ray decomposition and measurable gluing theorems for L1-optimal transport apply to the truncated residual plan.
Used in Steps 3-5 of the proof of Theorem 2.8.
standard math P1(Rd) endowed with W1 is a Polish space, and standard Prokhorov, Varadarajan, and measurable maximum theorems apply to the barycenter consistency proofs.
Used in Proposition 3.5 and Theorem 3.7 for compactness, selection, and almost-sure convergence.
domain assumption In two-stage sampling, N(n) tends to infinity and each estimated measure is the empirical measure of N i.i.d. observations from mu_i.
Required for the two-stage consistency statement in Proposition 3.5(ii).

how reviews work

0 comments

Cite this review

Pith. "Pith review of Huber-Wasserstein barycenters for robust distribution-valued data." pith.science (2026). https://pith.science/paper/ZF3A3OPB

@misc{pith2026260813131,
  author       = {Pith},
  title        = {Pith review of: Huber-Wasserstein barycenters for robust distribution-valued data},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ZF3A3OPB}},
  note         = {Machine review of arXiv:2608.13131}
}

abstract

We propose a robust barycenter for distribution-valued data by incorporating the Huber loss directly into the optimal transport cost. In contrast to metric-space Huber means, which apply the Huber loss to the Wasserstein distance after optimization, our construction acts on individual transport displacements, preserving quadratic behavior locally while limiting the influence of large displacements. The resulting Huber-Wasserstein barycenters form a natural interpolation between Wasserstein means and $L^1$-type Wasserstein medians. We establish the analytical and statistical foundations of this construction. For optimal transport with Huber loss, we prove regularity and uniqueness properties of dual potentials, existence of optimal transport maps, and stability as the Huber parameter varies. For the associated barycenter problem, we prove existence and characterization results, consistency of empirical plug-in estimators, and a finite-sample breakdown point essentially equal to $1/2$. In dimension one, we further derive the pointwise influence function and asymptotic distribution, quantify the associated robustness-efficiency trade-off, and show that displacement-wise Huberization can retain first-order information that is lost by distance-based Huberization under localized shape contamination. Numerical experiments on contaminated distribution-valued data demonstrate the robustness of the proposed barycenters and illustrate their interpolation between mean- and median-like behavior.

Figures

Figures reproduced from arXiv: 2608.13131 by the authors.

**Figure 2.** MNIST digit contamination experiment. The upper panel displays the nine input [PITH_FULL_IMAGE:figures/full_fig_p021_2.png] view at source ↗

**Figure 3.** London Underground pollution experiment. Thin gray curves are clean morning [PITH_FULL_IMAGE:figures/full_fig_p022_3.png] view at source ↗

Figures from the paper (1 more)

**Figure 4.** Figure 4: Algorithmic-bias score-distribution experiment. The left panel shows the mixed [PITH_FULL_IMAGE:figures/full_fig_p025_4.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

110 extracted references · 46 canonical work pages

[1]

The Annals of Statistics , volume=

The influence function and maximum bias of Tukey's median , author=. The Annals of Statistics , volume=. 2002 , publisher=

2002
[2]

Villani , TITLE =

C. Villani , TITLE =. 2009 , PAGES =

2009
[3]

Rachev and Ludger Rüschendorf , title =

Svetlozar T. Rachev and Ludger Rüschendorf , title =. 1998 , publisher =. doi:10.1007/b98893 , isbn =

work page doi:10.1007/b98893 1998
[4]

Advances in neural information processing systems , volume=

Entropic optimal transport between unbalanced gaussian measures has a closed form , author=. Advances in neural information processing systems , volume=
[5]

LeCun, Yann and Cortes, Corinna and Burges, Christopher J. C. , title =
[6]

Tackling Algorithmic Bias in Neural-Network Classifiers Using

Risser, Laurent and Gonz. Tackling Algorithmic Bias in Neural-Network Classifiers Using. J. Math. Imaging Vis. , fjournal =. 2022 , doi =

2022
[7]

arXiv preprint arXiv:2305.09745 , year=

Weak limits for empirical entropic optimal transport: Beyond smooth costs , author=. arXiv preprint arXiv:2305.09745 , year=

arXiv
[8]

Comptes Rendus

An entropic generalization of Caffarelli’s contraction theorem via covariance inequalities , author=. Comptes Rendus. Math

Show all 110 references

[9]

Calculus of Variations and Partial Differential Equations , volume=

Convergence rate of general entropic optimal transport costs , author=. Calculus of Variations and Partial Differential Equations , volume=. 2023 , publisher=

2023
[10]

arXiv preprint arXiv:2412.12007 , year=

The entropic optimal (self-) transport problem: Limit distributions for decreasing regularization with application to score function estimation , author=. arXiv preprint arXiv:2412.12007 , year=

arXiv
[11]

2011 , publisher=

Functional analysis, Sobolev spaces and partial differential equations , author=. 2011 , publisher=

2011
[12]

ESAIM: Control, Optimisation and Calculus of Variations , volume=

Displacement smoothness of entropic optimal transport , author=. ESAIM: Control, Optimisation and Calculus of Variations , volume=. 2024 , publisher=

2024
[13]

arXiv preprint arXiv:2207.07427 , year=

Weak limits of entropy regularized optimal transport; potentials, plans and divergences , author=. arXiv preprint arXiv:2207.07427 , year=

arXiv
[14]

Real Analysis Methods for Markov Processes: Singular Integrals and Feller Semigroups , ISBN =

Taira, Kazuaki , year =. Real Analysis Methods for Markov Processes: Singular Integrals and Feller Semigroups , ISBN =. doi:10.1007/978-981-97-3659-1 , publisher =

doi
[15]

ESAIM: Control, Optimisation and Calculus of Variations , pages =

Peyre, R\'emi , title =. ESAIM: Control, Optimisation and Calculus of Variations , pages =. 2018 , doi =

2018
[16]

Manole, Tudor and Balakrishnan, Sivaraman and Niles-Weed, Jonathan and Wasserman, Larry , TITLE =. Ann. Statist. , FJOURNAL =. 2024 , NUMBER =. doi:10.1214/24-aos2379 , URL =

2024 doi
[17]

del Barrio, Eustasio and Gordaliza, Paula and Loubes, Jean-Michel , title = ". Inf. Inference J. IMA , volume =
[18]

Proceedings of Machine Learning Research , pages =

Obtaining Fairness using Optimal Transport Theory , author =. Proceedings of Machine Learning Research , pages =. 2019 , volume =

2019
[19]

, title =

Bogachev, Vladimir I. , title =. 2007 , doi =

2007
[20]

International Conference on Machine Learning , pages=

Testing group fairness via optimal transport projections , author=. International Conference on Machine Learning , pages=. 2021 , organization=

2021
[21]

Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency , pages =

Taskesen, Bahar and Blanchet, Jose and Kuhn, Daniel and Nguyen, Viet Anh , title =. Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency , pages =. 2021 , isbn =

2021
[22]

, TITLE =

Gilbarg, David and Trudinger, Neil S. , TITLE =. 1983 , PAGES =. doi:10.1007/978-3-642-61798-0 , URL =

1983 doi
[23]

, TITLE =

Evans, Lawrence C. , TITLE =. 2010 , PAGES =. doi:10.1090/gsm/019 , URL =

2010 doi
[24]

Annals of Statistics , VOLUME =

Hallin, Marc and del Barrio, Eustasio and Cuesta-Albertos, Juan and Matr\'an, Carlos , TITLE =. Annals of Statistics , VOLUME =. 2021 , NUMBER =

2021
[25]

and Galichon, A

Chernozhukov, V. and Galichon, A. and Hallin, M. and Henry, M. , TITLE =. Annals of Statistics , VOLUME =. 2017 , NUMBER =

2017
[26]

G. Peyr. Computational optimal transport: with applications to data science , year =. Foundations and Trends in Machine Learning , _issn =
[27]

1986 , publisher=

Robust statistics: the approach based on influence functions , author=. 1986 , publisher=

1986
[28]

Research Papers in Statistical Inference for Time Series and Related Models: Essays in Honor of Masanobu Taniguchi , pages=

Robustness Aspects of Optimal Transport , author=. Research Papers in Statistical Inference for Time Series and Related Models: Essays in Honor of Masanobu Taniguchi , pages=. 2023 , publisher=

2023
[29]

Research Papers in Statistical Inference for Time Series and Related Models , editor =

Ronchetti, Elvezio , title =. Research Papers in Statistical Inference for Time Series and Related Models , editor =. 2023 , publisher =

2023
[30]

On the breakdown point of transport-based quantiles

Marco Avella-Medina and Alberto González-Sanz. On the breakdown point of transport-based quantiles. arXiv preprint arxiv:2410.16554. 2024

2024
[31]

arXiv preprint arXiv:2410.19596 , year=

On the robustness of semi-discrete optimal transport , author=. arXiv preprint arXiv:2410.19596 , year=

arXiv
[32]

Proceedings of the 38th International Conference on Machine Learning , series =

Outlier-Robust Optimal Transport , author =. Proceedings of the 38th International Conference on Machine Learning , series =
[33]

Proceedings of the 24th International Conference on Artificial Intelligence and Statistics , series =

When OT Meets MoM: Robust Estimation of Wasserstein Distance , author =. Proceedings of the 24th International Conference on Artificial Intelligence and Statistics , series =
[34]

Advances in Neural Information Processing Systems , volume =

Robust Optimal Transport with Applications in Generative Modeling and Domain Adaptation , author =. Advances in Neural Information Processing Systems , volume =
[35]

Proceedings of the 25th International Conference on Artificial Intelligence and Statistics , series =

Outlier-Robust Optimal Transport: Duality, Structure, and Statistical Analysis , author =. Proceedings of the 25th International Conference on Artificial Intelligence and Statistics , series =
[36]

Advances in Neural Information Processing Systems , volume =

On Robust Optimal Transport: Computational Complexity and Barycenter Computation , author =. Advances in Neural Information Processing Systems , volume =
[37]

Proceedings of the 2024 SIAM International Conference on Data Mining , pages =

On Robust Wasserstein Barycenter: The Model and Algorithm , author =. Proceedings of the 2024 SIAM International Conference on Data Mining , pages =. 2024 , doi =

2024
[38]

arXiv preprint arXiv:2603.07563 , year =

Robust Wasserstein Barycenter , author =. arXiv preprint arXiv:2603.07563 , year =

arXiv
[39]

arXiv preprint arXiv:2603.16005 , year =

Breakdown Properties of Optimal Transport Maps: General Transportation Costs , author =. arXiv preprint arXiv:2603.16005 , year =
[40]

arXiv preprint arXiv:2607.19080 , year =

The Influence Function of Transport-based Quantiles , author =. arXiv preprint arXiv:2607.19080 , year =

arXiv
[41]

and Grabarnik, Genady Ya

Rubshtein, Ben-Zion A. and Grabarnik, Genady Ya. and Muratov, Mustafa A. and Pashkova, Yulia S. , TITLE =. 2016 , PAGES =. doi:10.1007/978-3-319-42758-4 , URL =

2016 doi
[42]

and Fournier, John J

Adams, Robert A. and Fournier, John J. F. , TITLE =. 2003 , PAGES =

2003
[43]

Nardi, Giacomo , TITLE =. Enseign. Math. , FJOURNAL =. 2014 , NUMBER =. doi:10.4171/LEM/60-3/4-9 , URL =

2014 doi
[44]

2017 , PAGES =

Figalli, Alessio , TITLE =. 2017 , PAGES =. doi:10.4171/170 , URL =

2017 doi
[45]

Linearization of

Gonz. Linearization of. arXiv preprint arXiv:2408.06534 , year=
[46]

, TITLE =

Loeper, G. , TITLE =. Acta Math. , FJOURNAL =. 2009 , NUMBER =. doi:10.1007/s11511-009-0037-8 , URL =

2009 doi
[47]

del Barrio, Eustasio and Loubes, Jean-Michel , TITLE =. Ann. Probab. , FJOURNAL =. 2019 , NUMBER =

2019
[48]

Annales de l'Institut Henri Poincar

del Barrio, Eustasio and Gonz\'alez-Sanz, Alberto and Loubes, Jean-Michel , TITLE =. Annales de l'Institut Henri Poincar. 2024 , NUMBER =

2024
[49]

2022 , journal=

Graphical and uniform consistency of estimated optimal transport plans , author=. 2022 , journal=

2022
[50]

Second order stability for the Monge–Ampère equation and strong Sobolev convergence of optimal transport maps , volume =

De Philippis, Guido and Figalli, Alessio , year =. Second order stability for the Monge–Ampère equation and strong Sobolev convergence of optimal transport maps , volume =. Analysis & amp; PDE , publisher =. doi:10.2140/apde.2013.6.993 , number =

doi
[51]

, TITLE =

Caffarelli, Luis A. , TITLE =. Annals of Mathematics , YEAR =
[52]

, TITLE =

Caffarelli, Luis A. , TITLE =. Journal of the American Mathematical Society , VOLUME =. 1992 , NUMBER =

1992
[53]

, year =

Gutiérrez, Cristian E. , year =. The Monge-Ampère Equation , ISBN =. doi:10.1007/978-3-319-43374-5 , journal =

doi
[54]

and Cabr\'e, Xavier , TITLE =

Caffarelli, Luis A. and Cabr\'e, Xavier , TITLE =. 1995 , PAGES =. doi:10.1090/coll/043 , URL =

1995 doi
[55]

A characterization of random variables with minimum

R. A characterization of random variables with minimum. J. Multivariate Anal. , VOLUME =. 1990 , PAGES =

1990
[56]

Polar factorization and monotone rearrangement of vector-valued functions , author=. Comm. Pure Appl. Math. , volume=. 1991 , publisher=

1991
[57]

Notes on the

Juan A. Notes on the. Ann. Probab. , year=
[58]

1978 , PAGES =

Dellacherie, Claude and Meyer, Paul-Andr\'e , TITLE =. 1978 , PAGES =

1978
[59]

2003 , PAGES =

Villani, C\'edric , TITLE =. 2003 , PAGES =. doi:10.1090/gsm/058 , URL =

2003 doi
[60]

2023 , eprint=

Banach-Saks Theorem for L^1 revisited , author=. 2023 , eprint=

2023
[61]

Chen, Yaqing and Lin, Zhenhua and M\"uller, Hans-Georg , TITLE =. J. Amer. Statist. Assoc. , FJOURNAL =. 2023 , NUMBER =. doi:10.1080/01621459.2021.1956937 , URL =

2023
[62]

Petersen, Alexander and M\"uller, Hans-Georg , TITLE =. Ann. Statist. , FJOURNAL =. 2016 , NUMBER =. doi:10.1214/15-AOS1363 , URL =

2016 doi
[63]

, TITLE =

Petersen, Alexander and Liu, Xi and Divani, Afshin A. , TITLE =. Ann. Statist. , FJOURNAL =. 2021 , NUMBER =. doi:10.1214/20-AOS1971 , URL =

2021 doi
[64]

van den Boogaart, Karl Gerald and Egozcue, Juan Jos\'e. Bayes. Aust. N. Z. J. Stat. , FJOURNAL =. 2014 , NUMBER =. doi:10.1111/anzs.12074 , URL =

2014 doi
[65]

Journ\'ees

Bigot, J\'er\'emie , TITLE =. Journ\'ees. 2020 , MRCLASS =. doi:10.1051/proc/202068001 , URL =

2020
[66]

2025 , journal =

Improved learning theory for kernel distribution regression with two-stage sampling , author=. 2025 , journal =

2025
[67]

International Conference on Machine Learning , pages=

A distributional framework for data valuation , author=. International Conference on Machine Learning , pages=. 2020 , volume=

2020
[68]

Advances in Neural Information Processing Systems , volume=

Data distributional properties drive emergent in-context learning in transformers , author=. Advances in Neural Information Processing Systems , volume=
[69]

Agueh, Martial and Carlier, Guillaume , TITLE =. SIAM J. Math. Anal. , FJOURNAL =. 2011 , NUMBER =. doi:10.1137/100805741 , URL =

2011 doi
[70]

Le Gouic, Thibaut and Loubes, Jean-Michel , TITLE =. Probab. Theory Related Fields , FJOURNAL =. 2017 , NUMBER =. doi:10.1007/s00440-016-0727-z , URL =

2017 doi
[71]

A Fixed-Point Approach to Barycenters in

Pedro C. A Fixed-Point Approach to Barycenters in. J. Math. Anal. Appl. , fjournal =. 2016 , doi =

2016
[72]

Proceedings of the 31st International Conference on Machine Learning , series =

Marco Cuturi and Arnaud Doucet , title =. Proceedings of the 31st International Conference on Machine Learning , series =. 2014 , publisher =

2014
[73]

Iterative

Jean-David Benamou and Guillaume Carlier and Marco Cuturi and Luca Nenna and Gabriel Peyr. Iterative. SIAM J. Sci. Comput. , fjournal =. 2015 , doi =

2015
[74]

Characterization of Barycenters in the

J. Characterization of Barycenters in the. ESAIM Probab. Stat. , fjournal =. 2018 , doi =

2018
[75]

On the Complexity of Approximating

Alexey Kroshnin and Nazarii Tupitsa and Darina Dvinskikh and Pavel Dvurechensky and Alexander Gasnikov and C. On the Complexity of Approximating. Proceedings of the 36th International Conference on Machine Learning , series =. 2019 , publisher =

2019
[76]

Altschuler and Enric Boix-Adsera , title =

Jason M. Altschuler and Enric Boix-Adsera , title =. J. Mach. Learn. Res. , fjournal =
[77]

Florian Heinemann and Axel Munk and Yoav Zemel , title =. SIAM J. Math. Data Sci. , fjournal =. 2022 , doi =

2022
[78]

Computational Optimal Transport , journal =

Gabriel Peyr. Computational Optimal Transport , journal =. 2019 , doi =

2019
[79]

Panaretos and Yoav Zemel , title =

Victor M. Panaretos and Yoav Zemel , title =. Annu. Rev. Stat. Appl. , fjournal =. 2019 , doi =

2019
[80]

and Ronchetti, Elvezio M

Huber, Peter J. and Ronchetti, Elvezio M. , TITLE =. 2009 , PAGES =. doi:10.1002/9780470434697 , URL =

2009 doi
[81]

Guillaume Carlier and Enis Chenchene and Katharina Eichinger , title =. SIAM J. Math. Anal. , fjournal =. 2024 , doi =

2024
[82]

Staudt, Thomas and Hundrieser, Shayan and Munk, Axel , TITLE =. SIAM J. Math. Anal. , FJOURNAL =. 2025 , NUMBER =. doi:10.1137/24M1658966 , URL =

2025 doi
[83]

, TITLE =

Gangbo, Wilfrid and McCann, Robert J. , TITLE =. Acta Math. , FJOURNAL =. 1996 , NUMBER =. doi:10.1007/BF02392620 , URL =

1996 doi
[84]

Caffarelli and Mikhail Feldman and Robert J

Luis A. Caffarelli and Mikhail Feldman and Robert J. McCann , title =. J. Amer. Math. Soc. , fjournal =. 2002 , doi =

2002
[85]

Evans and Wilfrid Gangbo , title =

Lawrence C. Evans and Wilfrid Gangbo , title =. 1999 , pages =

1999
[86]

Optimal Transportation and Applications , series =

Luigi Ambrosio and Aldo Pratelli , title =. Optimal Transportation and Applications , series =. 2003 , doi =

2003
[87]

Duke Math

Thierry Champion and Luigi De Pascale , title =. Duke Math. J. , fjournal =. 2011 , doi =

2011
[88]

Aliprantis and Kim C

Charalambos D. Aliprantis and Kim C. Border , title =
[89]

Lee, Jongmin and Jung, Sungkyu , TITLE =. J. R. Stat. Soc. Ser. B. Stat. Methodol. , FJOURNAL =. 2026 , NUMBER =. doi:10.1093/jrsssb/qkaf054 , URL =

2026 doi
[90]

V. S. Varadarajan , journal =. On the Convergence of Sample Probability Distributions , urldate =
[91]

2026 , eprint=

Wasserstein Spatial Depth , author=. 2026 , eprint=

2026
[92]

Nonlinear Anal

Brizzi, Camilla and Friesecke, Gero and Ried, Tobias , TITLE =. Nonlinear Anal. , FJOURNAL =. 2025 , PAGES =. doi:10.1016/j.na.2024.113687 , URL =

2025
[93]

2013 , booktitle =

Cuturi, Marco , issn =. 2013 , booktitle =

2013
[94]

Borgwardt and Malte J

Arthur Gretton and Karsten M. Borgwardt and Malte J. Rasch and Bernhard Sch. A Kernel Two-Sample Test , journal =. 2012 , volume =

2012
[95]

The Journal of Machine Learning Research , volume=

Learning theory for distribution regression , author=. The Journal of Machine Learning Research , volume=. 2016 , publisher=

2016
[96]

Trimmed Comparison of Distributions , journal =

Pedro C. Trimmed Comparison of Distributions , journal =. 2008 , volume =

2008
[97]

Wide Consensus Aggregation in the

Pedro C. Wide Consensus Aggregation in the. Bernoulli , year =
[98]

Cuesta-Albertos and Carlos Matr

Eustasio del Barrio and Juan A. Cuesta-Albertos and Carlos Matr. Robust Clustering Tools Based on Optimal Transportation , journal =. 2019 , volume =

2019
[99]

Kisung You and Dennis Shung and Mauro Giuffr. On the. Journal of Computational and Graphical Statistics , year =
[100]

International Statistical Review , year =

Yiming Ma and Hang Liu and Davide La Vecchia and Matthieu Lerasle , title =. International Statistical Review , year =
[101]

Journal of Mathematical Analysis and Applications , year =

Camilla Brizzi and Gero Friesecke and Tobias Ried , title =. Journal of Mathematical Analysis and Applications , year =
[102]

2009 IEEE 12th International Conference on Computer Vision , year =

Ofir Pele and Michael Werman , title =. 2009 IEEE 12th International Conference on Computer Vision , year =

2009
[103]

, title =

Huber, Peter J. , title =. The Annals of Mathematical Statistics , year =
[104]

, title =

Hampel, Frank R. , title =. Journal of the American Statistical Association , year =
[105]

and Ronchetti, Elvezio M

Hampel, Frank R. and Ronchetti, Elvezio M. and Rousseeuw, Peter J. and Stahel, Werner A. , title =. 1986 , isbn =

1986
[106]

Bernoulli , year =

Sinova, Beatriz and Gonz. Bernoulli , year =
[107]

and Welsch, Roy E

Holland, Paul W. and Welsch, Roy E. , title =. Communications in Statistics---Theory and Methods , year =
[108]

, title =

van der Vaart, Aad W. , title =. 1998 , series =

1998
[109]

and Zamar, Ruben H

Alqallaf, Fatemah and Van Aelst, Stefan and Yohai, Victor J. and Zamar, Ruben H. , title =. The Annals of Statistics , volume =. 2009 , doi =

2009
[110]

and Van den Bossche, Wannes , title =

Rousseeuw, Peter J. and Van den Bossche, Wannes , title =. Technometrics , volume =. 2018 , doi =

2018

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.

[1] [1]

The Annals of Statistics , volume=

The influence function and maximum bias of Tukey's median , author=. The Annals of Statistics , volume=. 2002 , publisher=

2002

[2] [2]

Villani , TITLE =

C. Villani , TITLE =. 2009 , PAGES =

2009

[3] [3]

Rachev and Ludger Rüschendorf , title =

Svetlozar T. Rachev and Ludger Rüschendorf , title =. 1998 , publisher =. doi:10.1007/b98893 , isbn =

work page doi:10.1007/b98893 1998

[4] [4]

Advances in neural information processing systems , volume=

Entropic optimal transport between unbalanced gaussian measures has a closed form , author=. Advances in neural information processing systems , volume=

[5] [5]

LeCun, Yann and Cortes, Corinna and Burges, Christopher J. C. , title =

[6] [6]

Tackling Algorithmic Bias in Neural-Network Classifiers Using

Risser, Laurent and Gonz. Tackling Algorithmic Bias in Neural-Network Classifiers Using. J. Math. Imaging Vis. , fjournal =. 2022 , doi =

2022

[7] [7]

arXiv preprint arXiv:2305.09745 , year=

Weak limits for empirical entropic optimal transport: Beyond smooth costs , author=. arXiv preprint arXiv:2305.09745 , year=

arXiv

[8] [8]

Comptes Rendus

An entropic generalization of Caffarelli’s contraction theorem via covariance inequalities , author=. Comptes Rendus. Math