arxiv: 2605.11270 · v1 · submitted 2026-05-11 · 🧮 math.OC

Recognition: 2 theorem links

· Lean Theorem

A Unified Approach for Computing Wasserstein Barycenters of Discrete and Continuous Measures

Changbo Zhu, Peng Xu, Xiaohui Chen

Pith reviewed 2026-05-13 01:23 UTC · model grok-4.3

classification 🧮 math.OC

keywords Wasserstein barycentermirror descentoptimal transportFisher-Rao geometrysemi-discrete optimal transportdiscrete measurescontinuous measures

0 comments

The pith

A primal mirror descent algorithm computes exact Wasserstein barycenters for both discrete and continuous measures in a unified framework.

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

This paper introduces a primal mirror descent algorithm in the Fisher-Rao geometry that computes the unregularized Wasserstein barycenter for probability measures. The method applies the same sequence of updates whether the inputs are discrete point clouds or absolutely continuous densities, and it carries convergence guarantees in both cases. When every input measure is discrete, the procedure begins from an arbitrary probability density, solves a series of semi-discrete optimal transport subproblems, and produces absolutely continuous iterates that converge to the exact discrete barycenter. This matters because earlier algorithms were specialized to one regime or the other and often required regularization or grid approximations that could alter the solution.

Core claim

The central claim is that a single primal mirror descent scheme operating in the Fisher-Rao geometry on the space of probability measures computes the exact Wasserstein barycenter for any collection of discrete or absolutely continuous input measures. When all inputs are discrete, the algorithm starts from any initial density and generates a sequence of absolutely continuous functions by repeatedly solving semi-discrete optimal transport problems; these iterates converge to the target discrete barycenter. The identical update rule applies directly when inputs are absolutely continuous, again with proven convergence.

What carries the argument

Primal mirror descent updates performed in the Fisher-Rao geometry on probability measures, each step reducing to a semi-discrete optimal transport subproblem.

If this is right

The identical algorithm handles mixed discrete and continuous input measures without any change in formulation.
When inputs are discrete the iterates remain absolutely continuous throughout and converge to the exact discrete barycenter.
No regularization is required; the method targets the unregularized problem in both regimes.
Numerical tests on synthetic and real data indicate competitive accuracy and running time compared with specialized solvers.

Where Pith is reading between the lines

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

The ability to begin from any density and still converge suggests the procedure may be robust to poor initializations in practice.
Because the iterates stay absolutely continuous, the method could supply smooth interpolants between a point cloud and a density without additional post-processing.
The unified treatment may simplify pipelines that combine sensor data arriving as point sets with image data arriving as densities.

Load-bearing premise

The input measures are discrete or absolutely continuous and the mirror descent steps remain well-defined and convergent inside the Fisher-Rao geometry.

What would settle it

Run the algorithm from a uniform density on a simple discrete instance whose barycenter is known exactly, such as two uniform distributions supported on two distinct points, and check whether the final iterate recovers that known barycenter to machine precision.

Figures

Figures reproduced from arXiv: 2605.11270 by Changbo Zhu, Peng Xu, Xiaohui Chen.

**Figure 2.** Figure 2: Left to right: ground truth (LP), mirror descent barycenter (FRBary), entropic barycenter (EWB), and debiased entropic barycenter (DSB), where λ = 0.0004 for the latter two. 5.3 2D/3D Gaussian point clouds We evaluate our method on point clouds generated from Gaussian distributions. Since the Wasserstein barycenter of Gaussian measures remains Gaussian and admits a closed-form solution, this setting allows… view at source ↗

**Figure 3.** Figure 3: 2D (left) and 3D (right) samples drawn from the true ( [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗

**Figure 4.** Figure 4: Distances between the true and estimated means and covariances. [PITH_FULL_IMAGE:figures/full_fig_p010_4.png] view at source ↗

**Figure 5.** Figure 5: Results of color averaging with stochastic palette barycenter. [PITH_FULL_IMAGE:figures/full_fig_p010_5.png] view at source ↗

**Figure 6.** Figure 6: Comparison with Bures–Wasserstein gradient descent. Performance is measured by the [PITH_FULL_IMAGE:figures/full_fig_p019_6.png] view at source ↗

**Figure 7.** Figure 7: Swiss roll. Comaprison with free-support approaches. Using the Swiss roll data, we further compared our method (with both discretization and neural network-based method) against several representative free-support barycenter approaches: the neural network–based CW2B [Korotin et al., 2021], as well as exact and entropic solvers from the POT package. We computed the barycenter of 4 Swiss roll distributions, … view at source ↗

**Figure 8.** Figure 8: Comparison with various free-support barycenter methods. [PITH_FULL_IMAGE:figures/full_fig_p023_8.png] view at source ↗

**Figure 9.** Figure 9: Barycenters computed from n = 500 images. D.3 Comparison on high-resolution digit data In this experiment, we compute the barycenter of 10 high-resolution (100 × 100 and 200 × 200) handwritten digit “3” ( [PITH_FULL_IMAGE:figures/full_fig_p024_9.png] view at source ↗

**Figure 10.** Figure 10: 10 handwritten digit “3” from the HWD+ dataset. [PITH_FULL_IMAGE:figures/full_fig_p025_10.png] view at source ↗

**Figure 11.** Figure 11: Left to right: Barycenters computed through FRBary, sGS-ADMM, debiased Sinkhorn (λ = 0.005), and WDHA. Solver 100 × 100 Images 200 × 200 Images Iterations Time (seconds) Iterations Time (seconds) FRBary 200 147 200 423 sGA-ADMM 1650 473 2950 16470 Debiased Sinkhorn 100 0.84 100 6.73 WDHA 1000 5.5 1000 67 [PITH_FULL_IMAGE:figures/full_fig_p025_11.png] view at source ↗

**Figure 12.** Figure 12: Results of color averaging with discretization-based barycenter. [PITH_FULL_IMAGE:figures/full_fig_p026_12.png] view at source ↗

**Figure 13.** Figure 13: Color palettes of the input images (left and middle) and the barycenter (right). [PITH_FULL_IMAGE:figures/full_fig_p026_13.png] view at source ↗

**Figure 14.** Figure 14: Optimality gap with variable step sizes. [PITH_FULL_IMAGE:figures/full_fig_p027_14.png] view at source ↗

read the original abstract

Computing the unregularized Wasserstein barycenter for measure-valued data is a challenging optimization task. Recent algorithms have been tailored to either discrete measures as point clouds or continuous measures discretized on regular grids. In this work, we propose a primal mirror descent algorithm for computing the exact Wasserstein barycenter in the Fisher-Rao geometry. Our algorithm is a unified approach that is flexible enough to simultaneously cover discrete and absolutely continuous input measures, with convergence guarantees in both settings. In particular, when all input measures are discrete, our algorithm, initialized from any probability density, solves a sequence of semi-discrete optimal transport subproblems and produces absolutely continuous iterates that converge to the discrete barycenter. We use synthetic and real data examples to demonstrate the promising result in terms of accuracy and computational cost.

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.

Desk Editor's Note private letter to a colleague

The paper gives a mirror-descent scheme that unifies exact barycenter computation for discrete and continuous measures, but the claimed convergence from continuous iterates to a discrete barycenter rests on an incompletely justified limit step.

read the letter

The paper's core idea is a primal mirror descent algorithm in Fisher-Rao geometry that computes unregularized Wasserstein barycenters. It treats discrete and absolutely continuous inputs in one framework by repeatedly solving semi-discrete optimal transport subproblems. The distinctive claim is that, for fully discrete inputs, the method can start from an arbitrary density, keep all iterates absolutely continuous, and still converge to the exact discrete barycenter. That unification and the specific initialization behavior are the genuinely new pieces; prior work handled the two cases separately or used regularization or discretization from the start. The numerical section shows the approach on synthetic examples and real data, reporting competitive accuracy and run times, which is useful evidence that the scheme is at least practical. The paper therefore earns credit for delivering a single algorithm plus working code-level demonstrations rather than just another discretization trick. The soft spot is the convergence argument for the discrete limit. The abstract states that the iterates converge, but the stress-test concern is on target: as the densities concentrate onto atoms, one needs to confirm that the Fisher-Rao updates stay well-defined, that the semi-discrete subproblems remain compatible with the geometry, and that the limit is taken in the right topology without hidden smoothing or positivity assumptions. If the full proof only invokes standard mirror-descent theory without addressing the vanishing-density regime explicitly, the central claim is not yet fully supported. Readers who already work on optimal-transport algorithms for mixed data types will get the most out of this; the paper is not a broad theoretical advance but a concrete algorithmic tool. It is worth sending to peer review so that referees can check the limit justification and the precise statement of the convergence theorem.

Referee Report

2 major / 2 minor

Summary. The manuscript proposes a primal mirror descent algorithm in the Fisher-Rao geometry to compute exact (unregularized) Wasserstein barycenters. The method is presented as unified: it handles both discrete point-cloud measures and absolutely continuous measures, with claimed convergence guarantees in each case. For the discrete-input setting the algorithm starts from an arbitrary probability density, solves a sequence of semi-discrete optimal-transport subproblems at each iteration, and produces a sequence of absolutely continuous iterates asserted to converge to the exact discrete barycenter. Synthetic and real-data experiments are used to illustrate accuracy and computational cost.

Significance. A single algorithmic framework that provably recovers exact discrete barycenters from continuous iterates while also handling continuous inputs would be a useful contribution to the optimal-transport literature. The Fisher-Rao mirror-descent formulation and the explicit semi-discrete OT subproblem structure are technically interesting; if the convergence statements are rigorously established they would constitute a clear advance over existing specialized discrete or grid-based methods.

major comments (2)

[Abstract, §3–4] The central convergence claim (abstract and §3–4) states that absolutely continuous iterates generated by Fisher-Rao mirror descent converge to the discrete barycenter when all input measures are discrete. The manuscript must supply an explicit argument for the limit: the topology in which convergence holds, the behavior of the mirror-descent updates as densities concentrate on atoms (where the target has zero density), and the compatibility of the semi-discrete OT solutions with the geometry. Without this justification the claim that the algorithm “produces absolutely continuous iterates that converge to the discrete barycenter” remains unverified and is load-bearing for the unified approach.
[§4] Theorem statements on convergence (presumably in §4) should clarify whether any auxiliary regularity (strict positivity, smoothing, or bounded support) is assumed that would be violated in the discrete limit; if such assumptions are used they must be shown to be removable or the result must be stated with the corresponding restrictions.

minor comments (2)

[§2] Notation for the Fisher-Rao metric and the mirror-descent step should be introduced once and used consistently; currently the abstract and early sections mix “Fisher-Rao geometry” with “primal mirror descent” without a clear reference to the underlying Riemannian structure.
[§5] The numerical section would benefit from a direct comparison table (runtime and Wasserstein error) against at least one established discrete barycenter solver (e.g., the iterative Bregman projection method) and one grid-based continuous solver on the same synthetic instances.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the detailed and constructive report. The comments highlight important points regarding the rigor of our convergence analysis for the discrete-input case. We agree that additional explicit arguments are needed to fully substantiate the claims and will revise the manuscript accordingly to strengthen the presentation of the unified approach.

read point-by-point responses

Referee: [Abstract, §3–4] The central convergence claim (abstract and §3–4) states that absolutely continuous iterates generated by Fisher-Rao mirror descent converge to the discrete barycenter when all input measures are discrete. The manuscript must supply an explicit argument for the limit: the topology in which convergence holds, the behavior of the mirror-descent updates as densities concentrate on atoms (where the target has zero density), and the compatibility of the semi-discrete OT solutions with the geometry. Without this justification the claim that the algorithm “produces absolutely continuous iterates that converge to the discrete barycenter” remains unverified and is load-bearing for the unified approach.

Authors: We agree that the current exposition would benefit from a more self-contained argument. In the revised version we will insert a new subsection (after the statement of the main convergence result in §4) that explicitly addresses the three points raised: (i) convergence holds in the narrow (weak-*) topology on the space of probability measures, which is the natural topology for Wasserstein barycenters; (ii) the Fisher-Rao mirror-descent update, when the objective value approaches the discrete barycenter value, forces the density to concentrate at the atoms because the gradient of the semi-discrete OT functional becomes unbounded away from the support; (iii) compatibility follows from the fact that the dual potentials of the semi-discrete OT problems remain bounded and the transport plans converge weakly to the optimal coupling between the limiting discrete measure and the input measures. We will include a short lemma establishing the concentration behavior and a remark on the passage to the limit inside the mirror-descent recursion. revision: yes
Referee: [§4] Theorem statements on convergence (presumably in §4) should clarify whether any auxiliary regularity (strict positivity, smoothing, or bounded support) is assumed that would be violated in the discrete limit; if such assumptions are used they must be shown to be removable or the result must be stated with the corresponding restrictions.

Authors: We thank the referee for this clarification request. The theorems in §4 are currently stated under the standing assumption that the iterates remain positive and have bounded support (inherited from the continuous-input analysis). In the revision we will (a) explicitly list these assumptions at the beginning of each theorem, (b) add a separate corollary for the purely discrete-input setting that removes the strict-positivity requirement by working with the closure of the positive cone in the narrow topology, and (c) include a short argument showing that the smoothing or bounded-support hypotheses used for the continuous case are not needed once the target is known to be discrete, because the mirror-descent step automatically drives mass to the atoms. The revised statement will therefore distinguish the two regimes clearly. revision: yes

Circularity Check

0 steps flagged

No circularity: algorithm and convergence claims are independent of inputs

full rationale

The paper proposes a primal mirror descent algorithm in Fisher-Rao geometry that unifies discrete and continuous cases by iteratively solving semi-discrete OT subproblems, with convergence of AC iterates to the discrete barycenter asserted as a theorem. No equations, definitions, or steps in the abstract or described chain reduce the claimed convergence or barycenter computation to a tautology, fitted parameter renamed as prediction, or self-citation load-bearing premise. The initialization from arbitrary density and handling of both measure types are presented as external algorithmic properties rather than self-referential. The derivation remains self-contained against the stated assumptions on input measures and geometry.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review; no explicit free parameters, axioms, or invented entities can be extracted. The algorithm implicitly relies on the existence of solutions to semi-discrete OT problems and on the geometry of the Fisher-Rao metric.

pith-pipeline@v0.9.0 · 5430 in / 995 out tokens · 27231 ms · 2026-05-13T01:23:46.728056+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

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

IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear
Our algorithm... produces absolutely continuous iterates that converge to the discrete barycenter... ρ_{k+1} ∝ ρ_k exp(−η_k ∑ w_i ϕ_{ρ_k→μ_i})
IndisputableMonolith/Foundation/AlexanderDuality.lean alexander_duality_circle_linking unclear
Theorem 4.2... min E(λ_k)−E(λ⋆)=O(log T/√T) via Gaussian smoothing

Reference graph

Works this paper leans on

112 extracted references · 112 canonical work pages

[1]

Journal of Machine Learning Research , volume=

A fast globally linearly convergent algorithm for the computation of Wasserstein barycenters , author=. Journal of Machine Learning Research , volume=

work page
[2]

ESAIM: Probability and Statistics , volume=

Characterization of barycenters in the Wasserstein space by averaging optimal transport maps , author=. ESAIM: Probability and Statistics , volume=

work page
[3]

Journal of Computational Physics , pages=

Large-scale semi-discrete optimal transport with distributed Voronoi diagrams , author=. Journal of Computational Physics , pages=. 2025 , publisher=

work page 2025
[4]

Numerical Algorithms , volume=

Solving semi-discrete optimal transport problems: star shapedeness and Newton’s method , author=. Numerical Algorithms , volume=. 2025 , publisher=

work page 2025
[5]

Annales de l'Institut Henri Poincaré, Probabilités et Statistiques , number =

Jérémie Bigot and Raúl Gouet and Thierry Klein and Alfredo López , title =. Annales de l'Institut Henri Poincaré, Probabilités et Statistiques , number =. 2017 , doi =

work page 2017
[6]

International Conference on Learning Representations (ICLR) , URL =

Sobolev Gradient Ascent for Optimal Transport: Barycenter Optimization and Convergence Analysis , author=. International Conference on Learning Representations (ICLR) , URL =

work page
[7]

Scalable Computations of

Fan, Jiaojiao and Taghvaei, Amirhossein and Chen, Yongxin , booktitle =. Scalable Computations of. 2025 , series =

work page 2025
[8]

Proceedings of the 41st International Conference on Machine Learning , pages =

Estimating Barycenters of Distributions with Neural Optimal Transport , author =. Proceedings of the 41st International Conference on Machine Learning , pages =. 2024 , volume =

work page 2024
[9]

Journal of Mathematical Analysis and Applications , volume=

A fixed-point approach to barycenters in. Journal of Mathematical Analysis and Applications , volume=. 2016 , publisher=

work page 2016
[10]

Interior-point methods strike back: Solving the

Ge, Dongdong and Wang, Haoyue and Xiong, Zikai and Ye, Yinyu , journal=. Interior-point methods strike back: Solving the

work page
[11]

Learning Gaussian mixtures using the

Yan, Yuling and Wang, Kaizheng and Rigollet, Philippe , journal=. Learning Gaussian mixtures using the. 2024 , publisher=

work page 2024
[12]

Bulletin of the London Mathematical Society , volume=

Uniqueness of the Fisher--Rao metric on the space of smooth densities , author=. Bulletin of the London Mathematical Society , volume=. 2016 , publisher=

work page 2016
[13]

Convergence analysis of the

Zhu, Shuailong and Chen, Xiaohui , year=. Convergence analysis of the

work page
[14]

On the Global Convergence of

Boufadene, Siwan and Vialard, Fran. On the Global Convergence of. SIAM Journal on Mathematical Analysis , volume =. 2025 , doi =. https://doi.org/10.1137/24M1647631 , abstract =

work page doi:10.1137/24m1647631 2025
[15]

Mirror descent and nonlinear projected subgradient methods for convex optimization , volume =

Amir Beck and Marc Teboulle , keywords =. Mirror descent and nonlinear projected subgradient methods for convex optimization , journal =. 2003 , issn =. doi:https://doi.org/10.1016/S0167-6377(02)00231-6 , url =

work page doi:10.1016/s0167-6377(02)00231-6 2003
[16]

Mathematical Methods of Operations Research , volume=

Discrete Wasserstein barycenters: Optimal transport for discrete data , author=. Mathematical Methods of Operations Research , volume=. 2016 , publisher=

work page 2016
[17]

Probability Theory and Related Field , volume =

Le Gouic, Thibaut and Loubes, Jean-Michel , title =. Probability Theory and Related Field , volume =. 2017 , doi =

work page 2017
[18]

Advances in Neural Information Processing Systems , volume=

A combinatorial algorithm for the semi-discrete optimal transport problem , author=. Advances in Neural Information Processing Systems , volume=

work page
[19]

Mathematical Programming , volume=

Semi-discrete optimal transport: Hardness, regularization and numerical solution , author=. Mathematical Programming , volume=. 2023 , publisher=

work page 2023
[20]

The Annals of Applied Probability , volume=

Stability and statistical inference for semidiscrete optimal transport maps , author=. The Annals of Applied Probability , volume=. 2024 , publisher=

work page 2024
[21]

Polar factorization and monotone rearrangement of vector-valued functions , url =

Brenier, Yann , date-added =. Polar factorization and monotone rearrangement of vector-valued functions , url =. Communications on Pure and Applied Mathematics , number =. 1991 , bdsk-url-1 =. doi:https://doi.org/10.1002/cpa.3160440402 , eprint =

work page doi:10.1002/cpa.3160440402 1991
[22]

Monge, Gaspard , keywords =. M

work page
[23]

Optimal Transport: Old and New , year =

Villani, C\'edric , date-added =. Optimal Transport: Old and New , year =

work page
[24]

Topics in Optimal Transportation , url =

Villani, C\'edric , date-added =. Topics in Optimal Transportation , url =. 2003 , bdsk-url-1 =

work page 2003
[25]

Electronic Journal of Statistics , volume=

Two-sample and change-point inference for non-Euclidean valued time series , author=. Electronic Journal of Statistics , volume=. 2024 , publisher=

work page 2024
[26]

Journal of the American Statistical Association , volume=

Wasserstein regression , author=. Journal of the American Statistical Association , volume=. 2023 , publisher=

work page 2023
[27]

and Gigli, N

Ambrosio, L. and Gigli, N. , TITLE =. Modelling and optimisation of flows on networks , SERIES =. 2013 , MRCLASS =

work page 2013
[28]

2013 , publisher=

Introductory lectures on convex optimization: A basic course , author=. 2013 , publisher=

work page 2013
[29]

, isbn =

Lan, G. , isbn =. First-order and Stochastic Optimization Methods for Machine Learning , url =

work page
[30]

Gangbo, W. and \'. Optimal maps for the multidimensional. Comm. Pure Appl. Math. , FJOURNAL =. 1998 , NUMBER =

work page 1998
[31]

doi:10.1051/m2an/2015033

work page doi:10.1051/m2an/2015033
[32]

Proceedings of the 34th International Conference on Machine Learning - Volume 70 , pages =

Ho, Nhat and Nguyen, Xuan Long and Yurochkin, Mikhail and Bui, Hung Hai and Huynh, Viet and Phung, Dinh , title =. Proceedings of the 34th International Conference on Machine Learning - Volume 70 , pages =. 2017 , publisher =

work page 2017
[33]

Gradient descent algorithms for

Chewi, Sinho and Maunu, Tyler and Rigollet, Philippe and Stromme,. Gradient descent algorithms for. Proceedings of Thirty Third Conference on Learning Theory , pages =. 2020 , editor =

work page 2020
[34]

2004 , url =

Wilfrid Gangbo , title =. 2004 , url =

work page 2004
[35]

2025 , booktitle =

Kim, Kaheon and Yao, Rentian and Zhu, Changbo and Chen, Xiaohui , month = jul, title =. 2025 , booktitle =

work page 2025
[36]

Sample complexity and weak limits of nonsmooth multimarginal

Li, Pengtao and Chen, Xiaohui , year=. Sample complexity and weak limits of nonsmooth multimarginal

work page
[37]

2025 , archivePrefix =

Learning Density Evolution from Snapshot Data , author=. 2025 , archivePrefix =. 2502.17738 , primaryClass =

work page arXiv 2025
[38]

2017 , issn =

Wasserstein barycenters over Riemannian manifolds , journal =. 2017 , issn =. doi:https://doi.org/10.1016/j.aim.2016.11.026 , url =

work page doi:10.1016/j.aim.2016.11.026 2017
[39]

and Castanon, David A

Bertsekas, Dimitri P. and Castanon, David A. The auction algorithm for the transportation problem. Annals of Operations Research. 1989

work page 1989
[40]

and Ye, Yinyu , title =

Luenberger, David G. and Ye, Yinyu , title =. 2008 , publisher =

work page 2008
[41]

SIAM Journal on Mathematical Analysis , volume =

Carlier, Guillaume and Eichinger, Katharina and Kroshnin, Alexey , title =. SIAM Journal on Mathematical Analysis , volume =. 2021 , doi =

work page 2021
[43]

SIAM Journal on Mathematical Analysis , volume =

Bigot, J\'. SIAM Journal on Mathematical Analysis , volume =. 2019 , doi =

work page 2019
[44]

Li, Lingxiao and Genevay, Aude and Yurochkin, Mikhail and Solomon, Justin M , booktitle =

work page
[45]

Efficient and Exact Multimarginal Optimal Transport with Pairwise Costs , url =

Zhou, Bohan and Parno, Matthew , date =. Efficient and Exact Multimarginal Optimal Transport with Pairwise Costs , url =. Journal of Scientific Computing , number =. 2024 , bdsk-url-1 =. doi:10.1007/s10915-024-02572-8 , id =

work page doi:10.1007/s10915-024-02572-8 2024
[46]

International Conference on Learning Representations , year=

Continuous Wasserstein-2 Barycenter Estimation without Minimax Optimization , author=. International Conference on Learning Representations , year=

work page
[47]

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

The back-and-forth method for Wasserstein gradient flows , author=. ESAIM: Control, Optimisation and Calculus of Variations , volume=. 2021 , publisher=

work page 2021
[48]

SIAM journal on numerical analysis , volume=

Fast Legendre--Fenchel transform and applications to Hamilton--Jacobi equations and conservation laws , author=. SIAM journal on numerical analysis , volume=. 1996 , publisher=

work page 1996
[49]

Advances in Neural Information Processing Systems , volume=

Continuous regularized wasserstein barycenters , author=. Advances in Neural Information Processing Systems , volume=

work page
[50]

Wasserstein Iterative Networks for Barycenter Estimation , volume =

Korotin, Alexander and Egiazarian, Vage and Li, Lingxiao and Burnaev, Evgeny , booktitle =. Wasserstein Iterative Networks for Barycenter Estimation , volume =

work page
[51]

Convolutional

Solomon, Justin and de Goes, Fernando and Peyr\'. Convolutional. ACM Transactions on Graphs , number =

work page
[52]

Kuhn, H. W. , title =. Naval Research Logistics Quarterly , volume =

work page
[53]

, journal=

Srivastava, Sanvesh and Li, Cheng and Dunson, David B. , journal=. Scalable

work page
[54]

Fast Computation of

Cuturi, Marco and Doucet, Arnaud , booktitle =. Fast Computation of

work page
[55]

and Carlier, G

Agueh, M. and Carlier, G. , TITLE =. SIAM J. Math. Anal. , FJOURNAL =. 2011 , NUMBER =

work page 2011
[56]

Zhuang, Yubo and Chen, Xiaohui and Yang, Yun , booktitle =

work page
[57]

Continuous Regularized

Li, Lingxiao and Genevay, Aude and Yurochkin, Mikhail and Solomon, Justin , booktitle =. Continuous Regularized. 2020 , pages =

work page 2020
[58]

Convergence rate of entropy-regularized multi-marginal optimal transport costs , journal=

Nenna, Luca and Pegon, Paul , year=. Convergence rate of entropy-regularized multi-marginal optimal transport costs , journal=

work page
[59]

Wasserstein proximal coordinate gradient algorithms , volume =

Yao, Rentian and Chen, Xiaohui and Yang, Yun , date-added =. Wasserstein proximal coordinate gradient algorithms , volume =. Journal of Machine Learning Research , month = may, number =

work page
[60]

Journal of Machine Learning Research , year =

Jason Altschuler and Enric Boix-Adsera , title =. Journal of Machine Learning Research , year =

work page
[61]

Journal of Multivariate Analysis , volume =

On the computation of. Journal of Multivariate Analysis , volume =. 2020 , issn =

work page 2020
[62]

Wasserstein barycenters are

Altschuler, Jason and Boix-Adsera, Enric , journal=. Wasserstein barycenters are. 2022 , publisher=

work page 2022
[63]

Foundations and Trends

Computational optimal transport: With applications to data science , author=. Foundations and Trends

work page
[64]

The Journal of Machine Learning Research , volume=

On the complexity of approximating multimarginal optimal transport , author=. The Journal of Machine Learning Research , volume=. 2022 , publisher=

work page 2022
[65]

On the linear convergence of the multimarginal

Carlier, Guillaume , journal=. On the linear convergence of the multimarginal. 2022 , publisher=

work page 2022
[66]

Penalization of barycenters in the

Bigot, J\'. Penalization of barycenters in the. SIAM Journal on Mathematical Analysis , volume =

work page
[67]

SIAM Journal on Mathematical Analysis , volume =

Carlier, Guillaume and Eichinger, Katharina and Kroshnin, Alexey , title =. SIAM Journal on Mathematical Analysis , volume =

work page
[68]

Doubly Regularized Entropic

Lénaïc Chizat , year=. Doubly Regularized Entropic. arXiv preprint arXiv:2303.11844 , URL=

work page arXiv
[69]

Jacobs, M. and L\'. A fast approach to optimal transport: the back-and-forth method , JOURNAL =. 2020 , NUMBER =

work page 2020
[70]

, TITLE =

Santambrogio, F. , TITLE =. 2015 , PAGES =

work page 2015
[71]

A gradient descent solution to the

Chartrand, Rick and Wohlberg, Brendt and Vixie, Kevin and Bollt, Erik , journal=. A gradient descent solution to the

work page
[72]

Gradient descent algorithms for

Chewi, Sinho and Maunu, Tyler and Rigollet, Philippe and Stromme, Austin J , booktitle=. Gradient descent algorithms for

work page
[73]

, TITLE =

Zemel, Yoav and Panaretos, Victor M. , TITLE =. Bernoulli , FJOURNAL =. 2019 , NUMBER =. doi:10.3150/17-bej1009 , URL =

work page doi:10.3150/17-bej1009 2019
[74]

Mikha ilov, V. P. , TITLE =. 1978 , PAGES =

work page 1978
[75]

2010 , PAGES =

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

work page doi:10.1090/gsm/019 2010
[76]

Proceedings of The 25th International Conference on Artificial Intelligence and Statistics , pages =

On the complexity of the optimal transport problem with graph-structured cost , author =. Proceedings of The 25th International Conference on Artificial Intelligence and Statistics , pages =. 2022 , publisher =

work page 2022
[77]

International Conference on Machine Learning , pages=

On gradient descent ascent for nonconvex-concave minimax problems , author=. International Conference on Machine Learning , pages=

work page
[78]

International Conference on Artificial Intelligence and Statistics , pages=

Regularity as regularization: Smooth and strongly convex brenier potentials in optimal transport , author=. International Conference on Artificial Intelligence and Statistics , pages=

work page
[79]

The Annals of Statistics , volume=

Plugin estimation of smooth optimal transport maps , author=. The Annals of Statistics , volume=. 2024 , publisher=

work page 2024
[80]

The Annals of Statistics , number =

Jan-Christian H. The Annals of Statistics , number =

work page
[81]

On a theorem of

Bernhard, Pierre and Rapaport, Alain , journal=. On a theorem of. 1995 , publisher=

work page 1995

Showing first 80 references.