arxiv: 2605.03266 · v1 · submitted 2026-05-05 · 📊 stat.ML · cs.LG· math.ST· stat.CO· stat.ME· stat.TH

Recognition: unknown

Intrinsic effective sample size for manifold-valued Markov chain Monte Carlo via kernel discrepancy

Kisung You

Authors on Pith no claims yet

Pith reviewed 2026-05-07 13:44 UTC · model grok-4.3

classification 📊 stat.ML cs.LGmath.STstat.COstat.MEstat.TH

keywords effective sample sizeMarkov chain Monte Carlomanifold-valued datakernel discrepancyintrinsic diagnosticsgeometric statisticsmixing conditions

0 comments

The pith

Kernel discrepancy defines a coordinate-free effective sample size for manifold-valued Markov chain Monte Carlo.

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

The paper establishes an effective sample size for MCMC output on manifolds that avoids dependence on arbitrary coordinate choices or embeddings. Standard ESS measures rely on scalar summaries whose values can shift under rotations or chart changes even when the underlying chain is unchanged. The proposed quantity instead counts the number of independent draws that would produce the same expected squared kernel discrepancy to the target distribution. This construction supplies an exact finite-sample risk reading, an asymptotic link to integrated autocorrelation, and invariance properties whenever the kernel is transported consistently with the manifold geometry.

Core claim

An intrinsic effective sample size is defined as the number of independent samples that would match the expected squared kernel discrepancy of the MCMC empirical measure to the target measure. The definition yields an exact finite-sample risk interpretation, an asymptotic integrated-autocorrelation representation, invariance under transported kernels, and operator and principal-direction interpretations. A lag-window estimator is shown to be consistent under boundedness and absolute-regularity conditions, and valid kernel constructions on manifolds are discussed, noting that geodesic Gaussian kernels are not generally positive definite on curved spaces.

What carries the argument

Squared kernel discrepancy between the empirical distribution of the chain and the target distribution, used to equate MCMC performance to an equivalent number of independent draws.

If this is right

The diagnostic remains unchanged under rotations, chart changes, or alternative embeddings of the same manifold path.
A lag-window estimator of the quantity is consistent whenever the chain meets standard mixing conditions.
The measure admits an interpretation in terms of the reproducing kernel Hilbert space operator and its principal directions.
Only kernels that respect the manifold geometry are admissible; geodesic Gaussian kernels generally fail positive-definiteness on curved spaces.

Where Pith is reading between the lines

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

The same discrepancy-based construction could serve as a diagnostic for other geometry-preserving samplers such as manifold Hamiltonian Monte Carlo.
Researchers could calibrate the measure on additional manifolds like the Stiefel or Grassmannian to check whether the finite-sample risk interpretation holds in practice.
The approach suggests a route to intrinsic diagnostics for other distributional functionals beyond effective sample size.

Load-bearing premise

Kernels exist that are positive definite and compatible with the manifold geometry, and the chain satisfies boundedness and absolute regularity so that the lag-window estimator converges.

What would settle it

On the sphere, rotate the MCMC samples and recompute both the proposed effective sample size and the empirical distributional error; if the effective sample size changes while the error stays fixed, the invariance claim fails.

Figures

Figures reproduced from arXiv: 2605.03266 by Kisung You.

**Figure 1.** Figure 1: Rotation experiment on S 2 . Panel (a) compares ordinary coordinate-wise ESS values with the intrinsic kernel ESS after applying random rotations to the same stored path. The coordinatewise values are shown separately for the three rotated coordinate axes, while the kernel ESS is unchanged; orange lines and green triangles denote medians and means, respectively, and the dashed horizontal line marks the un… view at source ↗

**Figure 2.** Figure 2: Multimodal spherical mixture experiment. Panel (a) shows estimated intrinsic kernel view at source ↗

**Figure 3.** Figure 3: Kernel-risk calibration in the spherical mixture experiment. Panel (a) compares the view at source ↗

read the original abstract

Effective sample size is a standard summary of Markov chain Monte Carlo output, but it is usually attached to scalar or Euclidean summaries chosen by the analyst. For manifold-valued samples this choice is not canonical: coordinate-wise effective sample sizes can change under rotations, chart changes, or alternative embeddings of the same underlying path. We propose an intrinsic effective sample size based on kernel discrepancy. The proposed quantity is the number of independent draws that would yield the same expected squared kernel discrepancy between the empirical distribution and the target distribution. This gives an exact finite-sample risk interpretation, an asymptotic integrated-autocorrelation representation, and a coordinate-free diagnostic whenever the kernel respects the geometry of the state space. We establish invariance under transported kernels, operator and principal-direction interpretations, and consistency of a lag-window estimator under boundedness and absolute-regularity conditions. We also discuss valid kernel constructions on manifolds, emphasizing that geodesic Gaussian kernels are not generally positive definite on curved spaces. Sphere experiments illustrate rotation invariance and calibration of the proposed diagnostic against empirical distributional error.

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 defines an intrinsic ESS for manifold MCMC by equating kernel discrepancy of the chain to that of i.i.d. draws, which cleanly removes coordinate dependence when a suitable kernel exists.

read the letter

Hey, the main thing to know is that this work defines effective sample size for manifold-valued MCMC as the number of independent samples that would produce the same expected squared kernel discrepancy to the target. That construction automatically delivers coordinate-free behavior and a direct finite-sample risk reading, plus the usual asymptotic tie to integrated autocorrelation time. They also sketch operator and principal-direction views and prove consistency for a lag-window estimator under boundedness plus absolute regularity. Sphere runs are used to check that the diagnostic stays stable under rotations and tracks empirical error reasonably well. All of that is new relative to the usual scalar or Euclidean ESS summaries. The paper does a decent job flagging that geodesic Gaussians fail to be positive definite on curved manifolds and spends time on valid kernel constructions instead. That honesty is useful. The soft spots are mostly around practicality. The whole proposal only works once you have a kernel that is both positive definite and geometry-respecting on the manifold of interest; while alternatives are discussed, it is not clear how routine it is to find or tune one for a new manifold without extra theory or computation. The consistency result also leans on mixing conditions that may be hard to verify for some chains. Experiments are mentioned but stay at the illustrative level with no strong baselines or quantitative error tables. This is for people already running MCMC on spheres, rotation groups, or shape spaces who care about diagnostics that do not change with arbitrary embeddings. A reader comfortable with kernel methods and manifold statistics will extract the most. I would send it to peer review; the core definition is coherent and the invariance claim is worth checking in detail even if the kernel side needs more concrete guidance.

Referee Report

3 major / 2 minor

Summary. The manuscript proposes an intrinsic effective sample size (ESS) for manifold-valued Markov chain Monte Carlo (MCMC) based on kernel discrepancy. The quantity is defined as the number of i.i.d. draws that would produce the same expected squared kernel discrepancy between the empirical distribution and the target as the MCMC chain does. It claims this yields an exact finite-sample risk interpretation, an asymptotic integrated-autocorrelation representation, invariance under transported kernels, operator and principal-direction interpretations, and consistency of a lag-window estimator under boundedness and absolute-regularity conditions. Valid kernel constructions on manifolds are discussed (noting geodesic Gaussian kernels are not generally positive definite on curved spaces), and sphere experiments illustrate rotation invariance and calibration against empirical distributional error.

Significance. If the mathematical claims hold, the work would provide a coordinate-free diagnostic for MCMC efficiency on manifolds, filling a gap where standard scalar or Euclidean ESS summaries are not invariant to coordinate choices, chart changes, or embeddings. The finite-sample risk interpretation and invariance properties would be useful in applications such as directional statistics and shape analysis. Credit is due for the explicit discussion of kernel constructions and the operator interpretations, which strengthen the proposal beyond ad-hoc definitions.

major comments (3)

[Kernel constructions and invariance claims] The central definition of the intrinsic ESS rests on the kernel being positive definite and respecting manifold geometry so that the discrepancy is a valid squared RKHS distance and the invariance claims attach. The abstract notes that geodesic Gaussian kernels fail to be positive definite on curved spaces; the manuscript must therefore supply explicit, verifiable constructions for the manifolds of interest (e.g., spheres, SO(3)) together with proofs that the resulting discrepancy yields the claimed finite-sample risk and asymptotic IAC representations, otherwise the load-bearing interpretations do not hold.
[Consistency of lag-window estimator] The abstract asserts consistency of the lag-window estimator under boundedness and absolute-regularity conditions, yet no derivation, error bounds, or explicit statement of the required mixing rates appears in the provided summary. Because this consistency underpins the practical use of the diagnostic, the full manuscript must include the proof (or a clear reference to standard results) with the precise conditions stated.
[Experiments] Sphere experiments are invoked to illustrate rotation invariance and calibration against empirical distributional error, but the summary provides no quantitative results, baselines, or error analysis. Without these, it is impossible to evaluate whether the proposed ESS indeed calibrates correctly or outperforms coordinate-wise alternatives on the claimed manifolds.

minor comments (2)

[Notation and definitions] Clarify the precise definition of the kernel discrepancy (including any normalization) and the exact formula for the intrinsic ESS early in the manuscript so that the finite-sample risk interpretation is immediately readable.
[Assumptions] Ensure all assumptions (positive-definiteness, boundedness, absolute regularity) are collected in a single statement rather than scattered across the text.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the careful and constructive review of our manuscript. The comments identify areas where additional detail will strengthen the presentation, and we address each point below with our planned revisions.

read point-by-point responses

Referee: [Kernel constructions and invariance claims] The central definition of the intrinsic ESS rests on the kernel being positive definite and respecting manifold geometry so that the discrepancy is a valid squared RKHS distance and the invariance claims attach. The abstract notes that geodesic Gaussian kernels fail to be positive definite on curved spaces; the manuscript must therefore supply explicit, verifiable constructions for the manifolds of interest (e.g., spheres, SO(3)) together with proofs that the resulting discrepancy yields the claimed finite-sample risk and asymptotic IAC representations, otherwise the load-bearing interpretations do not hold.

Authors: We agree that explicit constructions are needed to make the invariance and risk interpretations fully verifiable. The manuscript already discusses valid kernel constructions on manifolds and explicitly notes the failure of geodesic Gaussians on curved spaces, proposing geometry-respecting alternatives. To address the request, we will expand this section with concrete, verifiable examples for the sphere (e.g., kernels based on spherical harmonics or the von Mises-Fisher family) and for SO(3) (using bi-invariant kernels on the rotation group), together with short arguments or references confirming that the resulting discrepancy satisfies the finite-sample expected squared risk and the asymptotic integrated autocorrelation representations. revision: yes
Referee: [Consistency of lag-window estimator] The abstract asserts consistency of the lag-window estimator under boundedness and absolute-regularity conditions, yet no derivation, error bounds, or explicit statement of the required mixing rates appears in the provided summary. Because this consistency underpins the practical use of the diagnostic, the full manuscript must include the proof (or a clear reference to standard results) with the precise conditions stated.

Authors: The consistency result is stated in the manuscript under the given conditions, but we accept that a self-contained derivation or precise reference will improve accessibility. In the revision we will add the proof sketch, drawing on standard results for kernel mean embeddings under absolute regularity, and explicitly state the required mixing rates (summability of the alpha-mixing coefficients). If space is limited we will cite the relevant theorem from the dependent-data literature and verify that our boundedness assumption suffices. revision: yes
Referee: [Experiments] Sphere experiments are invoked to illustrate rotation invariance and calibration against empirical distributional error, but the summary provides no quantitative results, baselines, or error analysis. Without these, it is impossible to evaluate whether the proposed ESS indeed calibrates correctly or outperforms coordinate-wise alternatives on the claimed manifolds.

Authors: The sphere experiments in the manuscript demonstrate rotation invariance and calibration, but we acknowledge that more quantitative detail is warranted. We will expand the experimental section to include numerical tables, explicit comparisons against coordinate-wise ESS, baseline methods, and error bars or calibration plots that quantify agreement with empirical distributional error. This will allow direct assessment of performance on the manifolds considered. revision: yes

Circularity Check

0 steps flagged

Intrinsic ESS defined directly via kernel discrepancy without circular reduction to inputs

full rationale

The paper defines the proposed intrinsic effective sample size explicitly as the number of i.i.d. draws yielding the same expected squared kernel discrepancy as the MCMC empirical measure. This is a direct definitional construction rather than a derivation that reduces by construction to fitted parameters, self-referential equations, or load-bearing self-citations. The abstract establishes invariance, operator interpretations, and lag-window consistency under external assumptions (boundedness, absolute regularity, and existence of suitable positive definite kernels), but these are independent properties attached to the definition, not reductions of it. No quoted steps in the provided material exhibit self-definition, fitted-input prediction, or ansatz smuggling; the kernel choice and manifold geometry respect are treated as external preconditions, not internal circularities. The chain is therefore self-contained against the stated benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The proposal depends on the availability of suitable positive definite kernels on the manifold and standard mixing conditions for the chain; no free parameters or new entities are introduced in the abstract.

axioms (2)

domain assumption Existence of positive definite kernels that respect the manifold geometry
Abstract explicitly notes geodesic Gaussian kernels are not generally positive definite on curved spaces, so the method assumes other valid kernels exist.
domain assumption Boundedness and absolute-regularity conditions on the Markov chain
Invoked for consistency of the lag-window estimator of the proposed ESS.

pith-pipeline@v0.9.0 · 5481 in / 1320 out tokens · 61193 ms · 2026-05-07T13:44:26.618273+00:00 · methodology

Review history (2 revisions) →

discussion (0)

Reference graph

Works this paper leans on

300 extracted references · 217 canonical work pages · 3 internal anchors

[1]

Geodesic exponential kernels:

Feragen, Aasa and Lauze, Francois and Hauberg, Soren , month = jun, year =. Geodesic exponential kernels:. 2015. doi:10.1109/CVPR.2015.7298922 , urldate =

work page doi:10.1109/cvpr.2015.7298922 2015
[2]

SIAM Journal on Matrix Analysis and Applications , author =

Geometric. SIAM Journal on Matrix Analysis and Applications , author =. 2007 , pages =. doi:10.1137/050637996 , number =

work page doi:10.1137/050637996 2007
[3]

and Salzmann, Mathieu and Hartley, Richard , editor =

Harandi, Mehrtash T. and Salzmann, Mathieu and Hartley, Richard , editor =. From. Computer. 2014 , pages =. doi:10.1007/978-3-319-10605-2_2 , language =

work page doi:10.1007/978-3-319-10605-2_2 2014
[4]

SIAM Journal on Matrix Analysis and Applications , author =

Theoretically and. SIAM Journal on Matrix Analysis and Applications , author =. 2022 , pages =. doi:10.1137/22M1471729 , language =

work page doi:10.1137/22m1471729 2022
[5]

Geodesics and

Thanwerdas, Yann and Pennec, Xavier , editor =. Geodesics and. Geometric. 2021 , pages =. doi:10.1007/978-3-030-80209-7_11 , language =

work page doi:10.1007/978-3-030-80209-7_11 2021
[6]

Journal of Machine Learning Research , author =

Universality,. Journal of Machine Learning Research , author =. 2011 , pages =

2011
[7]

Bernoulli , author =

Strictly and non-strictly positive definite functions on spheres , volume =. Bernoulli , author =. doi:10.3150/12-BEJSP06 , number =

work page doi:10.3150/12-bejsp06
[8]

Duke Mathematical Journal , author =

Positive definite functions on spheres , volume =. Duke Mathematical Journal , author =. doi:10.1215/S0012-7094-42-00908-6 , number =

work page doi:10.1215/s0012-7094-42-00908-6
[9]

Equivalence of distance-based and RKHS -based statistics in hypothesis testing

Equivalence of distance-based and. The Annals of Statistics , author =. doi:10.1214/13-AOS1140 , number =

work page doi:10.1214/13-aos1140
[10]

and Rasch, Malte J

Gretton, Arthur and Borgwardt, Karsten M. and Rasch, Malte J. and Schölkopf, Bernhard and Smola, Alexander , month = mar, year =. A. Journal of Machine Learning Research , publisher =
[11]

Positive

Bhatia, Rajendra , month = dec, year =. Positive. doi:10.1515/9781400827787 , urldate =

work page doi:10.1515/9781400827787
[12]

Medians and

Arnaudon, Marc and Barbaresco, Frédéric and Yang, Le , editor =. Medians and. Matrix. 2013 , note =. doi:10.1007/978-3-642-30232-9_8 , language =

work page doi:10.1007/978-3-642-30232-9_8 2013
[13]

Econometrics and Statistics , author =

Modeling. Econometrics and Statistics , author =. 2022 , pages =. doi:10.1016/j.ecosta.2021.04.004 , language =

work page doi:10.1016/j.ecosta.2021.04.004 2022
[14]

Riemannian

Petersen, Peter , year =. Riemannian. doi:10.1007/978-0-387-29403-2 , publisher =

work page doi:10.1007/978-0-387-29403-2
[15]

You, Kisung , year =. Finite. doi:10.48550/ARXIV.2604.24895 , abstract =

work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2604.24895
[16]

The Annals of Statistics , author =

Fréchet regression for random objects with. The Annals of Statistics , author =. doi:10.1214/17-AOS1624 , number =

work page doi:10.1214/17-aos1624
[17]

NeuroImage , author =

The. NeuroImage , author =. 2012 , pages =. doi:10.1016/j.neuroimage.2012.02.018 , language =

work page doi:10.1016/j.neuroimage.2012.02.018 2012
[18]

and Pedersen, M.S

Petersen, K.B. and Pedersen, M.S. , month = nov, year =. The
[19]

Science , author =

Prediction of. Science , author =. 2010 , pages =. doi:10.1126/science.1194144 , language =

work page doi:10.1126/science.1194144 2010
[20]

A wrapped normal distribution on hyperbolic space for gradient-based learning , volume =

Nagano, Yoshihiro and Yamaguchi, Shoichiro and Fujita, Yasuhiro and Koyama, Masanori , editor =. A wrapped normal distribution on hyperbolic space for gradient-based learning , volume =. Proceedings of the 36th international conference on machine learning , publisher =. 2019 , pages =

2019
[21]

, year =

Lee, John M. , year =. Riemannian
[22]

Springer Series in Statistics

Le Cam, Lucien and Lo Yang, Grace , year =. Asymptotics in. doi:10.1007/978-1-4612-1166-2 , publisher =

work page doi:10.1007/978-1-4612-1166-2
[23]

The Annals of Mathematical Statistics , author =

Asymptotic. The Annals of Mathematical Statistics , author =. 1971 , pages =. doi:10.1214/aoms/1177693066 , language =

work page doi:10.1214/aoms/1177693066 1971
[24]

The Annals of Mathematical Statistics , author =

Note on the. The Annals of Mathematical Statistics , author =. 1949 , pages =. doi:10.1214/aoms/1177729952 , language =

work page doi:10.1214/aoms/1177729952 1949
[25]

Local asymptotic minimax and admissibility in estimation , language =

Hajek, Jaroslav , year =. Local asymptotic minimax and admissibility in estimation , language =. Proceedings of the
[26]

, year =

Cox, D.R. , year =. Analysis of. doi:10.1201/9781315137391 , urldate =

work page doi:10.1201/9781315137391
[27]

Journal of the Royal Statistical Society Series B: Statistical Methodology , author =

An. Journal of the Royal Statistical Society Series B: Statistical Methodology , author =. 1964 , pages =. doi:10.1111/j.2517-6161.1964.tb00553.x , abstract =

work page doi:10.1111/j.2517-6161.1964.tb00553.x 1964
[28]

Computational Statistics & Data Analysis , author =

Parameter estimation and model-based clustering with spherical normal distribution on the unit hypersphere , volume =. Computational Statistics & Data Analysis , author =. 2022 , pages =. doi:10.1016/j.csda.2022.107457 , urldate =

work page doi:10.1016/j.csda.2022.107457 2022
[29]

, month = nov, year =

Li, Songzi and Rizzo, Maria L. , month = nov, year =. K-groups:. doi:10.48550/arXiv.1711.04359 , abstract =

work page doi:10.48550/arxiv.1711.04359
[30]

Directional

Hauberg, Soren , month = jul, year =. Directional. 2018 21st. doi:10.23919/ICIF.2018.8455242 , urldate =

work page doi:10.23919/icif.2018.8455242 2018
[31]

R., Falorsi, L., De Cao, N., Kipf, T., and Tomczak, J

Davidson, Tim R. and Falorsi, Luca and Cao, Nicola De and Kipf, Thomas and Tomczak, Jakub M. , month = sep, year =. Hyperspherical. doi:10.48550/arXiv.1804.00891 , abstract =

work page doi:10.48550/arxiv.1804.00891
[32]

2022 , note =

Directional. 2022 , note =. doi:10.1007/978-981-19-1044-9 , language =

work page doi:10.1007/978-981-19-1044-9 2022
[33]

On the spherical

You, Kisung and Shung, Dennis , year =. On the spherical. doi:10.48550/ARXIV.2208.11929 , abstract =

work page doi:10.48550/arxiv.2208.11929
[34]

Physical Review E , author =

Fuzzy communities and the concept of bridgeness in complex networks , volume =. Physical Review E , author =. 2008 , pages =. doi:10.1103/PhysRevE.77.016107 , number =

work page doi:10.1103/physreve.77.016107 2008
[35]

Journal of Anthropological Research , author =

An. Journal of Anthropological Research , author =. 1977 , pages =. doi:10.1086/jar.33.4.3629752 , number =

work page doi:10.1086/jar.33.4.3629752 1977
[36]

Journal of Complex Networks , author =

Hydra: a method for strain-minimizing hyperbolic embedding of network- and distance-based data , volume =. Journal of Complex Networks , author =. 2020 , pages =. doi:10.1093/comnet/cnaa002 , number =

work page doi:10.1093/comnet/cnaa002 2020
[37]

Numerische Mathematik , author =

A note on two problems in connexion with graphs , volume =. Numerische Mathematik , author =. 1959 , pages =. doi:10.1007/BF01386390 , number =

work page doi:10.1007/bf01386390 1959
[38]

Foundations of Data Science , author =

Tverberg's theorem and multi-class support vector machines , volume =. Foundations of Data Science , author =. 2025 , pages =. doi:10.3934/fods.2024039 , number =

work page doi:10.3934/fods.2024039 2025
[39]

Biometrika , author =

Multivariate output analysis for. Biometrika , author =. 2019 , pages =. doi:10.1093/biomet/asz002 , language =

work page doi:10.1093/biomet/asz002 2019
[40]

Biometrika , author =

Fréchet analysis of variance for random objects , volume =. Biometrika , author =. 2019 , pages =. doi:10.1093/biomet/asz052 , abstract =

work page doi:10.1093/biomet/asz052 2019
[41]

Proceedings of the American Mathematical Society , author =

Omnibus. Proceedings of the American Mathematical Society , author =. 2016 , pages =. doi:10.1090/proc/13216 , number =

work page doi:10.1090/proc/13216 2016
[42]

, month = may, year =

Ginestet, Cedric E. , month = may, year =. Strong
[43]

Journal of Multivariate Analysis , author =

The. Journal of Multivariate Analysis , author =. 1982 , pages =. doi:10.1016/0047-259X(82)90077-X , language =

work page doi:10.1016/0047-259x(82)90077-x 1982
[44]

Annales de l'institut Henri Poincaré , author =

Les éléments aléatoires de nature quelconque dans un espace distancié , volume =. Annales de l'institut Henri Poincaré , author =. 1948 , pages =

1948
[45]

Boumal, Nicolas , year =. An. doi:10.1017/9781009166164 , language =

work page doi:10.1017/9781009166164
[46]

Mishne, Gal and Wan, Zhengchao and Wang, Yusu and Yang, Sheng , year =. The. Proceedings of the 40th
[47]

and Bedo, Justin , month = jun, year =

Song, Le and Smola, Alex and Gretton, Arthur and Borgwardt, Karsten M. and Bedo, Justin , month = jun, year =. Supervised feature selection via dependence estimation , isbn =. Proceedings of the 24th international conference on. doi:10.1145/1273496.1273600 , language =

work page doi:10.1145/1273496.1273600
[48]

Dryden, I. L. and Mardia, K. V. , year =. Statistical shape analysis , isbn =
[49]

Dryden, I. L. and Mardia, K. V. , year =. Statistical shape analysis with applications in
[50]

Mardia, K. V. and Jupp, Peter E. , year =. Directional statistics , isbn =
[51]

Journal of the Royal Statistical Society: Series B (Methodological) , author =

Maximum. Journal of the Royal Statistical Society: Series B (Methodological) , author =. 1977 , pages =. doi:10.1111/j.2517-6161.1977.tb01600.x , number =

work page doi:10.1111/j.2517-6161.1977.tb01600.x 1977
[52]

Differentiating through the

Lou, Aaron and Katsman, Isay and Jiang, Qingxuan and Belongie, Serge and Lim, Ser-Nam and De Sa, Christopher , year =. Differentiating through the. Proceedings of the 37th
[53]

Qiu, Rui and Yu, Zhou and Zhu, Ruoqing , month = jan, year =. Random. Journal of Machine Learning Research , publisher =
[54]

Computational Statistics & Data Analysis , author =

Medoid splits for efficient random forests in metric spaces , volume =. Computational Statistics & Data Analysis , author =. 2024 , pages =. doi:10.1016/j.csda.2024.107995 , language =

work page doi:10.1016/j.csda.2024.107995 2024
[55]

Journal of Statistical Planning and Inference , author =

Design and relative efficiency in two-phase studies , volume =. Journal of Statistical Planning and Inference , author =. 2012 , pages =. doi:10.1016/j.jspi.2012.04.013 , language =

work page doi:10.1016/j.jspi.2012.04.013 2012
[56]

Riemannian radial distributions on

Chen, Hengchao , year =. Riemannian radial distributions on. doi:10.48550/ARXIV.2405.07852 , abstract =

work page doi:10.48550/arxiv.2405.07852
[57]

IEEE Transactions on Information Theory , author =

Gaussian. IEEE Transactions on Information Theory , author =. 2018 , pages =. doi:10.1109/TIT.2017.2713829 , number =

work page doi:10.1109/tit.2017.2713829 2018
[58]

Journal of Computational and Graphical Statistics , author =

Bayesian. Journal of Computational and Graphical Statistics , author =. 2024 , pages =. doi:10.1080/10618600.2024.2308219 , language =

work page doi:10.1080/10618600.2024.2308219 2024
[59]

Scientific Reports , author =

Navigability of. Scientific Reports , author =. 2017 , pages =. doi:10.1038/s41598-017-08872-4 , abstract =

work page doi:10.1038/s41598-017-08872-4 2017
[60]

Sarkar, Rik , editor =. Low. 2012 , pages =. doi:10.1007/978-3-642-25878-7_34 , booktitle =

work page doi:10.1007/978-3-642-25878-7_34 2012
[61]

Learning continuous hierarchies in the

Nickel, Maximillian and Kiela, Douwe , editor =. Learning continuous hierarchies in the. Proceedings of the 35th. 2018 , pages =

2018
[62]

Poincaré embeddings for learning hierarchical representations , volume =

Nickel, Maximillian and Kiela, Douwe , editor =. Poincaré embeddings for learning hierarchical representations , volume =. Advances in
[63]

Journal of the American Statistical Association , author =

Hyperbolic. Journal of the American Statistical Association , author =. 2026 , pages =. doi:10.1080/01621459.2026.2635077 , language =

work page doi:10.1080/01621459.2026.2635077 2026
[64]

, collaborator =

Abramowitz, Milton and Stegun, Irene A. , collaborator =. Handbook of
[65]

Acemoglu, Daron and Autor, David , year =. Skills,. Handbook of. doi:10.1016/S0169-7218(11)02410-5 , language =

work page doi:10.1016/s0169-7218(11)02410-5
[66]

American Economic Review , author =

The. American Economic Review , author =. 2018 , pages =. doi:10.1257/aer.20160696 , abstract =

work page doi:10.1257/aer.20160696 2018
[67]

An in depth look at the

Adamo, Davide and Corneli, Marco and Vuillien, Manon and Vila, Emmanuelle , year =. An in depth look at the. Forty-second
[68]

doi:10.18637/jss.v050.i05 , language =

Journal of Statistical Software , author =. doi:10.18637/jss.v050.i05 , language =

work page doi:10.18637/jss.v050.i05
[69]

Proceedings of the American Mathematical Society , author =

Riemannian \. Proceedings of the American Mathematical Society , author =. 2011 , pages =. doi:10.1090/S0002-9939-2010-10541-5 , language =

work page doi:10.1090/s0002-9939-2010-10541-5 2011
[70]

SIAM Journal on Control and Optimization , author =

On the. SIAM Journal on Control and Optimization , author =. 2013 , pages =. doi:10.1137/12086282X , language =

work page doi:10.1137/12086282x 2013
[71]

IEEE Transactions on Pattern Analysis and Machine Intelligence , author =

Generalized. IEEE Transactions on Pattern Analysis and Machine Intelligence , author =. 2015 , pages =. doi:10.1109/TPAMI.2014.2353625 , number =

work page doi:10.1109/tpami.2014.2353625 2015
[72]

and Chowdhury, A.K.R

Aggarwal, G. and Chowdhury, A.K.R. and Chellappa, R. , year =. A system identification approach for video-based face recognition , isbn =. Proceedings of the 17th. doi:10.1109/ICPR.2004.1333732 , urldate =

work page doi:10.1109/icpr.2004.1333732 2004
[73]

Barycenters in the

Barycenters in the. SIAM Journal on Mathematical Analysis , author =. 2011 , pages =. doi:10.1137/100805741 , language =

work page doi:10.1137/100805741 2011
[74]

Information

Akaike, Hirotogu , editor =. Information. 1998 , pages =. doi:10.1007/978-1-4612-1694-0_15 , booktitle =

work page doi:10.1007/978-1-4612-1694-0_15 1998
[75]

Data Mining and Knowledge Discovery , author =

Extreme-value-theoretic estimation of local intrinsic dimensionality , volume =. Data Mining and Knowledge Discovery , author =. 2018 , pages =. doi:10.1007/s10618-018-0578-6 , number =

work page doi:10.1007/s10618-018-0578-6 2018
[76]

Zico , editor =

Amos, Brandon and Xu, Lei and Kolter, J. Zico , editor =. Input convex neural networks , volume =. Proceedings of the 34th international conference on machine learning , publisher =. 2017 , pages =

2017
[77]

Scientific Reports , author =

The quest for identifiability in human functional connectomes , volume =. Scientific Reports , author =. 2018 , pages =. doi:10.1038/s41598-018-25089-1 , language =

work page doi:10.1038/s41598-018-25089-1 2018
[78]

Gradient flows: in metric spaces and in the space of probability measures , isbn =

Ambrosio, Luigi and Gigli, Nicola and Savaré, Giuseppe , year =. Gradient flows: in metric spaces and in the space of probability measures , isbn =
[79]

Lectures on

Ambrosio, Luigi and Brué, Elia and Semola, Daniele , year =. Lectures on. doi:10.1007/978-3-72162-6 , language =

work page doi:10.1007/978-3-72162-6
[80]

2003 , keywords =

Optimal transportation and applications: lectures given at the. 2003 , keywords =

2003

Showing first 80 references.