Pith. sign in

REVIEW 1 major objections 5 minor 297 references

A quotient-affine Riemannian Gaussian on full-rank correlation matrices has finite moments, exact p=2/Fisher likelihood, and a center-dependent normalizer for p=3 that separates MLE from Fréchet estimation.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-04 03:25 UTC pith:VISBLEE2

load-bearing objection A serious, honest construction of a quotient-affine Riemannian Gaussian on correlation matrices, with a genuine but not fully exhibited p=3 center-dependence result that needs referee verification. the 1 major comments →

arxiv 2608.02593 v1 pith:VISBLEE2 submitted 2026-08-03 stat.ME

Exact Likelihood and Sampling for Riemannian Gaussian Distributions on Correlation Matrices

Kisung You This is my paper
classification stat.ME
keywords matriceslikelihoodcorrelationexactgaussiansamplingdimensionsdispersion
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

Correlation matrices describe relationships among many variables, but they are not ordinary vectors. If you rescale each variable, the correlation stays the same, so a correlation matrix is an equivalence class of covariance matrices up to diagonal scaling. This paper defines a probability distribution on all full-rank correlation matrices by using a Riemannian distance that respects that equivalence. The density is a Gaussian-like radial model: probability is high near a central correlation matrix, proportional to exp(-distance²/(2σ²)), measured against the manifold's natural volume.

For one correlation coefficient (2×2 matrices), the natural distance is exactly a constant multiple of the Fisher z-transform difference, so the model reproduces classical Fisher inference. In higher dimensions the manifold is curved. The paper computes the scalar curvature of the 3×3 correlation manifold and uses a small-dispersion expansion to show that the normalizing constant—the number making the density integrate to one—depends on the chosen center. That dependence means fitting by maximum likelihood and fitting by Fréchet mean (minimizing average squared distance) are not equivalent; the Fréchet estimate ignores a drift term.

Practically, the normalizer must be estimated by numerical integration, and sampling uses Markov-chain and importance-resampling methods. The paper validates against the exact 2×2 case, compares integration schemes, and demonstrates applications to rolling financial correlations and as a Bayesian prior. Costs are high: evaluation is O(p^6) per volume-density point and efficiency degrades near the boundary, so the method is currently practical mainly for small to moderate dimensions.

Core claim

The quotient-affine Riemannian Gaussian on full-rank correlation matrices is a proper probability model with finite radial moments, exact score and profiled-scale equations, and a normalizing constant that can depend on the center in dimension three: the paper proves Z(C0,σ) ≠ Z(C1,σ) for all sufficiently small σ via the scalar curvature computation Scal(C_a) = -3/(4(1+a²)). If correct, the density in Definition 1 is a valid normalized likelihood, and exact MLE solves Σ Log_{C̄}(C_i) = n E_{C̄,σ}[Log_{C̄}(C)] rather than the Fréchet equation with zero right-hand side. The key sentence from the Discussion: "Beyond the two-dimensional Fisher-transform case, the model center and the population Fréchet mean need not coincide because normalization may vary across the manifold."

Load-bearing premise

The center-dependence of the normalizing constant (Proposition 6) rests on the scalar curvature calculation along C_a in Supplementary S2.9. That proof displays a metric expansion with "omitted terms" justified only by a parity argument and then states the Ricci tensor; the load-bearing premise is that this algebraic computation is correct. If Scal(C_a) = -3/[4(1+a²)] is wrong, the proof that Z varies with the center collapses, and with it the MLE-versus-Fréchet separation. The calculation is checked only by an accompanying script rather than fully exhibited in the manuscript.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

1 major / 5 minor

Summary. This paper proposes a Riemannian Gaussian distribution on the manifold of full-rank correlation matrices, using the quotient-affine geometry obtained from the SPD cone by positive diagonal scaling. The density is proportional to exp{-rho(C,Cbar)^2/(2 sigma^2)} with respect to the quotient-affine volume. The authors prove propriety and finite radial moments, derive exact score and profiled-scale equations, recover Fisher-transformed Gaussian inference for p=2, and claim that in dimension p=3 the normalizing constant depends on the center, implying that exact MLE and Frechet estimation can have different population targets. They develop numerical integration (cubature, RQMC, importance sampling), likelihood fitting, and sampling algorithms, and evaluate them in simulations and two applications. The central theoretical claim beyond p=2 rests on a scalar-curvature calculation in Proposition 6.

Significance. If the curvature computation is correct, the paper makes a substantive contribution: a normalized intrinsic likelihood on the elliptope, a clear distinction between the model center and the Frechet mean, and a principled maximum-entropy justification for the radial density. The exact score equations, the p=2 closed-form benchmark, and the sampling-correctness theorems are valuable. The computational framework is carefully reported with reproducibility materials, a failure audit, and diagnostics. However, the main novelty depends on a single algebraic curvature calculation that is not fully exhibited in the manuscript.

major comments (1)
  1. [Supplementary S2.9, Eq. (16)] Proposition 6 is the linchpin of the paper's central claim. The second-order metric expansion is asserted with 'Direct substitution in Lemma 1 gives' without showing the algebra; the parity argument only explains the zero quadratic terms in G_rs and G_rt, not the coefficients in G_rr, G_ss, G_tt, and G_st. The stated Ricci tensor is then introduced without intermediate steps. The accompanying script numerically evaluates the same construction and is not an independent verification. Please supply a complete derivation or a symbolic check (e.g., exact rational arithmetic or a CAS transcript).
minor comments (5)
  1. [Throughout] There are numerous spelling/ligature errors such as 'efficiency' rendered as 'efficiency' in the abstract, Section 5.4, and Section 6.2. A careful copyedit is needed.
  2. [Section 5.2] The phrase 'approximately 6 < d ≤ 15' is ambiguous; likely '6 ≤ d ≤ 15' or 'about 6 to 15' was intended.
  3. [Algorithm 1] The multistart set D_0 is not specified. Its size and construction affect the reliability of the distance/gradient computation; please provide the default choice.
  4. [Section 6.2] The reported discrepancies '−0.0239 at p=3' and '−0.2667 at p=5' should state explicitly that these are differences in log Z, to avoid confusion.
  5. [Supplementary S2.9] The script verify_p3_scalar_curvature.py is mentioned but not included in the supplement; if it is the sole numerical check, its code should be reproduced or referenced in an archived repository.

Circularity Check

0 steps flagged

No circularity: the central derivation is self-contained and the p=3 center-dependence is derived from an independent curvature computation plus a standard local expansion.

full rationale

I examined the full derivation chain. Definition 1 is well-posed because Proposition 1 proves exponential volume growth and finite radial moments, so the normalizing constant in (11) is finite; this does not presuppose its center dependence. The p=2 result (Proposition 5) is derived explicitly from the quotient-affine line element, giving Z = sqrt(2)pi sigma independent of center, and serves as an external analytic benchmark. The main novelty, center dependence for p=3 (Proposition 6), is obtained by combining an explicit scalar-curvature calculation in Supplementary S2.9 with the small-dispersion expansion of Proposition 9. The curvature computation is carried out in local coordinates from Lemma 1; it does not define the conclusion into the input, and Proposition 9 is proved independently from normal-coordinate volume expansions. The score equation (18) follows from differentiating log Z under the integral using the domination argument in S2.11; the Fréchet drift (Corollary 2) is then a consequence of the score identity, not an assumption. Numerical sections are used only as implementation checks, and the paper explicitly reports failures and the limited scope of the curvature correction. I found no load-bearing self-citation, no fitted parameter renamed as a prediction, and no uniqueness theorem imported from the authors' prior work. The unexhibited quadratic metric coefficients and Ricci contraction in S2.9 are a correctness risk rather than a circularity: if the algebra were wrong, Proposition 6 would fail, but the argument does not reduce to its own conclusion.

Axiom & Free-Parameter Ledger

0 free parameters · 4 axioms · 0 invented entities

The model introduces no new free parameters: σ and the center C̄ are statistical parameters estimated from data, not ad hoc constants fitted to force the derivation. No new entities are postulated. The mathematical work rests on standard Riemannian geometry; the only paper-specific algebraic object is the metric tensor in the normalized-Cholesky chart, which is derived, not assumed.

axioms (4)
  • standard math The affine-invariant SPD manifold (S_p++, g_AI) is a complete nonpositively curved symmetric space whose sectional curvature is bounded below by a finite constant.
    Used in the proof of Proposition 1(d) to obtain exponential volume growth via Bishop–Gromov; see S2.3.
  • standard math For a free, proper, isometric action, the quotient has the structure of a Riemannian submersion; O'Neill's curvature formulas and Hopf-Rinow apply.
    Used in Proposition 1 to establish completeness, distance attainment, and the curvature lower bound; see S2.3.
  • standard math Bishop–Gromov volume comparison for manifolds with Ricci curvature bounded below.
    Gives the exponential bound on metric balls in Proposition 1(d).
  • standard math The Schur product theorem implies C^{-1}∘C is positive semidefinite for positive definite C.
    Used in Lemma 1 to guarantee that I + B(C) is positive definite and the horizontal-lift system has a unique solution.

pith-pipeline@v1.3.0-daily-deepseek · 25078 in / 15696 out tokens · 223925 ms · 2026-08-04T03:25:34.263584+00:00 · methodology

0 comments
read the original abstract

Correlation matrices arise when marginal scales are removed from covariance matrices, yet a normalized likelihood must account for both quotient distance and quotient volume. We propose a Riemannian Gaussian model for full-rank correlation matrices under quotient-affine geometry. The distribution is proper and has finite radial moments. We derive exact score and profiled-scale equations and recover Fisher-transformed Gaussian inference for two-dimensional matrices. In higher dimension, a curvature calculation shows that the normalizing constant can vary with the center. Exact maximum likelihood and Fr\'{e}chet estimation may therefore have different population targets. We develop chart-based methods for evaluating the normalizer, fitting the likelihood, and sampling. Numerical studies verify the analytic case and compare integration, estimation, and sampling procedures across dimensions and dispersion regimes. A rolling-finance application and a controlled prior study illustrate both the value and computational cost of the model. The method is most reliable in small to moderate dimensions, while proposal efficiency and numerical conditioning deteriorate near the boundary and at larger dispersion.

Figures

Figures reproduced from arXiv: 2608.02593 by Kisung You.

Figure 1
Figure 1. Figure 1: Schematic quotient structure. Curves across the gray ambient manifold [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Normalizer estimates for the 𝑝 = 3 block center. Stochastic points are means across 20 completed randomized replications, while deterministic curves are single evalua￾tions. Because standard errors are smaller than the main-panel markers, the inset magnifies paired importance-minus-RQMC differences. Inset bars show one Monte Carlo standard error of each paired mean difference. The leading expression’s mean… view at source ↗
Figure 5
Figure 5. Figure 5: Estimated centers of the two-component QA Fréchet mixture fitted to 50 nonover [PITH_FULL_IMAGE:figures/full_fig_p033_5.png] view at source ↗
Figure 3
Figure 3. Figure 3: Descriptive financial sequence based on 60-trading-day windows advanced by 20 [PITH_FULL_IMAGE:figures/full_fig_p039_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Held-out log densities for the 21 nonoverlapping financial test windows. The [PITH_FULL_IMAGE:figures/full_fig_p040_4.png] view at source ↗
Figure 6
Figure 6. Figure 6: Posterior comparison for the QA and LKJ priors over 50 replications per scenario. [PITH_FULL_IMAGE:figures/full_fig_p041_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Prior squared QA distance from the supplied template under the fixed hyperpa [PITH_FULL_IMAGE:figures/full_fig_p041_7.png] view at source ↗

discussion (0)

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

Reference graph

Works this paper leans on

297 extracted references · 32 canonical work pages

  1. [1]

    and Burton, R

    Aaronson, J. and Burton, R. and Dehling, H. and Gilat, D. and Hill, T. and Weiss, B. , year = 1996, journal =. Strong Laws for \. doi:10.1090/S0002-9947-96-01681-9 , urldate =

    doi:10.1090/s0002-9947-96-01681-9 1996

  2. [2]

    Geodesic

    Abbas, Kausar and Liu, Mintao and Venkatesh, Manasij and Amico, Enrico and Kaplan, Alan David and Ventresca, Mario and Pessoa, Luiz and Harezlak, Jaroslaw and Goñi, Joaquín , year = 2021, month = jun, journal =. Geodesic. doi:10.1089/brain.2020.0881 , urldate =

    arXiv 2021

  3. [3]

    , year = 1972, series =

    Abramowitz, Milton and Stegun, Irene A. , year = 1972, series =. Handbook of

    1972

  4. [5]

    and Mahony, R

    Absil, P.-A. and Mahony, R. and Sepulchre, R. , year = 2008, publisher =. Optimization

    2008

  5. [6]

    Acemoglu, Daron and Autor, David , year = 2011, volume =. Skills,. Handbook of. doi:10.1016/S0169-7218(11)02410-5 , urldate =

    doi:10.1016/s0169-7218(11)02410-5 2011

  6. [7]

    Acemoglu, Daron and Restrepo, Pascual , year = 2018, month = jun, journal =. The. doi:10.1257/aer.20160696 , urldate =

    doi:10.1257/aer.20160696 2018

  7. [8]

    An in Depth Look at the

    Adamo, Davide and Corneli, Marco and Vuillien, Manon and Vila, Emmanuelle , year = 2025, file =. An in Depth Look at the. Forty-Second

    2025

  8. [9]

    and Cook, R

    Adragni, Kofi P. and Cook, R. Dennis and Wu, Seongho , year = 2012, journal =. doi:10.18637/jss.v050.i05 , urldate =

    doi:10.18637/jss.v050.i05 2012

  9. [10]

    Riemannian \

    Afsari, Bijan , year = 2011, month = feb, journal =. Riemannian \. doi:10.1090/S0002-9939-2010-10541-5 , urldate =

    doi:10.1090/s0002-9939-2010-10541-5 2011

  10. [11]

    Afsari, Bijan and Tron, Roberto and Vidal, René , year = 2013, month = jan, journal =. On the. doi:10.1137/12086282X , urldate =

    doi:10.1137/12086282x 2013

  11. [12]

    Generalized

    Aftab, Khurrum and Hartley, Richard and Trumpf, Jochen , year = 2015, month = apr, journal =. Generalized. doi:10.1109/TPAMI.2014.2353625 , urldate =

    arXiv 2015

  12. [13]

    and Chowdhury, A.K.R

    Aggarwal, G. and Chowdhury, A.K.R. and Chellappa, R. , year = 2004, pages =. A System Identification Approach for Video-Based Face Recognition , booktitle =. doi:10.1109/ICPR.2004.1333732 , urldate =

    Pith/arXiv arXiv 2004

  13. [14]

    Barycenters in the

    Agueh, Martial and Carlier, Guillaume , year = 2011, month = jan, journal =. Barycenters in the. doi:10.1137/100805741 , langid =

    doi:10.1137/100805741 2011

  14. [15]

    Information

    Akaike, Hirotogu , editor =. Information. Selected. doi:10.1007/978-1-4612-1694-0_15 , isbn =

    doi:10.1007/978-1-4612-1694-0_15

  15. [16]

    , year = 1970, month = aug, journal =

    Akerlof, George A. , year = 1970, month = aug, journal =. The. doi:10.2307/1879431 , urldate =

    doi:10.2307/1879431 1970

  16. [17]

    Nursing Open , volume =

    An Integrative Review of the Use of the Concept of Reassurance in Clinical Practice , author =. Nursing Open , volume =. doi:10.1002/nop2.1102 , langid =

    doi:10.1002/nop2.1102

  17. [18]

    and Chib, Siddhartha , year = 1993, month = jun, journal =

    Albert, James H. and Chib, Siddhartha , year = 1993, month = jun, journal =. Bayesian. doi:10.1080/01621459.1993.10476321 , urldate =

    arXiv 1993

  18. [19]

    and Border, Kim C

    Aliprantis, Charalambos D. and Border, Kim C. , year = 2006, publisher =. Infinite. doi:10.1007/3-540-29587-9 , urldate =

    doi:10.1007/3-540-29587-9 2006

  19. [20]

    and Damaraju, Eswar and Plis, Sergey M

    Allen, Elena A. and Damaraju, Eswar and Plis, Sergey M. and Erhardt, Erik B. and Eichele, Tom and Calhoun, Vince D. , year = 2014, month = mar, journal =. Tracking. doi:10.1093/cercor/bhs352 , urldate =

    work page doi:10.1093/cercor/bhs352 2014

  20. [21]

    and Schapire, Robert E

    Allwein, Erin L. and Schapire, Robert E. and Singer, Yoram , year = 2001, month = sep, journal =. Reducing. doi:10.1162/15324430152733133 , abstract =

    doi:10.1162/15324430152733133 2001

  21. [22]

    Alon, Noga and Matias, Yossi and Szegedy, Mario , year = 1999, month = feb, journal =. The. doi:10.1006/jcss.1997.1545 , urldate =

    arXiv 1999

  22. [23]

    Averaging on the

    Altschuler, Jason and Chewi, Sinho and Gerber, Patrik R and Stromme, Austin , editor =. Averaging on the. Advances in

  23. [24]

    Journal of Machine Learning Research , volume =

    Wasserstein Barycenters Can Be Computed in Polynomial Time in Fixed Dimension , author =. Journal of Machine Learning Research , volume =

  24. [25]

    Pattern Recognition , volume =

    Comparative Study on Classifying Human Activities with Miniature Inertial and Magnetic Sensors , author =. Pattern Recognition , volume =. doi:10.1016/j.patcog.2010.04.019 , urldate =

    doi:10.1016/j.patcog.2010.04.019 2010

  25. [26]

    Journal of Mathematical Analysis and Applications , volume =

    A Fixed-Point Approach to Barycenters in. Journal of Mathematical Analysis and Applications , volume =. doi:10.1016/j.jmaa.2016.04.045 , urldate =

    doi:10.1016/j.jmaa.2016.04.045 2016

  26. [27]

    IEEE Transactions on Information Theory , volume =

    Information Geometry on Hierarchy of Probability Distributions , author =. IEEE Transactions on Information Theory , volume =. doi:10.1109/18.930911 , urldate =

    doi:10.1109/18.930911

  27. [28]

    Methods of

    Amari, Shun'ichi and Nagaoka, Hiroshi and Amari, Shun'ichi and Amari, Shun'ichi , translator =. Methods of

  28. [29]

    Optimal Transportation and Applications: Lectures given at the

  29. [30]

    Gradient Flows: In Metric Spaces and in the Space of Probability Measures , shorttitle =

    Ambrosio, Luigi and Gigli, Nicola and Savaré, Giuseppe , year = 2005, series =. Gradient Flows: In Metric Spaces and in the Space of Probability Measures , shorttitle =

    2005

  30. [31]

    Lectures on

    Ambrosio, Luigi and Brué, Elia and Semola, Daniele , year = 2021, series =. Lectures on. doi:10.1007/978-3-72162-6 , isbn =

    doi:10.1007/978-3-72162-6 2021

  31. [32]

    TEST , volume =

    Mode Testing, Critical Bandwidth and Excess Mass , author =. TEST , volume =. doi:10.1007/s11749-018-0611-5 , urldate =

    doi:10.1007/s11749-018-0611-5

  32. [33]

    Scientific Reports , volume =

    The Quest for Identifiability in Human Functional Connectomes , author =. Scientific Reports , volume =. doi:10.1038/s41598-018-25089-1 , urldate =

    doi:10.1038/s41598-018-25089-1

  33. [34]

    Zico , editor =

    Amos, Brandon and Xu, Lei and Kolter, J. Zico , editor =. Input Convex Neural Networks , booktitle =

  34. [35]

    Data Mining and Knowledge Discovery , volume =

    Extreme-Value-Theoretic Estimation of Local Intrinsic Dimensionality , author =. Data Mining and Knowledge Discovery , volume =. doi:10.1007/s10618-018-0578-6 , urldate =

    doi:10.1007/s10618-018-0578-6

  35. [36]

    Anderson, Edgar , year = 1936, month = sep, journal =. The. doi:10.2307/2394164 , urldate =. 2394164 , eprinttype =

    doi:10.2307/2394164 1936

  36. [37]

    Anderson, T. W. , year = 1963, month = mar, journal =. Asymptotic. doi:10.1214/aoms/1177704248 , urldate =

    arXiv 1963

  37. [38]

    Stochastic Processes and their Applications , volume =

    Reverse-Time Diffusion Equation Models , author =. Stochastic Processes and their Applications , volume =. doi:10.1016/0304-4149(82)90051-5 , urldate =

    doi:10.1016/0304-4149(82)90051-5

  38. [39]

    and Bates, Stephen , year = 2022, month = dec, number =

    Angelopoulos, Anastasios N. and Bates, Stephen , year = 2022, month = dec, number =. A. doi:10.48550/arXiv.2107.07511 , urldate =. arXiv , keywords =:2107.07511 , primaryclass =

    Pith/arXiv arXiv doi:10.48550/arxiv.2107.07511 2022

  39. [40]

    Minimizing

    Angenent, Sigurd and Haker, Steven and Tannenbaum, Allen , year = 2003, month = jan, journal =. Minimizing. doi:10.1137/S0036141002410927 , urldate =

    doi:10.1137/s0036141002410927 2003

  40. [41]

    Anselin, Luc , year = 2019, month = apr, journal =. A. doi:10.1111/gean.12164 , urldate =

    work page doi:10.1111/gean.12164 2019

  41. [42]

    Tobler’s

    Anselin, Luc and Li, Xun , year = 2020, month = oct, journal =. Tobler’s. doi:10.1111/gean.12237 , urldate =

    work page doi:10.1111/gean.12237 2020

  42. [43]

    Pattern Recognition , volume =

    An Extensive Comparative Study of Cluster Validity Indices , author =. Pattern Recognition , volume =. doi:10.1016/j.patcog.2012.07.021 , urldate =

    doi:10.1016/j.patcog.2012.07.021 2012

  43. [45]

    and Gine, Evarist , year = 1993, month = jul, journal =

    Arcones, Miguel A. and Gine, Evarist , year = 1993, month = jul, journal =. Limit. doi:10.1214/aop/1176989128 , urldate =

    arXiv 1993

  44. [46]

    Wasserstein Generative Adversarial Networks , booktitle =

    Arjovsky, Martin and Chintala, Soumith and Bottou, Léon , editor =. Wasserstein Generative Adversarial Networks , booktitle =

  45. [47]

    Medians and

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

    doi:10.1007/978-3-642-30232-9_8

  46. [48]

    , year = 1950, journal =

    Aronszajn, N. , year = 1950, journal =. Theory of. doi:10.1090/S0002-9947-1950-0051437-7 , urldate =

    doi:10.1090/s0002-9947-1950-0051437-7 1950

  47. [49]

    Arsigny, Vincent and Fillard, Pierre and Pennec, Xavier and Ayache, Nicholas , year = 2006, month = aug, journal =. Log‐. doi:10.1002/mrm.20965 , urldate =

    doi:10.1002/mrm.20965 2006

  48. [50]

    Geometric

    Arsigny, Vincent and Fillard, Pierre and Pennec, Xavier and Ayache, Nicholas , year = 2007, month = jan, journal =. Geometric. doi:10.1137/050637996 , urldate =

    doi:10.1137/050637996 2007

  49. [51]

    , year = 2018, month = oct, journal =

    Assa, Akbar and Plataniotis, Konstantinos N. , year = 2018, month = oct, journal =. Wasserstein-. doi:10.1109/LSP.2018.2865829 , urldate =

    arXiv 2018

  50. [52]

    Åström, Freddie and Petra, Stefania and Schmitzer, Bernhard and Schnörr, Christoph , year = 2017, month = jun, journal =. Image. doi:10.1007/s10851-016-0702-4 , urldate =

    work page doi:10.1007/s10851-016-0702-4 2017

  51. [53]

    The Annals of Statistics , volume =

    Generalized Random Forests , author =. The Annals of Statistics , volume =. doi:10.1214/18-AOS1709 , urldate =

    doi:10.1214/18-aos1709

  52. [54]

    Convergence of Descent Methods for Semi-Algebraic and Tame Problems: Proximal Algorithms, Forward–Backward Splitting, and Regularized

    Attouch, Hedy and Bolte, Jérôme and Svaiter, Benar Fux , year = 2013, month = feb, journal =. Convergence of Descent Methods for Semi-Algebraic and Tame Problems: Proximal Algorithms, Forward–Backward Splitting, and Regularized. doi:10.1007/s10107-011-0484-9 , urldate =

    doi:10.1007/s10107-011-0484-9 2013

  53. [55]

    Statistics and Computing , volume =

    Clustering via Nonparametric Density Estimation , author =. Statistics and Computing , volume =. doi:10.1007/s11222-006-9010-y , urldate =

    doi:10.1007/s11222-006-9010-y

  54. [56]

    and Baudin, Michel and Duda, Richard O

    Babaud, Jean and Witkin, Andrew P. and Baudin, Michel and Duda, Richard O. , year = 1986, month = jan, journal =. Uniqueness of the. doi:10.1109/TPAMI.1986.4767749 , urldate =

    arXiv 1986

  55. [57]

    Bahadur, R. R. , year = 1966, month = jun, journal =. A. doi:10.1214/aoms/1177699450 , urldate =

    arXiv 1966

  56. [58]

    24306018 , eprinttype =

    Bai, Zhidong and Saranadasa, Hewa , year = 1996, journal =. 24306018 , eprinttype =

    1996

  57. [59]

    Analysis and

    Bakry, Dominique and Gentil, Ivan and Ledoux, Michel , year = 2014, series =. Analysis and. doi:10.1007/978-3-319-00227-9 , urldate =

    doi:10.1007/978-3-319-00227-9 2014

  58. [60]

    Balasubramanian, Krishnakumar and Li, Tong and Yuan, Ming , year = 2021, journal =. On the

    2021

  59. [61]

    Neural Networks and Principal Component Analysis:

    Baldi, Pierre and Hornik, Kurt , year = 1989, month = jan, journal =. Neural Networks and Principal Component Analysis:. doi:10.1016/0893-6080(89)90014-2 , urldate =

    doi:10.1016/0893-6080(89)90014-2 1989

  60. [63]

    Understanding Uncertainty in

    Balocchi, Cecilia and Wade, Sara , year = 2025, publisher =. Understanding Uncertainty in. doi:10.48550/ARXIV.2506.16295 , urldate =

    doi:10.48550/arxiv.2506.16295 2025

  61. [64]

    Journal of Machine Learning Research , volume =

    Clustering on the Unit Hypersphere Using von Mises-Fisher Distributions , author =. Journal of Machine Learning Research , volume =

  62. [65]

    Biosystems , volume =

    Structural Distance and Evolutionary Relationship of Networks , author =. Biosystems , volume =. doi:10.1016/j.biosystems.2011.11.004 , urldate =

    work page doi:10.1016/j.biosystems.2011.11.004 2011

  63. [66]

    , year = 2005, month = dec, journal =

    Bang, Heejung and Robins, James M. , year = 2005, month = dec, journal =. Doubly. doi:10.1111/j.1541-0420.2005.00377.x , urldate =

    arXiv 2005

  64. [67]

    Bao, Dianbin and You, Kisung and Lin, Lizhen , year = 2022, month = dec, pages =. Network. 2022. doi:10.1109/BigData55660.2022.10020500 , urldate =

    arXiv 2022

  65. [68]

    Classification of Covariance Matrices Using a

    Barachant, Alexandre and Bonnet, Stéphane and Congedo, Marco and Jutten, Christian , year = 2013, month = jul, journal =. Classification of Covariance Matrices Using a. doi:10.1016/j.neucom.2012.12.039 , urldate =

    doi:10.1016/j.neucom.2012.12.039 2013

  66. [69]

    Barachant, Alexandre , year = 2015, month = jun, doi =

    2015

  67. [70]

    Colourful Linear Programming , booktitle =

    Bárány, Imre and Onn, Shmuel , editor =. Colourful Linear Programming , booktitle =. doi:10.1007/3-540-61310-2_1 , urldate =

    doi:10.1007/3-540-61310-2_1

  68. [71]

    Colourful

    Bárány, Imre and Onn, Shmuel , year = 1997, month = aug, journal =. Colourful

    1997

  69. [72]

    Birds of the

    Barberá, Pablo , year = 2015, journal =. Birds of the. doi:10.1093/pan/mpu011 , urldate =

    doi:10.1093/pan/mpu011 2015

  70. [73]

    NeuroImage , volume =

    Uncovering Multi-Site Identifiability Based on Resting-State Functional Connectomes , author =. NeuroImage , volume =. doi:10.1016/j.neuroimage.2019.06.045 , urldate =

    work page doi:10.1016/j.neuroimage.2019.06.045 2019

  71. [74]

    Metrika , volume =

    A Consistent Test for Multivariate Normality Based on the Empirical Characteristic Function , author =. Metrika , volume =. doi:10.1007/BF02613322 , urldate =

    doi:10.1007/bf02613322

  72. [75]

    Barnard, John and McCulloch, Robert and Meng, Xiao-Li , year = 2000, month = oct, journal =

    2000

  73. [76]

    Barnett, Ian and Onnela, Jukka-Pekka , year = 2016, month = may, journal =. Change. doi:10.1038/srep18893 , urldate =

    work page doi:10.1038/srep18893 2016

  74. [77]

    Nature Neuroscience , volume =

    Prefrontal Cortex and Decision Making in a Mixed-Strategy Game , author =. Nature Neuroscience , volume =. doi:10.1038/nn1209 , urldate =

    work page doi:10.1038/nn1209

  75. [78]

    Analyzing Biological and Artificial Neural Networks: Challenges with Opportunities for Synergy? , shorttitle =

    Barrett, David GT and Morcos, Ari S and Macke, Jakob H , year = 2019, month = apr, journal =. Analyzing Biological and Artificial Neural Networks: Challenges with Opportunities for Synergy? , shorttitle =. doi:10.1016/j.conb.2019.01.007 , urldate =

    work page doi:10.1016/j.conb.2019.01.007 2019

  76. [79]

    and Sheu, Chyong-Hwa , year = 1991, month = sep, journal =

    Barron, Andrew R. and Sheu, Chyong-Hwa , year = 1991, month = sep, journal =. Approximation of. doi:10.1214/aos/1176348252 , urldate =

    arXiv 1991

  77. [80]

    and Mendelson, Shahar , editor =

    Bartlett, Peter L. and Mendelson, Shahar , editor =. Rademacher and. Computational. doi:10.1007/3-540-44581-1_15 , urldate =

    doi:10.1007/3-540-44581-1_15

  78. [81]

    and Wegkamp, Marten H

    Bartlett, Peter L. and Wegkamp, Marten H. , year = 2008, journal =. Classification with a

    2008

  79. [82]

    , year = 1992, month = dec, journal =

    Barton, Russell R. , year = 1992, month = dec, journal =. doi:10.1080/03052159208941281 , urldate =

    work page doi:10.1080/03052159208941281 1992

  80. [83]

    Proceedings of the National Academy of Sciences , volume =

    Iterative Random Forests to Discover Predictive and Stable High-Order Interactions , author =. Proceedings of the National Academy of Sciences , volume =. doi:10.1073/pnas.1711236115 , urldate =

    doi:10.1073/pnas.1711236115

Showing first 80 references.