pith. sign in

arxiv: 2605.23207 · v1 · pith:6SH6PWPVnew · submitted 2026-05-22 · 📊 stat.ME

Mixture-of-Finite-Mixtures Wishart Model for Clustering Covariance Matrices with an Application to Brain Functional Connectivity

Zongyu Li , Stefano Castruccio , Zhiyong Zhang This is my paper

Pith reviewed 2026-05-25 04:12 UTC · model grok-4.3

classification 📊 stat.ME
keywords mixture modelWishart distributionclusteringcovariance matricesBayesian inferenceposterior consistencyfunctional connectivityMCMC
0
0 comments X

The pith

The MFM-Wishart model performs Bayesian clustering of covariance matrices while jointly inferring the number of clusters from the data.

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

The paper introduces a Bayesian model for clustering data represented as covariance matrices by combining Wishart mixture components with a mixture-of-finite-mixtures prior. This setup permits simultaneous posterior inference on both the cluster assignments and the total number of clusters. The authors prove that the model achieves posterior consistency for the number of clusters and contraction of the mixing measure under standard regularity conditions. They also provide an efficient MCMC algorithm and demonstrate the approach through simulations and an application to infant brain functional connectivity data from fNIRS.

Core claim

The MFM-Wishart model combines Wishart mixture components with a mixture-of-finite-mixtures (MFM) prior, allowing joint posterior inference on both the number of clusters and clustering assignments for covariance matrix data. Theoretical results establish posterior consistency for the number of clusters and posterior contraction of the mixing measure under standard regularity conditions. An efficient MCMC algorithm supports posterior inference, with simulations confirming competitive performance and accurate cluster number recovery even under misspecification.

What carries the argument

The mixture-of-finite-mixtures (MFM) prior paired with Wishart kernel components, which enables automatic determination of the number of clusters in the posterior distribution for covariance matrix clustering.

If this is right

  • The model recovers the true number of clusters accurately in simulation studies even when the data-generating process differs from the assumed model.
  • An efficient MCMC algorithm allows practical computation of the joint posterior.
  • Application to fNIRS data reveals interpretable heterogeneity in infant functional connectivity during sleep.
  • The theoretical consistency results hold under standard regularity conditions for the Wishart kernels.

Where Pith is reading between the lines

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

  • This framework could extend to clustering other positive definite matrix data in fields like finance or image processing.
  • Future work might explore robustness when the regularity conditions are violated in high-dimensional settings.
  • The joint inference on cluster number reduces the need for separate model selection steps compared to fixed-K mixture models.

Load-bearing premise

The Wishart kernel combined with standard regularity conditions is sufficient to guarantee the posterior consistency for the number of clusters and the contraction of the mixing measure.

What would settle it

A dataset generated from a known number of Wishart components where the MCMC posterior fails to concentrate on that number despite satisfying the regularity conditions.

Figures

Figures reproduced from arXiv: 2605.23207 by Stefano Castruccio, Zhiyong Zhang, Zongyu Li.

Figure 1
Figure 1. Figure 1: Cluster-specific scale matrix Σk in the small-matrix (p = 3) simulations. singleton clusters, and use a Gaussian random-walk proposal with standard deviation 1.0 to update ν. The final clustering estimate for MFM–Wishart is obtained by applying Dahl’s method to the retained MCMC samples. Following Yin et al. [2023], we compare our MFM–Wishart model with a DPM of Wishart kernels, hereafter referred to as DP… view at source ↗
Figure 2
Figure 2. Figure 2: Comparison of the prior behavior and empirical performance of MFM–Wishart [PITH_FULL_IMAGE:figures/full_fig_p019_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: ARI under the balanced and unbalanced settings for the small-matrix ( [PITH_FULL_IMAGE:figures/full_fig_p021_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Cluster-wise mean correlation matrices for the Dahl partition obtained by applying [PITH_FULL_IMAGE:figures/full_fig_p026_4.png] view at source ↗
read the original abstract

Data represented as covariance-type matrices arise in many fields, including brain functional connectivity and diffusion tensor imaging. We develop the MFM-Wishart, a Bayesian model-based clustering approach for such data that combines Wishart mixture components with a mixture-of-finite-mixtures (MFM) prior, allowing joint posterior inference on both the number of clusters and clustering assignments. Theoretically, we study the properties of Wishart kernels in the context of mixture models and then establish results for posterior consistency for the number of clusters and posterior contraction of the mixing measure under standard regularity conditions. Computationally, we develop an efficient Markov chain Monte Carlo (MCMC) algorithm for posterior inference. Simulation studies show competitive clustering performance and accurate recovery of the number of clusters, even under model misspecification. We apply MFM-Wishart to cluster infants based on functional connectivity during sleep, estimated from functional near-infrared spectroscopy (fNIRS) data, illustrating the practical utility of the model and revealing interpretable heterogeneity.

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.

Referee Report

1 major / 2 minor

Summary. The paper proposes the MFM-Wishart model, which combines Wishart mixture components with a mixture-of-finite-mixtures (MFM) prior for Bayesian clustering of covariance matrices. It claims to establish posterior consistency for the number of clusters and posterior contraction of the mixing measure under standard regularity conditions after studying Wishart kernel properties, develops an MCMC algorithm for inference, reports competitive simulation performance even under misspecification, and applies the model to cluster infants using fNIRS-derived functional connectivity matrices.

Significance. If the consistency results hold after explicit verification, the work provides a useful Bayesian tool for clustering matrix-valued data with automatic inference on cluster number, relevant to neuroimaging and diffusion tensor imaging. The MCMC development and real-data application to brain connectivity illustrate practical value, and the simulations offer some evidence of robustness.

major comments (1)
  1. [Theoretical results] Theoretical section (referenced in abstract): The central claim of posterior consistency for K and contraction of the mixing measure rests on the Wishart kernel satisfying the technical conditions (KL support, identifiability, tail behavior) of the referenced general MFM theorems. The manuscript invokes 'standard regularity conditions' but provides no explicit verification or additional arguments addressing the non-Euclidean geometry of the positive-definite cone or boundary behavior, which is load-bearing for whether the theorems apply directly to this kernel.
minor comments (2)
  1. [Simulation studies] Simulation studies: Competitive performance and accurate K recovery are reported, but the section lacks details on the number of Monte Carlo replications, standard errors, or variability measures across runs, which would allow better assessment of the claims.
  2. [Application] Application section: The fNIRS analysis illustrates interpretable heterogeneity, but additional details on preprocessing of the covariance estimates, sensitivity to hyperparameter choices, or comparison to alternative clustering methods on the same data would strengthen the practical utility demonstration.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the constructive feedback on our manuscript. We address the single major comment below.

read point-by-point responses

  1. Referee: [Theoretical results] Theoretical section (referenced in abstract): The central claim of posterior consistency for K and contraction of the mixing measure rests on the Wishart kernel satisfying the technical conditions (KL support, identifiability, tail behavior) of the referenced general MFM theorems. The manuscript invokes 'standard regularity conditions' but provides no explicit verification or additional arguments addressing the non-Euclidean geometry of the positive-definite cone or boundary behavior, which is load-bearing for whether the theorems apply directly to this kernel.

    Authors: We agree that the manuscript would benefit from an explicit verification of the technical conditions for the Wishart kernel. Although the paper studies Wishart kernel properties in the context of mixture models, it does not contain a dedicated verification addressing the geometry of the positive definite cone or boundary behavior. In the revised manuscript we will add a new appendix that explicitly checks the KL support, identifiability, and tail-behavior conditions of the referenced MFM theorems, adapting the arguments to the manifold of positive definite matrices. revision: yes

Circularity Check

0 steps flagged

No circularity; consistency claims rest on explicit study of Wishart kernel properties

full rationale

The abstract states the authors 'study the properties of Wishart kernels in the context of mixture models and then establish results for posterior consistency... under standard regularity conditions.' No quoted equations or steps reduce a claimed prediction to a fitted input by construction, nor does any load-bearing premise collapse to a self-citation whose verification is internal to the paper. The derivation is presented as building on general MFM theorems after kernel-specific checks, rendering it self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review supplies no identifiable free parameters, axioms, or invented entities; full text required for ledger.

pith-pipeline@v0.9.0 · 5711 in / 946 out tokens · 18411 ms · 2026-05-25T04:12:14.042956+00:00 · methodology

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

58 extracted references · 58 canonical work pages

  1. [1]

    and Morgan, Victoria L

    Rogers, Baxter P. and Morgan, Victoria L. and Newton, Allen T. and Gore, John C. , title =. Magnetic Resonance Imaging , volume =

    work page

  2. [2]

    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. , title =. Cerebral Cortex , volume =

    work page

  3. [3]

    and Li, Yimei and Hall, Carol and Lin, Weili , title =

    Zhu, Hongtu and Chen, Yasheng and Ibrahim, Joseph G. and Li, Yimei and Hall, Carol and Lin, Weili , title =. Journal of the American Statistical Association , volume =

    work page

  4. [4]

    Computer Vision -- ECCV 2006 , series =

    Tuzel, Oncel and Porikli, Fatih and Meer, Peter , title =. Computer Vision -- ECCV 2006 , series =

    work page 2006

  5. [5]

    and Shephard, Neil , title =

    Barndorff-Nielsen, Ole E. and Shephard, Neil , title =. Econometrica , volume =

    work page

  6. [6]

    Computers in Biology and Medicine , volume =

    Sakkalis, Vangelis , title =. Computers in Biology and Medicine , volume =

    work page

  7. [7]

    The Neuroscientist , volume =

    Niu, Haijing and He, Yong , title =. The Neuroscientist , volume =

    work page

  8. [8]

    and Brier, Matthew R

    Strain, Jeremy F. and Brier, Matthew R. and Tanenbaum, Aaron and Gordon, Brian A. and McCarthy, John E. and Dincer, Aylin and Marcus, Daniel S. and Chhatwal, Jasmeer P. and Graff-Radford, Neill R. and Day, Gregory S. and. Covariance-based vs. correlation-based functional connectivity dissociates healthy aging from Alzheimer disease , journal =

    work page

  9. [9]

    and Miller, Karla L

    Smith, Stephen M. and Miller, Karla L. and Salimi-Khorshidi, Gholamreza and Webster, Matthew and Beckmann, Christian F. and Nichols, Thomas E. and Ramsey, Joseph D. and Woolrich, Mark W. , title =. NeuroImage , volume =

    work page

  10. [10]

    Systematic Review of Functional MRI Applications for Psychiatric Disease Subtyping , journal =

    Miranda, Lucas and Paul, Riya and P. Systematic Review of Functional MRI Applications for Psychiatric Disease Subtyping , journal =

    work page

  11. [11]

    and Wu, W

    Zhang, Y. and Wu, W. and Toll, R. T. and others , title =. Nature Biomedical Engineering , volume =

    work page

  12. [12]

    and Koloydenko, Alexander and Zhou, Di , title =

    Dryden, Ian L. and Koloydenko, Alexander and Zhou, Di , title =. The Annals of Applied Statistics , volume =

    work page

  13. [13]

    and Pise, N

    Irani, J. and Pise, N. and Phatak, M. , title =. International Journal of Computer Applications , volume =

    work page

  14. [14]

    Wiley Interdisciplinary Reviews: Computational Statistics , volume =

    van de Velden, Michel and Iodice D'Enza, Angela and Markos, Angelos , title =. Wiley Interdisciplinary Reviews: Computational Statistics , volume =

    work page

  15. [15]

    Ikotun, A. M. and Ezugwu, A. E. and Abualigah, L. and Abuhaija, B. and Heming, J. , title =. Information Sciences , volume =

    work page

  16. [16]

    Wiley Interdisciplinary Reviews: Data Mining and Knowledge Discovery , volume =

    Murtagh, Fionn and Contreras, Pedro , title =. Wiley Interdisciplinary Reviews: Data Mining and Knowledge Discovery , volume =

    work page

  17. [17]

    Statistics and Computing , volume =

    von Luxburg, Ulrike , title =. Statistics and Computing , volume =

    work page

  18. [18]

    Studies in Neural Data Science , series =

    Cabassi, Andrea and Casa, Alberto and Fontana, Marco and Russo, Monica and Farcomeni, Alessio , title =. Studies in Neural Data Science , series =

    work page

  19. [19]

    Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) , year =

    Jayasumana, Sadeep and Hartley, Richard and Salzmann, Mathieu and Li, Hongdong and Harandi, Mehrtash , title =. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) , year =

    work page

  20. [20]

    Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) , year =

    Yin, Ming and Guo, Yi and Gao, Junbin and He, Zhaoshui and Xie, Shengli , title =. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) , year =

    work page

  21. [21]

    Computational Statistics & Data Analysis , volume =

    Bouveyron, Charles and Brunet-Saumard, Camille , title =. Computational Statistics & Data Analysis , volume =

    work page

  22. [22]

    and Murphy, Thomas B

    Gormley, Isobel C. and Murphy, Thomas B. and Raftery, Adrian E. , title =. Annual Review of Statistics and Its Application , volume =

    work page

  23. [23]

    and Lee, Sharon X

    McLachlan, Geoffrey J. and Lee, Sharon X. and Rathnayake, Suren I. , title =. Annual Review of Statistics and Its Application , volume =

    work page

  24. [24]

    Journal of the Royal Statistical Society: Series B (Statistical Methodology) , volume =

    Rousseau, Judith and Mengersen, Kerrie , title =. Journal of the Royal Statistical Society: Series B (Statistical Methodology) , volume =

    work page

  25. [25]

    , title =

    Nydick, Scott W. , title =. Electronic Journal of Statistics , volume =

    work page

  26. [26]

    Pattern Recognition Letters , volume =

    Hidot, Sullivan and Saint-Jean, Christophe , title =. Pattern Recognition Letters , volume =

    work page

  27. [27]

    Computational Statistics & Data Analysis , volume =

    Cappozzo, Andrea and Casa, Alessandro , title =. Computational Statistics & Data Analysis , volume =

    work page

  28. [28]

    IEEE Transactions on Pattern Analysis and Machine Intelligence , volume =

    Cherian, Anoop and Morellas, Vassilios and Papanikolopoulos, Nikolaos , title =. IEEE Transactions on Pattern Analysis and Machine Intelligence , volume =

    work page

  29. [29]

    Neural Networks , volume =

    Tokuda, Takashi and Yamashita, Okito and Yoshimoto, Jun , title =. Neural Networks , volume =

    work page

  30. [30]

    and Harrison, Matthew T

    Miller, Jeffrey W. and Harrison, Matthew T. , title =. Advances in Neural Information Processing Systems , volume =

    work page

  31. [31]

    and Harrison, Matthew T

    Miller, Jeffrey W. and Harrison, Matthew T. , title =. Journal of Machine Learning Research , volume =

    work page

  32. [32]

    Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences , volume =

    Wade, Sara , title =. Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences , volume =

    work page

  33. [33]

    and Harrison, Matthew T

    Miller, Jeffrey W. and Harrison, Matthew T. , title =. Journal of the American Statistical Association , volume =

    work page

  34. [34]

    Bernoulli , volume =

    Guha, Arindam and Ho, Nhat and Nguyen, XuanLong , title =. Bernoulli , volume =

    work page

  35. [35]

    and Hu, G

    Yin, F. and Hu, G. and Shen, W. , title =. Journal of Computational and Graphical Statistics , volume =

    work page

  36. [36]

    and Molnar, M

    Blanco, B. and Molnar, M. and Carreiras, M. and Caballero-Gaudes, C. , title =. Scientific Data , volume =

    work page

  37. [37]

    Neurophotonics , volume =

    Abdollahpour, Neda and Artan, Nabi Sertac , title =. Neurophotonics , volume =

    work page

  38. [38]

    Frontiers in Psychiatry , volume =

    Wang, Kai and Ji, Xiang and Li, Ting , title =. Frontiers in Psychiatry , volume =

    work page

  39. [39]

    Frontiers in Psychiatry , volume =

    Gao, Chenyang and Shu, Leijin and Li, Ting , title =. Frontiers in Psychiatry , volume =

    work page

  40. [40]

    , title =

    Neal, Radford M. , title =. The Annals of Statistics , volume =

    work page

  41. [41]

    , title =

    Dahl, David B. , title =. Bayesian Inference for Gene Expression and Proteomics , pages =

    work page

  42. [42]

    and Hu, Guanyu , title =

    Pan, Tianyu and Shen, Weining and Davis-Stober, Clintin P. and Hu, Guanyu , title =. British Journal of Mathematical and Statistical Psychology , volume =

    work page

  43. [43]

    The Annals of Applied Statistics , volume =

    Zhu, Bencong and Hu, Guanyu and Xu, Lin and Fan, Xiaodan and Li, Qiwei , title =. The Annals of Applied Statistics , volume =

    work page

  44. [44]

    , title =

    Neal, Radford M. , title =. Journal of Computational and Graphical Statistics , volume =

    work page

  45. [45]

    Maechler, Martin , title =

    work page

  46. [46]

    Di Lonardo Burr, S. M. and Pirazzoli, L. and Dopiera. Longitudinal assessments of functional near-infrared spectroscopy background functional connectivity in low- and middle-income infants during a social cognition task , journal =

    work page

  47. [47]

    and deKlerk, C

    Bulgarelli, C. and deKlerk, C. C. J. M. and Richards, J. E. and Southgate, V. and Hamilton, A. and Blasi, A. , title =. Human Brain Mapping , volume =

    work page

  48. [48]

    Frontiers in Neuroscience , volume =

    Seiler, Christian and Holmes, Susan , title =. Frontiers in Neuroscience , volume =

    work page

  49. [49]

    Nielsen, S. F. V. and Madsen, K. H. and Schmidt, M. N. and M. Modeling dynamic functional connectivity using a Wishart mixture model , booktitle =

    work page

  50. [50]

    Electronic Journal of Statistics , volume =

    Ho, Nhat and Nguyen, XuanLong , title =. Electronic Journal of Statistics , volume =

    work page

  51. [51]

    and Bandyopadhyay, Dipankar , title =

    Lan, Zhou and Reich, Brian J. and Bandyopadhyay, Dipankar , title =. Canadian Journal of Statistics , volume =

    work page

  52. [52]

    Partial correlation for functional brain interactivity investigation in functional MRI , journal =

    Marrelec, Guillaume and Krainik, Alexandre and Duffau, Hugues and P. Partial correlation for functional brain interactivity investigation in functional MRI , journal =

    work page

  53. [53]

    Regions, systems, and the brain: hierarchical measures of functional integration in fMRI , journal =

    Marrelec, Guillaume and Bellec, Pierre and Krainik, Alexandre and Duffau, Hugues and P. Regions, systems, and the brain: hierarchical measures of functional integration in fMRI , journal =

    work page

  54. [54]

    Encyclopedia of Machine Learning , publisher =

    Teh, Yee Whye , title =. Encyclopedia of Machine Learning , publisher =

    work page

  55. [55]

    Bayesian Nonparametric Data Analysis , publisher =

    M. Bayesian Nonparametric Data Analysis , publisher =

    work page

  56. [56]

    Frontiers in Psychiatry , volume =

    Clustering of Multiple Psychiatric Disorders Using Functional Connectivity in the Data-Driven Brain Subnetwork , author =. Frontiers in Psychiatry , volume =

    work page

  57. [57]

    Journal of Multivariate Analysis , volume =

    Lewandowski, Daniel and Kurowicka, Dorota and Joe, Harry , title =. Journal of Multivariate Analysis , volume =

    work page

  58. [58]

    Statistica Sinica , pages =

    Barnard, John and McCulloch, Robert and Meng, Xiao-Li , title =. Statistica Sinica , pages =

    work page