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

arXiv:2605.23207 · stat.ME · Submitted 2026-05-22 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Statistics. The day in statistics: methods, theory and the inference under them.

Tom: We're looking at the paper: "Mixture-of-Finite-Mixtures Wishart Model for Clustering Covariance Matrices with an Application to Brain Functional Connectivity".

Jane: Here's the gist of the paper. Mixture-of-Finite-Mixtures Wishart Model for Clustering Covariance Matrices with an Application to Brain Functional Connectivity develops a Bayesian model-based clustering approach that combines Wishart mixture components with a mixture-of-finite-mixtures (MFM) prior to jointly infer the number of clusters and their assignments for covariance and correlation matrices, which is relevant in fields like brain functional connectivity <ref:2605.23207#pg2> Model Formulation The paper proposes the MFM–Wishart model, which is a Wishart mixture model equipped with an MFM prior to enable joint posterior inference on the number of clusters and the clustering assignments <ref:2605.23207#pg5>.

Tom: This model places a prior directly on the number of mixture components by treating K as a random variable, unlike the Dirichlet process mixture (DPM) which assumes a countably infinite collection of components <ref:2605.23207#pg9>.

Jane: The model is specified in equation (2.6), which includes component-specific scale matrices Σk and degrees-of-freedom parameters νk <ref:2605.23207#pg10>.

Tom: A natural specialization involves imposing a shared degrees-of-freedom parameter across mixture components, ν1 = · · · = νK = ν, which is well suited for functional-connectivity studies.

Paper summary: Tom: So, wrapping up on the "Mixture-of-Finite-Mixtures Wishart Model for Clustering Covariance Matrices with an Application to Brain Functional Connectivity." The authors developed a Bayesian model-based clustering approach that mixes Wishart mixture components with a mixture-of-finite-mixtures prior to find both the number of clusters and their assignments

Allen et al., two thousand fourteen: <ref:2605.23207#pg1>.

Jane: It matters because it tackles the problem of clustering covariance matrices, which show second-order dependence structures in fields like brain functional connectivity, by providing a framework that allows for joint inference on the number of clusters and what those clusters actually are

Allen et al., two thousand fourteen: <ref:2605.23207#pg1>.

Lu: The key implication is that when you have data that truly comes from a finite mixture, this MFM approach gives a better estimate of how many components there are compared to the Dirichlet process mixture model

Allen et al., two thousand fourteen: <ref:2605.23207#pg1>.

Meng: For us deploying models, it means we can trust the number of clusters we get more when we know the underlying structure is finite

Allen et al., two thousand fourteen: <ref:2605.23207#pg1>.

Lalam: It also gives us a way to look at these brain connectivity matrices not just as static links, but as groupings with distinct patterns derived from a generative model

Allen et al., two thousand fourteen: <ref:2605.23207#pg1>.

Tom: So, the paper provides a competitive clustering method that recovers the true number of clusters more reliably than other methods in this context

Allen et al., two thousand fourteen: <ref:2605.23207#pg1>.

Conclusion: Tom: So we're wrapping up on this paper about clustering covariance matrices using a mixture model and they call it the Mixture-of-Finite-Mixtures Wishart Model

Allen et al., two thousand fourteen: .

Jane: It's about taking those complex brain connectivity matrices, which are essentially huge tables of how different brain regions talk to each other, and trying to group them into meaningful clusters.

Lu: The core idea is using a mixture model where you treat the number of clusters as a random thing you have to infer, instead of just picking one fixed number upfront.

Meng: So what's the big payoff here? They claim it recovers the true number of clusters better when the data actually comes from a finite set.

Tom: Right, that means if you're trying to find distinct functional patterns in brain activity, this model should give you a more reliable count of those patterns than some other methods we've seen.

Jane: It connects this to real science because they used it on infant sleep data from functional near-infrared spectroscopy, which is basically measuring blood flow in the brain.

Lu: They found these interpretable cluster-specific connectivity patterns in a specific region of channels, like channels forty through forty-six.

Tom: So for someone just listening to this show, the implication is that we can move past just seeing noise and start finding distinct functional groupings in brain networks with more confidence.

Jane: It's a step toward making sure the patterns we see in brain scans aren't just random fluctuations but represent something biologically real.

Lu: The authors suggest extending this model by adding some sparsity regularization to those cluster-specific scale matrices if you want to tackle even higher dimensions later on.

Meng: That makes sense for production because high dimensions mean a lot of computational cost, and sparsity helps keep things manageable when scaling up the analysis.

Tom: It sounds like a solid framework for getting better structure from messy brain data.

Jane: And now we have to look at how they actually ran the math behind it next.

Zongyu Li, Stefano Castruccio, Zhiyong Zhang

Department of Applied and Computational Mathematics and Statistics, University of Notre Dame · Department of Psychology, University of Notre Dame

stat.ME

Submitted: 2026-05-22

Updated: 2026-11-01

Code: https://github.com/Zongyu-Li/MFM-Wishart

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

Importance score: 79/100

The gist: Mixture-of-Finite-Mixtures Wishart Model for Clustering Covariance Matrices with an Application to Brain Functional Connectivity develops a Bayesian model-based clustering approach that combines…

Key concepts

Wishart Mixture Model
This is a statistical model used to describe covariance matrices by assuming they are a mixture of several different Wishart distributions. A Wishart distribution is a probability distribution over symmetric positive definite matrices, which are the mathematical objects representing the relationships between variables in data.
Mixture-of-Finite-Mixtures (MFM) Prior
The MFM prior is a specific way to model the number of mixture components. Unlike some other methods, this prior directly treats the number of clusters as a random variable, allowing the model to jointly determine both how many groups there are and which data points belong to which group.
Functional Connectivity
This refers to the statistical measure that describes the patterns of synchronization or coordination between different regions in the brain. In this study, it was used on near-infrared spectroscopy (fNIRS) data from infants to identify distinct functional patterns during sleep.

Terminology

Summary

Mixture-of-Finite-Mixtures Wishart Model for Clustering Covariance Matrices with an Application to Brain Functional Connectivity develops a Bayesian model-based clustering approach that combines Wishart mixture components with a mixture-of-finite-mixtures (MFM) prior to jointly infer the number of clusters and their assignments for covariance and correlation matrices, which is relevant in fields like brain functional connectivity <ref:2605. p. 023207#pg2>

Model Formulation

The paper proposes the MFM–Wishart model, which is a Wishart mixture model equipped with an MFM prior to enable joint posterior inference on the number of clusters and the clustering assignments <ref:2605.23207#pg5>. This model places a prior directly on the number of mixture components by treating K as a random variable, unlike the Dirichlet process mixture (DPM) which assumes a countably infinite collection of components <ref:2605. p. 323207#pg9>. The model is specified in equation (2. p. 26), which includes component-specific scale matrices Σk and degrees-of-freedom parameters νk <ref:2605. p. 723207#pg10>. A natural specialization involves imposing a shared degrees-of-freedom parameter across mixture components, ν1 = · · · = νK = ν, which is well suited for functional-connectivity studies. p. 7

Theoretical Guarantees

The theoretical analysis establishes large-sample theory for posterior consistency of the number of mixture components and posterior contraction of the mixing measure for the MFM–Wishart model under standard regularity conditions <ref:2605.23207#pg5>. Theorem 1 establishes that, as n → ∞, the posterior probability assigned to the true number of mixture components converges to one almost surely under PG0. p. 3 Furthermore, it shows that the posterior distribution over mixing measures contracts around the true mixing measure G0 with respect to the order-1 Wasserstein distance at a rate of (log n/n) 1/2. p. 8

Inference Algorithm

An efficient Markov chain Monte Carlo (MCMC) algorithm is developed for posterior inference, which consists of Gibbs updates of the cluster labels and a random-walk Metropolis–Hastings step for updating the shared ν. The full conditional distribution of the cluster label zi given z−i, ν, and the data is given in Proposition 2. p. 6 The update for the shared parameter ν is handled via a random-walk Metropolis–Hastings step targeting its collapsed full conditional posterior distribution.

Application and Performance

The method is applied 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 <ref:2605.23207#pg2>. Simulation studies show competitive clustering performance and accurate recovery of the number of clusters, even under model misspecification. p. 0 In the application, MFM–Wishart identified interpretable cluster-specific connectivity patterns in the region of channels 40–46. p. 3

Conclusion and Future Work

The MFM–Wishart model provides competitive clustering performance across various settings, yielding more reliable recovery of the true number of clusters than DPM–Wishart when data are generated from a finite mixture <ref:2605.23207#pg2>. The paper suggests that extending the model to incorporate explicit sparsity regularization in the cluster-specific scale matrices may be beneficial for higher-dimensional settings. p. 2 Future work could also explore incorporating spatial dependence into the model, as considered in Lan et al. p. 3 [2021]. The code used to implement the MFM–Wishart model is available for reproduction. p. 0

References

Elena A. Allen, Eswar Damaraju, Sergey M. p. 1 Plis, Erik B. Erhardt, Tom Eichele, and Vince D. Calhoun Tracking whole-brain connectivity dynamics in the resting state Cerebral Cortex 24(3):663–676 2014 <ref:2605. p. 123207#pg10>

Ole E. Barndorff-Nielsen and Neil Shephard Econometric analysis of realized covariation: High frequency based covariance, regression, and correlation in financial economics Econometrica 72(3):885–925 2004

B. Blanco, M. Molnar, M. Carreiras, and C. Caballero-Gaudes Open access dataset of taskfree hemodynamic activity in 4-month-old infants during sleep using fnirs Scientific Data 9(1):102 2022

Charles Bouveyron and Camille Brunet-Saumard Model-based clustering of highdimensional data: A review Computational Statistics & Data Analysis 71:52–78 2014

C. Bulgarelli, C. C. J. M. deKlerk, J. E. Richards, V. Southgate, A Hamilton, and A Blasi The developmental trajectory of fronto-temporoparietal connectivity as a proxy of the default mode network Human Brain Mapping 41:2717–2740 2020

Andrea Cappozzo and Alessandro Casa Model-based clustering for covariance matrices via penalized wishart mixture models Computational Statistics & Data Analysis 212:108232 2025

Anoop Cherian, Vassilios Morellas, and Nikolaos Papanikolopoulos Bayesian nonparametric clustering for positive definite matrices IEEE Transactions on Pattern Analysis and Machine Intelligence 38(5):862–874 2015

David B. Dahl Model-based clustering for expression data via a dirichlet process mixture model In Bayesian Inference for Gene Expression and Proteomics pages 201–218 2006

S. M. Di Lonardo Burr, L. Pirazzoli, A. W Dopiera la, V R Bejjanki, C A Nelson, and L L Emberson Longitudinal assessments of functional near-infrared spectroscopy background functional connectivity in low- and middle-income infants during a social cognition task Journal of Experimental Psychology: General 153(3):798–813 2024

Ian L Dryden, Alexander Koloydenko, and Di Zhou Non-euclidean statistics for covariance matrices with applications to diffusion tensor imaging The Annals of Applied Statistics 3(3):1102–1123 2009

Isobel C. Gormley, Thomas B. p. 2 Murphy, and Adrian E. Raftery Model-based clustering Annual Review of Statistics and Its Application 10(1):573–595 2023

Arindam Guha, Nhat Ho, and XuanLong Nguyen On posterior contraction of parameters and interpretability in bayesian mixture modeling Bernoulli 27(4):2159–2188 2021

Sullivan Hidot and Christophe Saint-Jean An expectation–maximization algorithm for the wishart mixture model Application to movement clustering Pattern Recognition Letters 31(14):2318–2324 2010

Nhat Ho and XuanLong Nguyen On strong identifiability and convergence rates of parameter estimation in finite mixtures Electronic Journal of Statistics 10(1):271–307 2016

A. M. Ikotun, A E Ezugwu, L Abualigah, B Abuhaija, and J Heming K-means clustering algorithms A comprehensive review variants analysis and advances in the era of big data Information Sciences 622:178–210 2023

J Irani, N Pise, and M Phatak Clustering techniques and the similarity measures used in clustering A survey International Journal of Computer Applications 134(7):9–14 2016

Sadeep Jayasumana, Richard Hartley, Mathieu Salzmann, Hongdong Li, and Mehrtash Harandi Kernel methods on the riemannian manifold of symmetric positive definite matrices In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition CVPR 2013

Zhou Lan, Brian J. Reich, and Dipankar Bandyopadhyay A spatial bayesian semiparametric mixture model for positive definite matrices with applications in diffusion tensor imaging Canadian Journal of Statistics 49(1):129–149 2021

Daniel Lewandowski, Dorota Kurowicka, and Harry Joe Generating random correlation matrices based on vines and extended onion method Journal of Multivariate Analysis 100(9):1989–2001 2009

Martin Maechler Finding Groups in Data Cluster Analysis Extended Rousseeuw et al. 2019 R package version 2. p. 20.0 Information Sciences 622:178–210 2023

Guillaume Marrelec, Alexandre Krainik, Hugues Duffau, M´elanie P´el´egrini-Issac Stéphane Leh´ericy Julien Doyon and Habib Benali Partial correlation for functional brain interactivity investigation in functional mri NeuroImage 32(1):228–237 2006

Guillaume Marrelec, Pierre Bellec, Alexandre Krainik, Hugues Duffau, M´elanie P´el´egriniIssac Stéphane Leh´ericy Habib Benali and Julien Doyon Regions systems and the brain hierarchical measures of functional integration in fmri Medical Image Analysis 12(4):484–496 2008

Geoffrey J. McLachlan, Sharon X. p. 2 Lee, and Suren I Rathnayake Finite mixture models Annual Review of Statistics and Its Application 6(1):355–378 2019

Jeffrey W. Miller and Matthew T Harrison A simple example of dirichlet process mixture inconsistency for the number of components In Advances in Neural Information Processing Systems volume 26 2013

Jeffrey W. Miller and Matthew T Harrison Inconsistency of pitman–yor process mixtures for the number of components Journal of Machine Learning Research 15(1):3333–3370 2014

Jeffrey W. Miller and Matthew T Harrison Mixture models with a prior on the number of components Journal of the American Statistical Association 113(521):340–356 2018

Lucas Miranda, Riya Paul, Benno Putz, Nikolaos Koutsouleris, and Bertram Muller-Myhsok Systematic review of functional mri applications for psychiatric disease subtyping Frontiers in Psychiatry 12 2021

Peter Muller, Fernando Andr´es Quintana, Alejandro Jara, and Tim Hanson Bayesian Nonparametric Data Analysis Springer New York 2015

Fionn Murtagh and Pedro Contreras Algorithms for hierarchical clustering an overview Wiley Interdisciplinary Reviews Data Mining and Knowledge Discovery 2(1):86–97 2012

Radford M Neal Markov chain sampling methods for dirichlet process mixture models Journal of Computational and Graphical Statistics 9(2):249–265 2000

S. F. V Nielsen, K H Madsen, M N Schmidt, and M Mørup Modeling dynamic functional connectivity using a wishart mixture model In 2017 International Workshop on Pattern Recognition in Neuroimaging PRNI pages 1–4 Toronto ON Canada 2017

Haijing Niu and Yong He Resting-state functional brain connectivity lessons from functional near-infrared spectroscopy The Neuroscientist 20(2):173–188 2014

Scott W. Nydick The wishart and inverse wishart distributions Electronic Journal of Statistics 6:1–19 2012

Yee Whye Teh Dirichlet process In Encyclopedia of Machine Learning pages 280–287 Springer 2011

Takashi Tokuda, Okito Yamashita, and Jun Yoshimoto Multiple clustering for identifying subject clusters and brain sub-networks using functional connectivity matrices without vectorization Neural Networks 142:269–287 2021a

Tomoki Tokuda, Okito Yamashita, Yuki Sakai, and Junichiro Yoshimoto Clustering of multiple psychiatric disorders using functional connectivity in the data-driven brain subnetwork Frontiers in Psychiatry 12:683280 2021b

Oncel Tuzel, Fatih Porikli, and Peter Meer Region covariance A fast descriptor for detection and classification In Computer Vision ECCV 2006 volume 3952 of Lecture Notes in Computer Science Berlin Heidelberg 2006

Michel van de Velden, Angela Iodice D’Enza, and Angelos Markos Distance-based clustering of mixed data Wiley Interdisciplinary Reviews Computational Statistics 11(3):e1456 2019

Ulrike von Luxburg A tutorial on spectral clustering Statistics and Computing 17(4):395–416 2007

Sara Wade Bayesian cluster analysis Philosophical Transactions of the Royal Society A Mathematical, Physical and Engineering Sciences 381(2247):20220149 2023

Kai Wang, Xiang Ji, and Ting Li Gender difference in functional activity of 4-months-old infants during sleep A functional near-infrared spectroscopy study Frontiers in Psychiatry 13 2023

F. Yin, G Hu, and W Shen Analysis of professional basketball field goal attempts via a bayesian matrix clustering approach Journal of Computational and Graphical Statistics 32(1):49–60 2023

Ming Yin, Yi Guo, Junbin Gao, Zhaoshui He, and Shengli Xie Kernel sparse subspace clustering on symmetric positive definite manifolds In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition CVPR 2016

Y.

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Enhanced clustering for covariance-type data: The MFM-Wishart model enables joint posterior inference on both clusters and assignments, allowing AI systems to identify interpretable heterogeneity in complex brain functional connectivity data during sleep by estimating the number of clusters and their specific connectivity patterns.

  2. Robust cluster count estimation: The MFM prior allows for consistent estimation of the number of clusters even when the true data-generating mixture has finitely many components, which can be crucial for correctly identifying subject subtypes or cognitive phenotypes in brain imaging analyses.

  3. Efficient posterior inference: A developed Markov chain Monte Carlo algorithm exploits Wishart inverse-Wishart conjugacy and the MFM prior to efficiently sample from the joint posterior distribution of parameters, enabling rapid computation of clustering assignments and model parameters.

  4. Robustness to model misspecification: The method demonstrates competitive performance in simulated settings where the observed matrices are not exactly Wishart (e.g., due to temporal dependence), suggesting that the MFM-Wishart framework is robust for real-world data like fNIRS functional connectivity from infants during sleep.

  5. Interpretable cluster characterization: For brain connectivity data, the system can report cluster-specific summaries by applying Dahl's method to MCMC samples, providing a representative partition and characterizing latent heterogeneity, such as identifying a cluster characterized by systematically stronger correlations among channels 40–46.

Related papers