Fast estimation of Gaussian mixture components via centering and singular value thresholding

Huan Qing

arxiv: 2604.19091 · v1 · submitted 2026-04-21 · 📊 stat.ML · cs.LG

Fast estimation of Gaussian mixture components via centering and singular value thresholding

Huan Qing This is my paper

Pith reviewed 2026-05-10 01:58 UTC · model grok-4.3

classification 📊 stat.ML cs.LG

keywords Gaussian mixture modelsnumber of components estimationsingular value thresholdinghigh-dimensional clusteringunsupervised learningconsistent estimationimbalanced data

0 comments

The pith

Centering the data matrix and thresholding its singular values consistently recovers the true number of Gaussian mixture components.

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

The paper introduces a direct estimator for the number of components in Gaussian mixture models: subtract the sample mean from each observation, form the centered data matrix, compute its singular values, and count how many exceed a fixed threshold. This procedure requires no iterative optimization, no likelihood evaluation, and no advance knowledge of the component count. The authors prove that the count matches the true number with high probability whenever the component centers obey a mild separation condition, even when the ambient dimension exceeds the number of observations and when the number of components itself grows with the data size. The same guarantee continues to hold under severe imbalance in the sizes of the individual components. Because the method reduces to a single matrix decomposition, it runs in near-linear time on large datasets.

Core claim

Under a mild separation condition on the component centers, centering the data matrix and counting its singular values above a threshold yields a consistent estimator of the true number of components K. The result continues to hold in regimes where the dimension p may greatly exceed the sample size n, where K itself may grow as fast as min(p, n), and where the component weights may be arbitrarily unbalanced.

What carries the argument

Singular-value thresholding applied to the centered data matrix, which isolates the low-rank signal contributed by the distinct component centers from the noise.

If this is right

The estimator remains consistent when the number of components grows with the data size.
No iterative fitting or likelihood computation is required.
The procedure scales to at least ten million observations in a few hundred dimensions.
Severe imbalance among component sizes does not invalidate the consistency guarantee.

Where Pith is reading between the lines

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

The same centering-plus-thresholding idea could be tested on mixtures whose component distributions are non-Gaussian but still possess a low-rank mean structure.
Because the method relies only on a single SVD, it is immediately compatible with streaming or distributed implementations for very large data.
Connections to random-matrix analyses of spiked covariance models may yield sharper finite-sample bounds than those given in the paper.

Load-bearing premise

The centers of the different Gaussian components must be sufficiently far apart from one another.

What would settle it

Generate data from a two-component Gaussian mixture whose centers lie closer together than the separation threshold used in the proof, then verify whether the estimator returns one component instead of two.

read the original abstract

Estimating the number of components is a fundamental challenge in unsupervised learning, particularly when dealing with high-dimensional data with many components or severely imbalanced component sizes. This paper addresses this challenge for classical Gaussian mixture models. The proposed estimator is simple: center the data, compute the singular values of the centered matrix, and count those above a threshold. No iterative fitting, no likelihood calculation, and no prior knowledge of the number of components are required. We prove that, under a mild separation condition on the component centers, the estimator consistently recovers the true number of components. The result holds in high-dimensional settings where the dimension can be much larger than the sample size. It also holds when the number of components grows to the smaller of the dimension and the sample size, even under severe imbalance among component sizes. Computationally, the method is extremely fast: for example, it processes ten million samples in one hundred dimensions within one minute. Extensive experimental studies confirm its accuracy in challenging settings such as high dimensionality, many components, and severe class imbalance.

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

Simple SVD thresholding on centered data gives a fast estimator for the number of GMM components with consistency claims in high dimensions and growing K, but the separation condition likely strengthens with small component weights and undercuts the severe-imbalance claim.

read the letter

This paper's main contribution is a non-iterative estimator for the number of components in a Gaussian mixture: center the data matrix and threshold its singular values. It proves consistency under a separation condition on the means, and claims this works even when the dimension exceeds the sample size, when the number of components grows, and when the mixture is severely imbalanced. The approach is new in its simplicity. Most methods for choosing K in GMMs rely on EM iterations or penalized likelihood. Here, everything reduces to one centering step and an SVD, which explains the speed. Processing ten million points in a hundred dimensions in a minute is a real practical plus. The experiments apparently test the hard regimes the abstract mentions, so that part looks solid. The potential issue is with the separation condition. The abstract calls it mild and says the result holds under severe imbalance. But the signal from a small component in the second-moment matrix is scaled by the square root of its weight. For the corresponding singular value to stand out from the noise bulk, the mean separation for that component likely has to be larger when its pi is tiny. If the theorem requires something like separation growing with 1 over sqrt of the smallest weight, then the condition is not mild for imbalanced cases, and the claim needs more qualification. I would want to see the exact statement in the theorem to check this. The rest of the paper seems to use standard tools from random matrix theory for the proof, which is fine if the assumptions are stated clearly. No obvious circularity or invented parameters. This is worth a serious look for anyone building fast unsupervised methods in high dimensions. A reader who needs something quick and doesn't want to tune many parameters could get value from it, provided they verify the separation for their data. It deserves peer review because the computational advantage is clear and the consistency result, if the conditions check out, would be useful. I recommend sending it to referees with a note to pay close attention to how the separation condition behaves under imbalance.

Referee Report

2 major / 1 minor

Summary. The paper proposes a simple, non-iterative estimator for the number of components K in a Gaussian mixture model: center the n×p data matrix, compute its singular values, and count how many exceed a threshold. The central claim is a consistency proof showing that this estimator recovers the true K with high probability under a mild separation condition on the component centers. The result is stated to hold in high-dimensional regimes (p ≫ n), when K grows up to min(p,n), and even under severe imbalance in component sizes π_i, with supporting experiments on large-scale and challenging instances.

Significance. If the consistency theorem holds with the claimed generality, the contribution is substantial: it supplies a computationally cheap (single SVD) alternative to likelihood-based selection methods that scale poorly in high dimensions or with large/imbalanced K. The high-dimensional and growing-K asymptotics, together with the explicit handling of imbalance without iterative fitting, address practically relevant gaps in unsupervised learning. The absence of free parameters in the estimator and the provision of reproducible experiments are additional strengths.

major comments (2)

[Theorem 4.1] Theorem 4.1 (or the main consistency result in §4): the abstract asserts consistency under a 'mild separation condition' that continues to hold 'even under severe imbalance among component sizes.' However, the population second-moment matrix is Σ + M M^T where the columns of M are scaled by sqrt(π_i). Consequently the (K-1) signal singular values of the centered matrix scale as sqrt(n π_i)·||μ_i - μ̄||. For these to exceed the Marchenko-Pastur bulk edge, the minimal separation must compensate for the smallest π_i. If the stated condition in the theorem contains an implicit 1/sqrt(π_min) factor (or equivalent), the separation is no longer mild when min π_i → 0, contradicting the abstract claim. The precise form of the separation assumption must be displayed and shown to be independent of min π_i.
[§3] §3 (population analysis) and the proof of Theorem 4.1: the derivation of the threshold and the control of the noise singular values under p,n→∞ with K growing must be checked for hidden dependence on the minimal component weight. If the proof relies on a uniform lower bound away from zero for all π_i, the 'severe imbalance' statement cannot be maintained without additional restrictions.

minor comments (1)

[Abstract] The abstract and introduction should explicitly reference the theorem number containing the separation condition so readers can immediately locate the precise assumption.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We appreciate the referee's thorough review and valuable feedback on our manuscript. We address each of the major comments in detail below.

read point-by-point responses

Referee: [Theorem 4.1] Theorem 4.1 (or the main consistency result in §4): the abstract asserts consistency under a 'mild separation condition' that continues to hold 'even under severe imbalance among component sizes.' However, the population second-moment matrix is Σ + M M^T where the columns of M are scaled by sqrt(π_i). Consequently the (K-1) signal singular values of the centered matrix scale as sqrt(n π_i)·||μ_i - μ̄||. For these to exceed the Marchenko-Pastur bulk edge, the minimal separation must compensate for the smallest π_i. If the stated condition in the theorem contains an implicit 1/sqrt(π_min) factor (or equivalent), the separation is no longer mild when min π_i → 0, contradicting the abstract claim. The precise form of the separation assumption must be displayed and shown to be independent of min π_i.

Authors: We thank the referee for this important observation. The separation condition in Theorem 4.1 is formulated as a lower bound on the minimal distance between component centers, without explicit dependence on the mixture weights π_i. However, as correctly noted, the singular values associated with each component scale with sqrt(π_i), meaning that the condition effectively requires stronger separation for components with smaller weights to ensure their signal exceeds the noise bulk. We will display the precise form of the assumption in the revised manuscript and include a discussion clarifying that while the condition itself does not depend on π_min, the regime of 'severe imbalance' is limited by the need for sufficient separation for the smallest components. This maintains the consistency result as stated, but we will adjust the abstract to avoid any potential misinterpretation of 'mild'. revision: partial
Referee: §3 (population analysis) and the proof of Theorem 4.1: the derivation of the threshold and the control of the noise singular values under p,n→∞ with K growing must be checked for hidden dependence on the minimal component weight. If the proof relies on a uniform lower bound away from zero for all π_i, the 'severe imbalance' statement cannot be maintained without additional restrictions.

Authors: The population analysis in §3 derives the eigenvalues of the second-moment matrix, which explicitly depend on the π_i through the rank-K perturbation. The threshold is set based on the Marchenko-Pastur edge for the noise covariance, independent of π. The proof of Theorem 4.1 uses Weyl's inequality or perturbation bounds to separate signal and noise singular values, with the noise control relying on random matrix theory results that hold as long as K = o(min(n,p)) or as per the assumptions, without requiring π_i bounded away from zero. We confirm that no uniform lower bound on π_i is used, allowing for severe imbalance. We will add a clarifying remark in the revised version. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The estimator is defined explicitly as centering the observed data matrix followed by singular-value thresholding and counting. The claimed consistency is a theorem proved under an external separation condition on the component means; the proof does not reduce any quantity to a fitted parameter from the same data, nor does it rely on self-definitional loops, load-bearing self-citations, or imported uniqueness results. The derivation chain remains a standard non-circular consistency argument for a direct spectral procedure.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on the standard assumption that the data follows a Gaussian mixture model together with a separation condition on the component centers. No free parameters or invented entities are mentioned in the abstract.

axioms (1)

domain assumption Data is generated from a Gaussian mixture model whose component centers satisfy a mild separation condition.
This condition is explicitly invoked for the consistency guarantee stated in the abstract.

pith-pipeline@v0.9.0 · 5470 in / 1228 out tokens · 47720 ms · 2026-05-10T01:58:55.451012+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

32 extracted references · 1 canonical work pages

[1]

ACM Co mputing Surveys (CSUR) 31(3), 264–323 (1999)

Jain, A.K., Murty, M.N., Flynn, P.J.: Data clustering: a review. ACM Co mputing Surveys (CSUR) 31(3), 264–323 (1999)

1999
[2]

Pattern R ecognition Letters 31(8), 651–666 (2010)

Jain, A.K.: Data clustering: 50 years beyond k-means. Pattern R ecognition Letters 31(8), 651–666 (2010)

2010
[3]

Annual Review of Statistics and Its Application 10(1), 573–595 (2023)

Gormley, I.C., Murphy, T.B., Raftery, A.E.: Model-based clustering . Annual Review of Statistics and Its Application 10(1), 573–595 (2023)

2023
[4]

contributions to the mathematical theory of ev olution

Pearson, K.: Iii. contributions to the mathematical theory of ev olution. Proceed- ings of the Royal Society of London 54(326-330), 329–333 (1894)
[5]

Lindsay, B.G.: Mixture models: theory, geometry, and application s. (1995). Ims

1995
[6]

McLachlan, G.J., Peel, D.: Finite mixture models (2000)

2000
[7]

Statistics Surveys 12(none), 18–65 (2018)

Fop, M., Murphy, T.B.: Variable selection methods for model-based clustering. Statistics Surveys 12(none), 18–65 (2018)

2018
[8]

Journal of the Royal Statistical Society : Series B (Methodological) 39(1), 1–22 (1977) 26

Dempster, A.P., Laird, N.M., Rubin, D.B.: Maximum likelihood from incomp lete data via the EM algorithm. Journal of the Royal Statistical Society : Series B (Methodological) 39(1), 1–22 (1977) 26

1977
[9]

arXiv preprint arXiv:1612.02099 (2016)

Lu, Y., Zhou, H.H.: Statistical and computational guarantees of lloyd’s algorithm and its variants. arXiv preprint arXiv:1612.02099 (2016)

work page arXiv 2016
[10]

Annals of Statistics 49(5), 2506–2530 (2021)

L¨ oﬄer, M., Zhang, A.Y., Zhou, H.H.: Optimality of spectral cluste ring in the gaussian mixture model. Annals of Statistics 49(5), 2506–2530 (2021)

2021
[11]

IEEE Transactions on Information Theory 67(6), 4223–4238 (2021)

Chen, X., Yang, Y.: Cutoﬀ for exact recovery of gaussian mixtu re models. IEEE Transactions on Information Theory 67(6), 4223–4238 (2021)

2021
[12]

Annals of Statistics 50(4), 2096–2126 (2022)

Ndaoud, M.: Sharp optimal recovery in the two component gaus sian mixture model. Annals of Statistics 50(4), 2096–2126 (2022)

2096
[13]

European Journal of Applied Mathematics 36(3), 491–523 (2025)

Li, Z.: Exact recovery of community detection in k-community ga ussian mixture models. European Journal of Applied Mathematics 36(3), 491–523 (2025)

2025
[14]

Advances in Neural In formation Processing Systems 37, 113698–113741 (2024)

Chen, X., Zhang, A.Y.: Achieving optimal clustering in gaussian mixt ure mod- els with anisotropic covariance structures. Advances in Neural In formation Processing Systems 37, 113698–113741 (2024)

2024
[15]

IEE E Transactions on Automatic Control 19(6), 716–723 (2003)

Akaike, H.: A new look at the statistical model identiﬁcation. IEE E Transactions on Automatic Control 19(6), 716–723 (2003)

2003
[16]

Annals of Statis tics, 461–464 (1978)

Schwarz, G.: Estimating the dimension of a model. Annals of Statis tics, 461–464 (1978)

1978
[17]

Journal of the royal statistical so ciety: series b (statistical methodology) 63(2), 411–423 (2001)

Tibshirani, R., Walther, G., Hastie, T.: Estimating the number of clu sters in a data set via the gap statistic. Journal of the royal statistical so ciety: series b (statistical methodology) 63(2), 411–423 (2001)

2001
[18]

: X-means: Extending k-means with eﬃcient esti- mation of the number of clusters

Pelleg, D., Moore, A.W., et al. : X-means: Extending k-means with eﬃcient esti- mation of the number of clusters. In: ICML, vol. 1, pp. 727–734 (2 000). Stanford, CA
[19]

Advances in Neu ral Informa- tion Processing Systems 16 (2003)

Hamerly, G., Elkan, C.: Learning the k in k-means. Advances in Neu ral Informa- tion Processing Systems 16 (2003)

2003
[20]

Advances in Neural Information Processing Systems 19 (2006)

Feng, Y., Hamerly, G.: Pg-means: learning the number of cluster s in data. Advances in Neural Information Processing Systems 19 (2006)

2006
[21]

Computational Sta tistics & Data Analysis 56(6), 1381–1395 (2012)

Melnykov, V., Melnykov, I.: Initializing the em algorithm in gaussian m ixture models with an unknown number of components. Computational Sta tistics & Data Analysis 56(6), 1381–1395 (2012)

2012
[22]

In: Proceedings Eighth International Confer ence on Artiﬁcial Intelligence and Statistics, pp

Corduneanu, A., Bishop, C.M.: Variational bayesian model select ion for mix- ture distributions. In: Proceedings Eighth International Confer ence on Artiﬁcial Intelligence and Statistics, pp. 27–34 (2001). Morgan Kaufmann 27

2001
[23]

IEEE Transactions on Pattern An alysis and Machine Intelligence 28(6), 1013–1018 (2006)

Constantinopoulos, C., Titsias, M.K., Likas, A.: Bayesian feature and model selec- tion for gaussian mixture models. IEEE Transactions on Pattern An alysis and Machine Intelligence 28(6), 1013–1018 (2006)

2006
[24]

Bishop, C.M., Nasrabadi, N.M.: Pattern recognition and machine lea rning 4(4) (2006)

2006
[25]

Statistics and Computing 26(1), 303–324 (2016)

Malsiner-Walli, G., Fr¨ uhwirth-Schnatter, S., Gr¨ un, B.: Model-b ased clustering based on sparse ﬁnite gaussian mixtures. Statistics and Computing 26(1), 303–324 (2016)

2016
[26]

IEEE Transactions on Signal Pro cessing 66(20), 5392–5406 (2018)

Teklehaymanot, F.K., Muma, M., Zoubir, A.M.: Bayesian cluster enu meration cri- terion for unsupervised learning. IEEE Transactions on Signal Pro cessing 66(20), 5392–5406 (2018)

2018
[27]

Statistica Sinica, 147–169 (2017)

Huang, T., Peng, H., Zhang, K.: Model selection for gaussian mixt ure models. Statistica Sinica, 147–169 (2017)

2017
[28]

Jour nal of Econo- metrics 249, 105958 (2025)

Budanova, S.: Penalized estimation of ﬁnite mixture models. Jour nal of Econo- metrics 249, 105958 (2025)

2025
[29]

Annals of Eugenics 7(2), 179–188 (1936)

Fisher, R.A.: The use of multiple measurements in taxonomic proble ms. Annals of Eugenics 7(2), 179–188 (1936)

1936
[30]

Australian Journal of Zoology 22(3), 417–425 (1974)

Campbell, N., Mahon, R.: A multivariate study of variation in two spe cies of rock crab of the genus leptograpsus. Australian Journal of Zoology 22(3), 417–425 (1974)

1974
[31]

IEEE Transac- tions on Pattern Analysis and Machine Intelligence 16(5), 550–554 (2002)

Hull, J.J.: A database for handwritten text recognition researc h. IEEE Transac- tions on Pattern Analysis and Machine Intelligence 16(5), 550–554 (2002)

2002
[32]

UCI Machine Learning Re pository (2002) 28

Cattral, R., Oppacher, F.: Poker Hand. UCI Machine Learning Re pository (2002) 28

2002

[1] [1]

ACM Co mputing Surveys (CSUR) 31(3), 264–323 (1999)

Jain, A.K., Murty, M.N., Flynn, P.J.: Data clustering: a review. ACM Co mputing Surveys (CSUR) 31(3), 264–323 (1999)

1999

[2] [2]

Pattern R ecognition Letters 31(8), 651–666 (2010)

Jain, A.K.: Data clustering: 50 years beyond k-means. Pattern R ecognition Letters 31(8), 651–666 (2010)

2010

[3] [3]

Annual Review of Statistics and Its Application 10(1), 573–595 (2023)

Gormley, I.C., Murphy, T.B., Raftery, A.E.: Model-based clustering . Annual Review of Statistics and Its Application 10(1), 573–595 (2023)

2023

[4] [4]

contributions to the mathematical theory of ev olution

Pearson, K.: Iii. contributions to the mathematical theory of ev olution. Proceed- ings of the Royal Society of London 54(326-330), 329–333 (1894)

[5] [5]

Lindsay, B.G.: Mixture models: theory, geometry, and application s. (1995). Ims

1995

[6] [6]

McLachlan, G.J., Peel, D.: Finite mixture models (2000)

2000

[7] [7]

Statistics Surveys 12(none), 18–65 (2018)

Fop, M., Murphy, T.B.: Variable selection methods for model-based clustering. Statistics Surveys 12(none), 18–65 (2018)

2018

[8] [8]

Journal of the Royal Statistical Society : Series B (Methodological) 39(1), 1–22 (1977) 26

Dempster, A.P., Laird, N.M., Rubin, D.B.: Maximum likelihood from incomp lete data via the EM algorithm. Journal of the Royal Statistical Society : Series B (Methodological) 39(1), 1–22 (1977) 26

1977

[9] [9]

arXiv preprint arXiv:1612.02099 (2016)

Lu, Y., Zhou, H.H.: Statistical and computational guarantees of lloyd’s algorithm and its variants. arXiv preprint arXiv:1612.02099 (2016)

work page arXiv 2016

[10] [10]

Annals of Statistics 49(5), 2506–2530 (2021)

L¨ oﬄer, M., Zhang, A.Y., Zhou, H.H.: Optimality of spectral cluste ring in the gaussian mixture model. Annals of Statistics 49(5), 2506–2530 (2021)

2021

[11] [11]

IEEE Transactions on Information Theory 67(6), 4223–4238 (2021)

Chen, X., Yang, Y.: Cutoﬀ for exact recovery of gaussian mixtu re models. IEEE Transactions on Information Theory 67(6), 4223–4238 (2021)

2021

[12] [12]

Annals of Statistics 50(4), 2096–2126 (2022)

Ndaoud, M.: Sharp optimal recovery in the two component gaus sian mixture model. Annals of Statistics 50(4), 2096–2126 (2022)

2096

[13] [13]

European Journal of Applied Mathematics 36(3), 491–523 (2025)

Li, Z.: Exact recovery of community detection in k-community ga ussian mixture models. European Journal of Applied Mathematics 36(3), 491–523 (2025)

2025

[14] [14]

Advances in Neural In formation Processing Systems 37, 113698–113741 (2024)

Chen, X., Zhang, A.Y.: Achieving optimal clustering in gaussian mixt ure mod- els with anisotropic covariance structures. Advances in Neural In formation Processing Systems 37, 113698–113741 (2024)

2024

[15] [15]

IEE E Transactions on Automatic Control 19(6), 716–723 (2003)

Akaike, H.: A new look at the statistical model identiﬁcation. IEE E Transactions on Automatic Control 19(6), 716–723 (2003)

2003

[16] [16]

Annals of Statis tics, 461–464 (1978)

Schwarz, G.: Estimating the dimension of a model. Annals of Statis tics, 461–464 (1978)

1978

[17] [17]

Journal of the royal statistical so ciety: series b (statistical methodology) 63(2), 411–423 (2001)

Tibshirani, R., Walther, G., Hastie, T.: Estimating the number of clu sters in a data set via the gap statistic. Journal of the royal statistical so ciety: series b (statistical methodology) 63(2), 411–423 (2001)

2001

[18] [18]

: X-means: Extending k-means with eﬃcient esti- mation of the number of clusters

Pelleg, D., Moore, A.W., et al. : X-means: Extending k-means with eﬃcient esti- mation of the number of clusters. In: ICML, vol. 1, pp. 727–734 (2 000). Stanford, CA

[19] [19]

Advances in Neu ral Informa- tion Processing Systems 16 (2003)

Hamerly, G., Elkan, C.: Learning the k in k-means. Advances in Neu ral Informa- tion Processing Systems 16 (2003)

2003

[20] [20]

Advances in Neural Information Processing Systems 19 (2006)

Feng, Y., Hamerly, G.: Pg-means: learning the number of cluster s in data. Advances in Neural Information Processing Systems 19 (2006)

2006

[21] [21]

Computational Sta tistics & Data Analysis 56(6), 1381–1395 (2012)

Melnykov, V., Melnykov, I.: Initializing the em algorithm in gaussian m ixture models with an unknown number of components. Computational Sta tistics & Data Analysis 56(6), 1381–1395 (2012)

2012

[22] [22]

In: Proceedings Eighth International Confer ence on Artiﬁcial Intelligence and Statistics, pp

Corduneanu, A., Bishop, C.M.: Variational bayesian model select ion for mix- ture distributions. In: Proceedings Eighth International Confer ence on Artiﬁcial Intelligence and Statistics, pp. 27–34 (2001). Morgan Kaufmann 27

2001

[23] [23]

IEEE Transactions on Pattern An alysis and Machine Intelligence 28(6), 1013–1018 (2006)

Constantinopoulos, C., Titsias, M.K., Likas, A.: Bayesian feature and model selec- tion for gaussian mixture models. IEEE Transactions on Pattern An alysis and Machine Intelligence 28(6), 1013–1018 (2006)

2006

[24] [24]

Bishop, C.M., Nasrabadi, N.M.: Pattern recognition and machine lea rning 4(4) (2006)

2006

[25] [25]

Statistics and Computing 26(1), 303–324 (2016)

Malsiner-Walli, G., Fr¨ uhwirth-Schnatter, S., Gr¨ un, B.: Model-b ased clustering based on sparse ﬁnite gaussian mixtures. Statistics and Computing 26(1), 303–324 (2016)

2016

[26] [26]

IEEE Transactions on Signal Pro cessing 66(20), 5392–5406 (2018)

Teklehaymanot, F.K., Muma, M., Zoubir, A.M.: Bayesian cluster enu meration cri- terion for unsupervised learning. IEEE Transactions on Signal Pro cessing 66(20), 5392–5406 (2018)

2018

[27] [27]

Statistica Sinica, 147–169 (2017)

Huang, T., Peng, H., Zhang, K.: Model selection for gaussian mixt ure models. Statistica Sinica, 147–169 (2017)

2017

[28] [28]

Jour nal of Econo- metrics 249, 105958 (2025)

Budanova, S.: Penalized estimation of ﬁnite mixture models. Jour nal of Econo- metrics 249, 105958 (2025)

2025

[29] [29]

Annals of Eugenics 7(2), 179–188 (1936)

Fisher, R.A.: The use of multiple measurements in taxonomic proble ms. Annals of Eugenics 7(2), 179–188 (1936)

1936

[30] [30]

Australian Journal of Zoology 22(3), 417–425 (1974)

Campbell, N., Mahon, R.: A multivariate study of variation in two spe cies of rock crab of the genus leptograpsus. Australian Journal of Zoology 22(3), 417–425 (1974)

1974

[31] [31]

IEEE Transac- tions on Pattern Analysis and Machine Intelligence 16(5), 550–554 (2002)

Hull, J.J.: A database for handwritten text recognition researc h. IEEE Transac- tions on Pattern Analysis and Machine Intelligence 16(5), 550–554 (2002)

2002

[32] [32]

UCI Machine Learning Re pository (2002) 28

Cattral, R., Oppacher, F.: Poker Hand. UCI Machine Learning Re pository (2002) 28

2002