High dimensional clustering for mixture models

Clustering is an essential subject in unsupervised learning. It is a common technique used in many fields, including machine learning, statistics, bioinformatics, and computer graphics. Classifying samples into homogeneous groups is based on different criterions. In this thesis, we focus on the clu...

Full description

Saved in:
Bibliographic Details
Main Author: Liu, Yiming
Other Authors: Pan Guangming
Format: Thesis-Doctor of Philosophy
Language:English
Published: Nanyang Technological University 2020
Subjects:
Online Access:https://hdl.handle.net/10356/142941
Tags: Add Tag
No Tags, Be the first to tag this record!
Institution: Nanyang Technological University
Language: English
id sg-ntu-dr.10356-142941
record_format dspace
spelling sg-ntu-dr.10356-1429412023-02-28T23:33:16Z High dimensional clustering for mixture models Liu, Yiming Pan Guangming School of Physical and Mathematical Sciences GMPAN@ntu.edu.sg Science::Mathematics Clustering is an essential subject in unsupervised learning. It is a common technique used in many fields, including machine learning, statistics, bioinformatics, and computer graphics. Classifying samples into homogeneous groups is based on different criterions. In this thesis, we focus on the clusters that are characterized by the different parameters (i.e., means and covariances), and we study the clustering method for the high dimensional mixture data. According to this setting, we propose two new methods, Covariance clustering method and {\it Two-step} method. Also, we investigate and develop the Mean clustering method from both theoretical and practical aspects by random matrix theory. Specifically, the first part focuses on the clustering when the data are collected from a mixture distribution with distinct covariance matrices. We provide a new algorithm to address this issue and find the misclustering rate theoretically. In the second part, for the data with different means, we provide a noncentered and centered version of Mean clustering method. Moreover, to give a theoretical justification of these two methods, we prove that the results of no eigenvalue outside the support of the limiting spectral distribution and exact separation of eigenvalues of large-dimensional sample covariance matrices can be extended to low rank information plus general noise models. In the third part, when either means or covariances are distinct, we propose a Two-step method to do clustering. Both theoretical and numerical properties of the Two-step method are discussed. Simulation studies and real data analysis also demonstrate that the Two-step method outperforms the other methods under a variety of settings. Doctor of Philosophy 2020-07-14T07:27:43Z 2020-07-14T07:27:43Z 2020 Thesis-Doctor of Philosophy Liu, Y. (2020). High dimensional clustering for mixture models. Doctoral thesis, Nanyang Technological University, Singapore. https://hdl.handle.net/10356/142941 10.32657/10356/142941 en This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License (CC BY-NC 4.0). application/pdf Nanyang Technological University
institution Nanyang Technological University
building NTU Library
continent Asia
country Singapore
Singapore
content_provider NTU Library
collection DR-NTU
language English
topic Science::Mathematics
spellingShingle Science::Mathematics
Liu, Yiming
High dimensional clustering for mixture models
description Clustering is an essential subject in unsupervised learning. It is a common technique used in many fields, including machine learning, statistics, bioinformatics, and computer graphics. Classifying samples into homogeneous groups is based on different criterions. In this thesis, we focus on the clusters that are characterized by the different parameters (i.e., means and covariances), and we study the clustering method for the high dimensional mixture data. According to this setting, we propose two new methods, Covariance clustering method and {\it Two-step} method. Also, we investigate and develop the Mean clustering method from both theoretical and practical aspects by random matrix theory. Specifically, the first part focuses on the clustering when the data are collected from a mixture distribution with distinct covariance matrices. We provide a new algorithm to address this issue and find the misclustering rate theoretically. In the second part, for the data with different means, we provide a noncentered and centered version of Mean clustering method. Moreover, to give a theoretical justification of these two methods, we prove that the results of no eigenvalue outside the support of the limiting spectral distribution and exact separation of eigenvalues of large-dimensional sample covariance matrices can be extended to low rank information plus general noise models. In the third part, when either means or covariances are distinct, we propose a Two-step method to do clustering. Both theoretical and numerical properties of the Two-step method are discussed. Simulation studies and real data analysis also demonstrate that the Two-step method outperforms the other methods under a variety of settings.
author2 Pan Guangming
author_facet Pan Guangming
Liu, Yiming
format Thesis-Doctor of Philosophy
author Liu, Yiming
author_sort Liu, Yiming
title High dimensional clustering for mixture models
title_short High dimensional clustering for mixture models
title_full High dimensional clustering for mixture models
title_fullStr High dimensional clustering for mixture models
title_full_unstemmed High dimensional clustering for mixture models
title_sort high dimensional clustering for mixture models
publisher Nanyang Technological University
publishDate 2020
url https://hdl.handle.net/10356/142941
_version_ 1759853325275103232