Regularized Gaussian Mixture Model for High-Dimensional Clustering.

Area of Science:

• Data Science
• Machine Learning
• Statistics

Background:

• High-dimensional data presents challenges for clustering due to clusters existing in different subspaces.
• Traditional Gaussian Mixture Models (GMMs) struggle with maximum-likelihood estimation in high-dimensional settings.
• Existing methods often fail to simultaneously identify clusters and their intrinsic subspaces.

Purpose of the Study:

• To propose a novel regularized Gaussian Mixture Model (GMM) for simultaneous clustering and subspace identification.
• To enhance the performance of GMMs in high-dimensional data analysis.
• To develop a method that improves the estimation of local feature correlations.

Main Methods:

• A regularized GMM is proposed, incorporating a novel regularization technique for component covariance matrices.
• The regularization aims to find low-dimensional representations of these matrices.
• The Expectation-Maximization (EM) algorithm is adapted, with the M-step modified to include the regularization, involving an efficient determinant maximization solution.

Main Results:

• The proposed regularization method leads to improved estimation of local feature correlations.
• The modified EM algorithm effectively incorporates regularization for better GMM parameter estimation.
• Simulations on synthetic datasets demonstrate the method's superior performance.
• The approach shows potential value in real-world applications using diverse datasets.

Conclusions:

• The regularized GMM offers a robust solution for clustering high-dimensional data with varying subspaces.
• The method enhances GMM performance by addressing limitations in high-dimensional settings.
• This approach facilitates more accurate data representation and analysis in complex datasets.