Abstract:In this paper, we study the subspace clustering problem under complex noise. A wide class of reconstruction-based methods model the subspace clustering problem by combining a quadratic data-fidelity term and a regularization term. In a statistical framework, the data-fidelity term assumes to be contaminated by a unimodal Gaussian noise, which is a popular setting in most current subspace clustering models. However, the realistic noise is much more complex than our assumptions. Besides, the coarse representation of the data-fidelity term may depress the clustering accuracy, which is often used to evaluate the models. To address this issue, we propose the mixture of Gaussian regression (MoG Regression) for subspace clustering. The MoG Regression seeks a valid way to model the unknown noise distribution, which approaches the real one as far as possible, so that the desired affinity matrix is better at characterizing the structure of data in the real world, and furthermore, improving the performance. Theoretically, the proposed model enjoys the grouping effect, which encourages the coefficients of highly correlated points are nearly equal. Drawing upon the ideal of the minimum message length, a model selection strategy is proposed to estimate the numbers of the Gaussian components that shows a way how to seek the number of Gaussian components besides determining it by empirical value. In addition, the asymptotic property of our model is investigated. The proposed model is evaluated on the challenging datasets. The experimental results show that the proposed MoG Regression model significantly outperforms several state-of-the-art subspace clustering methods.

A Noise-Resistant Clustering Algorithm for Switching Regression Models

An Accelerating Method For Fuzzy C-Regression Models

Multivariate Regime Switching Model Estimation and Asset Allocation

A New Switching Scheme for Multi-model Predictive Control Using Clustering Modeling

Regime Switching Model Estimation: Spectral Clustering Hidden Markov Model

Cluster-Based Input Selection for Transparant Fuzzy Modeling

Noise-Aware Subspace Clustering

Censored Regression with Noisy Input

Spectral Clustering on Subspace for Parameter Estimation of Jump Linear Models

A Segmented Regime-Switching Model with Its Application to Stock Market Indices

Robust Subspace Clustering Via Thresholding Ridge Regression

Subspace Clustering under Complex Noise

A robust adaptive linear regression method for severe noise

Correntropy Induced L2 Graph for Robust Subspace Clustering

Penalized model-based clustering with cluster-specific diagonal covariance matrices and grouped variables

A Subspace Technique for The Identification of Switched Affine Models

Comparison of Statistical Methods for Analysis of Threshold Model of Two-Stage Clustered Randomized Data

Dynamic noise estimation: A generalized method for modeling noise fluctuations in decision-making

Subspace Clustering By Mixture Of Gaussian Regression

Semisupervised Clustering by Iterative Partition and Regression with Neuroscience Applications