Robust Subspace Clustering by Cauchy Loss Function.

Area of Science:

• Data Science
• Machine Learning
• Statistics

Background:

• Subspace clustering aims to find low-dimensional structures in high-dimensional data.
• Current spectral clustering-based methods often overlook the impact of complex noise distributions.
• Real-world data frequently contains noise that can significantly affect clustering accuracy.

Purpose of the Study:

• To propose a robust subspace clustering method that effectively mitigates the influence of noise.
• To introduce a novel approach utilizing the Cauchy loss function (CLF) for noise suppression in subspace clustering.
• To theoretically validate the grouping efficacy of the proposed method.

Main Methods:

• Developed a subspace clustering algorithm incorporating the Cauchy loss function (CLF).
• Utilized CLF to penalize noise terms, leveraging its bounded influence function to reduce the impact of outliers.
• Provided theoretical proof for the grouping effect, demonstrating its ability to cluster correlated data.

Main Results:

• The proposed Cauchy loss function-based method effectively suppresses large noise in real-world datasets.
• Theoretical analysis confirmed the method's capability to group highly correlated data points.
• Experimental results on five real datasets showed superior performance over several representative clustering algorithms.

Conclusions:

• The Cauchy loss function-based subspace clustering method offers a robust solution for noisy high-dimensional data.
• The method demonstrates strong performance in identifying underlying data structures and grouping correlated samples.
• This approach advances subspace clustering techniques by addressing the critical challenge of complex noise patterns.