Fast Self-Guided Multi-View Subspace Clustering.

Area of Science:

• Computer Science
• Data Science
• Machine Learning

Background:

• Multi-view subspace clustering leverages complementary information from multiple object views.
• Existing methods struggle with noisy views and capturing cross-view diversity, impacting performance.
• View-shared anchor learning accelerates common data representation but has limitations.

Purpose of the Study:

• To propose a novel Fast Self-guided Multi-view Subspace Clustering (FSMSC) algorithm.
• To address limitations of existing methods, including noisy view impact and lack of cross-view diversity capture.
• To enhance the accuracy and robustness of multi-view clustering.

Main Methods:

• Integrates view-shared anchor learning with global-guided-local self-guidance learning into a unified model.
• Employs a local learning module for a locally-consistent shared representation.
• Utilizes a global learning module for a globally-discriminative representation from concatenated features.
• Incorporates an l2,1-norm constrained feature selection matrix for global-to-local guidance.

Main Results:

• The FSMSC algorithm effectively overcomes the negative impact of noisy views.
• It simultaneously utilizes multi-view consistency and diversity for improved feature representation.
• Demonstrates promising performance compared to state-of-the-art non-deep and deep multi-view clustering algorithms.
• Extensive experiments on various datasets validate the proposed model's effectiveness.

Conclusions:

• The proposed FSMSC algorithm offers a robust and effective solution for multi-view subspace clustering.
• It successfully balances multi-view consistency and diversity while mitigating noisy view effects.
• FSMSC shows superior performance, making it a valuable contribution to the field of multi-view clustering.