Essential Tensor Learning for Multi-View Spectral Clustering.

Area of Science:

• Machine Learning
• Data Mining
• Computer Vision

Background:

• Multi-view clustering leverages diverse data sources for improved performance.
• Existing self-representation methods face high computational complexity.
• Markov chain-based spectral clustering offers an alternative approach.

Purpose of the Study:

• To propose a novel essential tensor learning method for multi-view representation.
• To explore high-order correlations within multi-view data.
• To overcome the computational limitations of existing subspace clustering techniques.

Main Methods:

• Constructing a tensor from multi-view Markov chain transition probability matrices.
• Applying tensor singular value decomposition (t-SVD) with a tensor nuclear norm for low-rank preservation.
• Utilizing a tensor rotation operator to analyze inter-view relationships and reduce complexity.
• Employing the alternating direction method of multipliers (ADMM) for efficient optimization.

Main Results:

• The proposed tensor learning method effectively captures principle information across multiple views.
• The tensor rotation operator enhances the investigation of relationships among views.
• The ADMM optimization ensures efficient computation.
• Experimental results demonstrate superior performance compared to state-of-the-art methods on seven real-world datasets.

Conclusions:

• The novel essential tensor learning method provides an effective and efficient solution for multi-view clustering.
• The integration of Markov chains, t-SVD, and tensor rotation offers significant advantages.
• The method shows strong potential for various real-world applications.