Multiview Spectral Clustering With Bipartite Graph.

Area of Science:

• Machine Learning
• Data Mining
• Computer Vision

Background:

• Multi-view spectral clustering excels at correlating data across multiple perspectives.
• Existing methods face scalability issues due to high computational complexity (quadratic/cubic) and inability to integrate complementary information.
• Large-scale datasets in the big data era necessitate more efficient clustering approaches.

Purpose of the Study:

• To develop a novel multi-view spectral clustering model addressing computational inefficiency and information integration limitations.
• To encode complementary information between adjacency matrices and low-rank spatial structures.
• To enhance scalability for large datasets.

Main Methods:

• Developed a multi-view spectral clustering model utilizing Schatten p-norm regularization on a tensor of adjacency matrices.
• Introduced a fast model leveraging anchor graphs instead of full adjacency matrices for improved efficiency.
• Applied Schatten p-norm regularization on a tensor bipartite graph to encode complementary information in anchor graphs.
• Derived an efficient alternating algorithm for model optimization, proving convergence to a stationary KKT point.

Main Results:

• The proposed model effectively encodes complementary information from multiple views.
• The use of anchor graphs significantly improves computational efficiency, making it suitable for large-scale data.
• Experimental results demonstrate superior performance compared to existing methods.
• The optimization algorithm guarantees convergence.

Conclusions:

• The novel multi-view spectral clustering model offers an efficient and effective solution for large-scale clustering tasks.
• It successfully integrates complementary information across views and within data structures.
• The method provides a scalable and high-performing alternative to traditional spectral clustering techniques.