SimMTC: Simple Multi-View Tensor Clustering.

Area of Science:

• Machine Learning
• Data Mining
• Artificial Intelligence

Background:

• Tensor-based multi-view clustering shows promise but often treats views independently, missing complementary information and global context.
• Existing methods may lose crucial data due to weak consistency constraints when extracting shared information across views.

Purpose of the Study:

• To propose Simple Multi-view Tensor Clustering (SimMTC), a novel algorithm designed to achieve globality and strong consistency in multi-view clustering.
• To overcome the limitations of independent view processing and weak consistency constraints in traditional tensor-based clustering.

Main Methods:

• SimMTC utilizes Fast Fourier Transform (FFT) on bipartite graphs to capture high- and low-frequency information, encoding sample-anchor similarities across all views for global context.
• The algorithm employs orthogonal tensor factorization in the frequency domain and introduces a novel FFT-based strong consistency constraint.
• An efficient alternative optimization algorithm is developed to solve the proposed SimMTC optimization problem.

Main Results:

• SimMTC effectively captures global information by processing similarities across all views in the frequency domain.
• The novel strong consistency constraint enhances the extraction of relevant, consistent information across different data views.
• Extensive experiments on real-world datasets confirm that SimMTC achieves state-of-the-art clustering performance.

Conclusions:

• SimMTC offers a significant advancement in multi-view clustering by effectively integrating global information and enforcing strong consistency.
• The proposed method, leveraging FFT and orthogonal tensor factorization, demonstrates superior performance compared to existing state-of-the-art techniques.
• SimMTC provides a robust and efficient framework for tackling complex multi-view clustering tasks.