Multiple graphs learning with a new weighted tensor nuclear norm.

Area of Science:

• Machine Learning
• Data Science
• Computer Vision

Background:

• Multi-view clustering methods leverage information from multiple data sources.
• Tensor nuclear norm minimization is a convex relaxation for rank minimization in clustering.
• Existing methods often treat singular values uniformly, limiting flexibility.

Purpose of the Study:

• To propose a novel weighted tensor nuclear norm minimization (WTNNM) for multi-view spectral clustering.
• To address the limitation of uniform singular value regularization in existing methods.
• To improve the capability and flexibility of multi-view clustering.

Main Methods:

• Constructing a 3-order tensor from transition probability matrices of different views.
• Learning a latent high-order transition probability matrix using a proposed weighted tensor nuclear norm.
• Incorporating prior knowledge of singular value importance.
• Performing spectral clustering on the learned transition probability matrix.

Main Results:

• The proposed WTNNM method effectively characterizes complementary and high-order information in multi-view data.
• Extensive experiments on five benchmarks demonstrate superior performance compared to state-of-the-art methods.
• The method shows improved clustering accuracy and robustness.

Conclusions:

• The weighted tensor nuclear norm minimization offers a more flexible and effective approach to multi-view spectral clustering.
• WTNNM successfully addresses the limitations of uniform singular value treatment.
• The proposed method provides a significant advancement in multi-view data analysis and clustering.