A Tensor Approach for Uncoupled Multiview Clustering.

Area of Science:

• Machine Learning
• Data Science
• Artificial Intelligence

Background:

• Multiview clustering is crucial for unsupervised learning.
• Existing methods struggle with uncoupled data, leading to performance degradation.
• Uncoupled data lacks clear cross-view correlation, hindering complementary information exploration.

Purpose of the Study:

• To propose a novel tensor approach for uncoupled multiview clustering (T-UMC).
• To address the limitations of existing methods in handling uncoupled data.
• To improve clustering performance by exploring high-order correlations and preserving local structures.

Main Methods:

• T-UMC identifies a reliable view using view-specific silhouette coefficient (VSSC).
• It employs pairwise cross-view coupling learning to integrate self-representation matrices.
• Tensor singular value decomposition (t-SVD)-based tensor nuclear norm (TNN) explores high-order correlations.
• Manifold learning preserves view-specific local structures.

Main Results:

• T-UMC demonstrates superior performance on six benchmark datasets.
• The method effectively handles uncoupled data, a limitation of prior approaches.
• Experiments confirm the effectiveness of exploring high-order correlations and preserving local structures.

Conclusions:

• T-UMC offers a robust solution for uncoupled multiview clustering.
• The proposed tensor approach significantly enhances clustering accuracy.
• This work advances unsupervised learning techniques for real-world, uncoupled datasets.