High-order correlation and consistency-aware multi-view clustering via anchor graph learning.

Area of Science:

• Machine Learning
• Data Mining
• Artificial Intelligence

Background:

• Multi-view clustering integrates information from multiple data sources.
• Traditional methods face challenges with computational efficiency, high-order correlations, and cross-view consistency, especially for large-scale, heterogeneous data.

Purpose of the Study:

• To propose a novel framework, HCAGL, addressing limitations of traditional multi-view clustering.
• To enhance computational efficiency, capture high-order correlations, and maintain cross-view consistency.

Main Methods:

• Leveraging anchor graph learning with a compact set of anchor points for dimensionality reduction and efficiency.
• Employing tensor Schatten p-norm on low-dimensional anchor embeddings to capture high-order correlations and propagate global consistency.
• Incorporating adaptive neighborhood graph learning for dynamic view-specific weight adjustment to enhance cross-view consistency.

Main Results:

• HCAGL demonstrates superior performance in capturing cross-view consistency and high-order correlations on six benchmark datasets.
• Achieved higher accuracy and better clustering quality compared to existing multi-view methods.
• Component analyses confirm positive contributions of design choices, ensuring stable and reliable results.

Conclusions:

• HCAGL offers an effective and computationally efficient solution for complex multi-view clustering.
• The framework successfully addresses challenges associated with large-scale and heterogeneous multi-view data.