Align Then Tensorize: Multi-Level Consistent Anchor Graph Learning for Scalable Multi-View Clustering

Area of Science:

• Machine Learning
• Data Mining
• Computer Vision

Background:

• Multi-view clustering aims to leverage data from multiple sources.
• Existing tensor-based methods face scalability and structural consistency issues.
• Limitations include high computational complexity and inconsistent cross-view representations.

Purpose of the Study:

• To propose a novel, scalable multi-view clustering framework (ATTMVC).
• To address limitations of existing tensorial methods in scalability and cross-view consistency.
• To enhance the modeling of high-order correlations in multi-view data.

Main Methods:

• Developed an anchor-based graph learning framework for efficient reconstruction.
• Introduced cross-view anchor alignment into a shared latent space.
• Employed a Threshold Tensor Rank (TTR) surrogate for improved low-rank regularization.

Main Results:

• ATTMVC significantly reduces computational complexity compared to existing methods.
• The proposed alignment strategy enforces structural consistency across views.
• Experiments show ATTMVC outperforms state-of-the-art multi-view clustering algorithms.

Conclusions:

• ATTMVC offers a scalable and effective solution for multi-view clustering.
• The framework enhances cross-view comparability and high-order correlation modeling.
• The study provides a publicly available implementation for reproducibility.