Multi-view subspace tensorization with attentive clustering embedding.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Data Science

Background:

• Deep learning-based multi-view clustering (DMVC) excels at nonlinear representation but struggles with high-order correlations.
• Traditional multi-view clustering (MVC) effectively uses tensor methods for inter-view dependencies.

Purpose of the Study:

• To propose STANCE (multi-view Subspace Tensorization with Attentive Clustering Embedding), a novel DMVC framework.
• To integrate tensor modeling into deep learning for improved high-order correlation capture.
• To enhance inter-view consistency using an attention-based adaptive fusion module.

Main Methods:

• Utilizes view-specific auto-encoders for robust subspace representations.
• Constructs a third-order tensor from latent features for low-rank constraint application.
• Employs an attention mechanism for sample-level adaptive fusion of view-specific representations.

Main Results:

• STANCE effectively captures high-order inter-view dependencies via tensorization and low-rank constraints.
• Attention-based fusion enhances consistency by dynamically weighting sample-level features.
• Experimental results show STANCE significantly outperforms state-of-the-art DMVC methods on benchmark datasets.

Conclusions:

• STANCE offers a powerful new approach for DMVC by combining deep learning with tensor modeling.
• The proposed method demonstrates superior performance in capturing complex inter-view relationships.
• STANCE provides a robust and effective solution for multi-view clustering tasks.