Multi-view graph-regularized deep metric subspace clustering network

Area of Science:

• Artificial Intelligence
• Machine Learning
• Data Science

Background:

• Deep neural networks advance multi-view subspace clustering for nonlinear data.
• Multi-view Self-Expressive Subspace Clustering (MSESC) offers computational efficiency but struggles with high-order geometric structures and clustering distribution guidance.

Purpose of the Study:

• To address limitations in existing multi-view subspace clustering methods.
• To enhance representation discriminability for improved clustering performance.

Main Methods:

• Proposes Multi-View Graph Regularized Deep Metric Subspace Clustering (MVGR-DMSC).
• Introduces a dual-order graph regularization module to capture complex local geometric relationships.
• Incorporates an adaptive view-weighted deep clustering module using Kullback-Leibler divergence for guided representation learning.

Main Results:

• MVGR-DMSC demonstrates superior performance compared to state-of-the-art methods, including MSESC.
• The framework achieves better accuracy and robustness in multi-view subspace clustering.
• Evaluations on five benchmark datasets validate the proposed method's effectiveness.

Conclusions:

• MVGR-DMSC effectively captures high-order geometric structures and leverages clustering distribution guidance.
• The novel framework significantly improves upon existing multi-view subspace clustering techniques.
• The proposed method offers a robust and accurate solution for complex multi-view data analysis.