One-Step Multiview Clustering via Adaptive Graph Learning and Spectral Rotation

Area of Science:

• Machine Learning
• Data Mining
• Computer Science

Background:

• Graph-based multiview clustering methods typically involve multiple steps, separating graph learning from partition generation.
• This separation can lead to suboptimal performance as it fails to closely integrate these processes.

Purpose of the Study:

• To propose a novel one-step multiview clustering method that unifies graph learning and partition generation.
• To enhance clustering performance by overcoming the limitations of traditional multistep approaches.

Main Methods:

• Adaptive Graph Learning and Spectral Rotation (AGLSR) method is introduced.
• It adaptively learns affinity graphs for each view to capture sample relationships.
• A spectral embedding technique leverages shared feature spaces, and spectral rotation directly yields discrete clustering labels.

Main Results:

• The proposed AGLSR method demonstrates effectiveness across six metrics on benchmark datasets.
• An effective updating algorithm with proven convergence was developed for optimization.
• The one-step approach integrates graph learning and clustering more closely than traditional methods.

Conclusions:

• AGLSR offers a more effective and unified approach to multiview clustering.
• The method achieves superior performance by directly generating discrete labels through spectral rotation.
• The study provides a robust algorithm with demonstrated experimental validation.