Incremental multi-view spectral clustering with sparse and connected graph learning.

Area of Science:

• Data Science
• Machine Learning
• Computer Vision

Background:

• Traditional multi-view clustering struggles with scalability as the number of views increases.
• Existing methods often require fusing all views simultaneously, leading to high computational and storage costs.
• Re-fusing views incrementally is computationally expensive and requires storing all historical data.

Purpose of the Study:

• To propose an efficient incremental multi-view spectral clustering method.
• To address the challenge of increasing numbers of views in multi-view clustering.
• To reduce the storage and computational burden of traditional methods.

Main Methods:

• Developed an efficient incremental multi-view spectral clustering method (SCGL).
• Employed sparse and connected graph learning to enhance clustering performance.
• Utilized a single consensus similarity matrix to represent historical view information.
• Reconstructed the consensus matrix incrementally with new views.

Main Results:

• The proposed SCGL method demonstrates superior clustering accuracy compared to traditional approaches.
• SCGL effectively handles multi-view clustering scenarios with a growing number of views.
• The method reduces noise and preserves essential cluster connections through graph learning.

Conclusions:

• SCGL offers an efficient and scalable solution for incremental multi-view clustering.
• The integration of sparse and connected graph learning improves robustness and accuracy.
• This method is well-suited for dynamic multi-view learning environments where views accumulate over time.