Restarted multiple kernel algorithms with self-guiding for large-scale multi-view clustering.

Area of Science:

• Data Science
• Machine Learning
• Artificial Intelligence

Background:

• Multi-view clustering methods often struggle with adaptively capturing relationships between diverse data views.
• Existing spectral clustering techniques using fixed or alternately updated similarity matrices are computationally expensive or inefficient for large datasets.

Purpose of the Study:

• To address the limitations of current multi-view spectral clustering methods by proposing an efficient framework.
• To develop a model that constructs multi-view similarity matrices inexpensively while preserving clustering insights.

Main Methods:

• Introduced a Sum-Ratio Multi-view Normalized Cut (Ncut) model with a shared representation embedding.
• Proposed a restarted multi-view multiple kernel clustering framework with self-guiding capabilities.
• Utilized similarity matrices with a strict block diagonal representation and an efficient multiple kernel selection technique.

Main Results:

• The proposed restarted algorithms demonstrated significant performance improvements, boosting results by 5-10 times for some popular multi-view clustering methods.
• The framework effectively enhances state-of-the-art multi-view clustering algorithms, particularly those with lower initial performance.
• Experiments on benchmark datasets confirmed the efficiency and effectiveness of the proposed approach, even with random initializations.

Conclusions:

• The developed framework offers a cost-effective solution for improving multi-view clustering performance.
• This approach provides a valuable boosting effect for various multi-view clustering algorithms, enhancing their ability to discover underlying data structures.
• The method successfully balances computational efficiency with the preservation of crucial clustering information across multiple data views.