One-Step Multi-View Clustering With Diverse Representation.

Area of Science:

• Machine Learning
• Data Mining
• Artificial Intelligence

Background:

• Multi-view clustering leverages consistent and complementary information across multiple data sources.
• Existing methods often face high complexity, limiting scalability, and suboptimal results due to two-step processes.
• Matrix factorization methods for multi-view clustering can be limited by fixed-dimension mapping and suboptimal outcomes.

Purpose of the Study:

• To propose a novel one-step multi-view clustering method (OMVCDR) that integrates representation learning and k-means.
• To address the limitations of high complexity and suboptimal clustering in existing multi-view clustering techniques.
• To improve the expressiveness and efficiency of multi-view clustering for large-scale applications.

Main Methods:

• The proposed OMVCDR method projects data into diverse latent spaces for comprehensive information capture.
• It employs a self-supervised approach for auto-weighting these diverse representations.
• Representation learning and k-means clustering are unified into a single framework for direct consensus label generation.

Main Results:

• The unified framework of representation learning and clustering significantly boosts the quality of clustering results.
• An efficient optimization algorithm with proven convergence properties was developed.
• Experiments on various datasets demonstrate the promising clustering performance of the OMVCDR method.

Conclusions:

• OMVCDR offers an effective and efficient solution for large-scale multi-view clustering.
• The one-step approach overcomes the limitations of traditional two-step methods, leading to improved clustering accuracy.
• The method's ability to utilize diverse representations enhances its overall performance and applicability.