Multi-View Clustering With Incremental Instances and Views.

Area of Science:

• Machine Learning
• Data Mining
• Artificial Intelligence

Background:

• Multi-view clustering (MVC) is increasingly important with diverse data sources.
• Dynamic environments require methods to handle continuously arriving instances and expanding views.
• Existing unsupervised MVC methods struggle with simultaneous instance and view increments.

Purpose of the Study:

• To propose a novel Multi-View Clustering method (MVC-IIV) for dynamic environments.
• To address the underexplored challenge of simultaneous instance and view increments in unsupervised learning.
• To develop an efficient and effective MVC approach for real-world data streams.

Main Methods:

• MVC-IIV employs a two-stage approach: initial and incremental learning.
• The incremental stage reuses a trained model to guide learning on new data and views.
• Key modules include multi-view embedding and consensus centroids, enhanced by consistency regularization.

Main Results:

• The proposed MVC-IIV method demonstrates effectiveness and efficiency.
• It successfully transfers historical knowledge to new data batches, improving clustering.
• Experimental results show superior performance compared to state-of-the-art approaches.

Conclusions:

• MVC-IIV provides an effective solution for unsupervised clustering in dynamic, multi-view data scenarios.
• The method offers linear time and space complexity, ensuring scalability.
• It advances the field of incremental multi-view clustering by addressing simultaneous instance and view growth.