Self-Guided Partial Graph Propagation for Incomplete Multiview Clustering.

Area of Science:

• Machine Learning
• Data Science
• Computer Vision

Background:

• Multiview clustering (MVC) faces challenges with incomplete data, known as incomplete MVC (IMVC).
• Existing IMVC methods often struggle with data recovery due to insufficient information.
• Exploiting complementary and consistency information is crucial for effective IMVC.

Purpose of the Study:

• To develop a robust approach for incomplete multiview clustering.
• To address the limitations of existing methods in handling missing data.
• To effectively utilize both consistency and complementary information in IMVC.

Main Methods:

• A graph propagation framework is proposed for IMVC.
• Partial graphs represent sample similarity for incomplete views, translating missing instances to missing graph entries.
• A common graph is adaptively learned to self-guide propagation, refined iteratively by view-specific propagated graphs.

Main Results:

• The method successfully infers missing data entries through graph propagation.
• It effectively exploits consistency information across all views.
• An exclusive regularization term is incorporated to leverage complementary information, which is often overlooked.

Conclusions:

• The proposed graph propagation method significantly enhances incomplete multiview clustering performance.
• It offers a more effective way to handle missing data by exploiting both consistency and complementary information.
• Experimental results validate the superiority of the proposed method over state-of-the-art approaches.