Incomplete Multiview Spectral Clustering With Adaptive Graph Learning.

Area of Science:

• Machine Learning
• Data Science
• Computer Vision

Background:

• Incomplete multiview clustering presents challenges due to missing data across different data sources.
• Existing methods often struggle to effectively integrate information from incomplete views.

Purpose of the Study:

• To propose a general framework for incomplete multiview clustering.
• To develop a method that learns a common representation from incomplete multi-view data.

Main Methods:

• Utilizes graph learning to construct adaptive graphs for each view, leveraging low-rank representation.
• Applies spectral clustering to obtain low-dimensional representations.
• Introduces a co-regularization term for learning a unified sample representation across all views.
• Employs k-means for final data partitioning.

Main Results:

• The proposed method demonstrates superior performance on seven incomplete multiview datasets.
• Achieves state-of-the-art results compared to existing incomplete multiview clustering techniques.

Conclusions:

• The developed framework is effective for incomplete multiview clustering.
• The integration of graph learning, spectral clustering, and co-regularization offers a robust approach to handling missing data in multiview learning.