Incomplete Multiview Clustering via Late Fusion.

Area of Science:

• Machine Learning
• Data Mining
• Computer Vision

Background:

• Real-world multiview clustering often encounters incomplete data due to noise or sensor failures.
• Existing early fusion methods for incomplete multiview clustering may not fully leverage reliable clustering from visible instances.

Purpose of the Study:

• To propose a novel late fusion method for incomplete multiview clustering that addresses data incompleteness.
• To develop a method that effectively integrates clustering results from multiple incomplete views.

Main Methods:

• Kernel k-means clustering is applied to visible instances within each view.
• Clustering results are encoded as zero-one matrices for late fusion.
• An alternating updating algorithm is designed to learn a unified clustering decision.

Main Results:

• The proposed late fusion method shows superior clustering performance compared to imputation and early fusion methods.
• Experiments on six benchmark datasets validate the effectiveness of the proposed approach.

Conclusions:

• Late fusion is a natural and effective strategy for incomplete multiview clustering.
• The proposed method provides a robust solution for handling missing data in multiview clustering applications.