Robust Multi-View Clustering With Incomplete Information.

Area of Science:

• Machine Learning
• Data Science
• Computer Vision

Background:

• Multi-view clustering methods often assume complete data, which is unrealistic.
• Incomplete data issues include partially view-unaligned (PVP) and sample-missing (PSP) problems.
• Existing methods struggle with these incomplete information challenges.

Purpose of the Study:

• To propose a novel, unified framework called SURE (robuSt mUlti-view clusteRing with incomplEte information).
• To address both PVP and PSP simultaneously in multi-view clustering.
• To introduce a noise-robust contrastive loss for handling false negatives.

Main Methods:

• SURE employs a contrastive learning paradigm.
• It utilizes available data pairs as positives and randomly selected cross-view samples as negatives.
• A noise-robust contrastive loss mitigates the impact of false negatives from random sampling.

Main Results:

• SURE effectively handles incomplete information challenges (PVP and PSP) in a unified manner.
• The noise-robust loss significantly reduces the influence of false negatives.
• Experiments show SURE outperforms 10 state-of-the-art methods in multi-view clustering.

Conclusions:

• SURE is the first method to simultaneously address PVP and PSP with a unified solution.
• This work pioneers the study of noisy correspondence (false negatives) in clustering.
• SURE demonstrates effectiveness and efficiency for robust multi-view clustering with incomplete data.