Prototype Matching Learning for Incomplete Multi-view Clustering.

Area of Science:

• Data Science
• Machine Learning
• Computer Vision

Background:

• Incomplete multi-view clustering (IMVC) faces challenges due to partial data loss from sensor or equipment failures.
• Existing prototype-based IMVC methods often assume cross-view prototype alignment, which may not hold true, leading to the prototype-unaligned problem (PUP) and potential overfitting.
• Data imputation in IMVC can suffer from performance instability (PIP) due to varying data quality under different missing rates.

Purpose of the Study:

• To propose a novel method, Prototype Matching Learning for Incomplete Multi-view Clustering (PMIMC), to effectively handle incomplete multi-view data.
• To address the challenges of prototype misalignment (PUP) and performance instability (PIP) in IMVC.
• To improve the robustness and clustering accuracy of incomplete multi-view clustering algorithms.

Main Methods:

• PMIMC utilizes relational consistency learning to manage multi-view data heterogeneity.
• A robust prototype contrastive learning loss is employed to mitigate the effects of PUP.
• A prototype-based imputation strategy is developed to reduce imputation instability, especially at high missing rates.

Main Results:

• PMIMC demonstrates superior clustering performance compared to 13 state-of-the-art methods.
• The proposed method shows enhanced robustness in handling incomplete multi-view data.
• Experimental results validate the effectiveness of PMIMC in addressing PUP and PIP.

Conclusions:

• PMIMC offers a robust solution for incomplete multi-view clustering by effectively handling data heterogeneity, prototype misalignment, and imputation instability.
• The method achieves state-of-the-art performance and improved robustness.
• The developed techniques provide a significant advancement in the field of multi-view clustering.