Generative Partial Multi-View Clustering With Adaptive Fusion and Cycle Consistency.

Area of Science:

• Machine Learning
• Data Science
• Artificial Intelligence

Background:

• Multi-view clustering (MVC) is widely used but often assumes complete data across all views.
• Existing MVC methods struggle with incomplete multi-view data caused by corruption or sensor failures.
• Addressing incomplete data is crucial for real-world multi-view data analysis.

Purpose of the Study:

• To develop a novel generative model for multi-view clustering that handles incomplete data.
• To propose a method that explicitly generates missing data within a multi-view framework.
• To enhance clustering accuracy by effectively utilizing available and generated multi-view information.

Main Methods:

• A generative partial multi-view clustering (GP-MVC) model is proposed.
• It employs multi-view encoder networks for common low-dimensional representations and a clustering layer.
• View-specific generative adversarial networks with multi-view cycle consistency are used to generate missing data.

Main Results:

• The GP-MVC model demonstrates effectiveness in handling incomplete multi-view data.
• Experimental results show superior performance compared to state-of-the-art methods on benchmark datasets.
• The proposed method successfully generates missing views and improves clustering accuracy.

Conclusions:

• The GP-MVC model offers a robust solution for multi-view clustering with incomplete data.
• The integration of data generation and adaptive fusion enhances clustering performance.
• This approach advances the field of multi-view learning by addressing data incompleteness.