Contrastive and adversarial regularized multi-level representation learning for incomplete multi-view clustering.

Area of Science:

• Machine Learning
• Data Science
• Artificial Intelligence

Background:

• Incomplete multi-view clustering is crucial for analyzing complex, partially observable systems.
• Existing algorithms struggle to balance data restoration with clustering and maintain representational consistency across views.

Purpose of the Study:

• To propose a novel method, Multi-level Representation Learning Contrastive and Adversarial Learning (MRL_CAL), for incomplete multi-view clustering.
• To jointly learn data restoration, consistent representation, and clustering by leveraging features from various subspaces.

Main Methods:

• Utilizes a variational auto-encoder for low-level, view-specific instance representation and adversarial learning for data restoration.
• Employs contrastive learning to integrate consistency across views and cluster labels into high-level representations.
• Formulates incomplete multi-view clustering as an overall objective where feature learning is guided by clustering.

Main Results:

• MRL_CAL successfully learns multi-level features across subspaces, mitigating representational conflicts and improving feature quality.
• The method demonstrates superior performance compared to state-of-the-art algorithms across various evaluation metrics.

Conclusions:

• MRL_CAL offers a promising approach to incomplete multi-view clustering by effectively addressing data restoration and representation consistency.
• The joint learning framework enhances the quality of learned features and the overall clustering performance.