Adaptively Weighted Multiview Proximity Learning for Clustering.

Area of Science:

• Machine Learning
• Data Mining
• Computer Science

Background:

• Proximity-based methods are successful in multiview clustering but depend heavily on predefined proximity matrices.
• Existing multiview proximity learning (MVPL) methods often overlook inter-view correlations or fail to account for view weight differences.

Purpose of the Study:

• To address limitations in existing MVPL methods by proposing a novel approach.
• To enhance the quality of learned proximity matrices for improved multiview clustering performance.

Main Methods:

• Proposed adaptively weighted multiview proximity learning (AWMVPL).
• Incorporated both intra-view relation and inter-view correlation learning.
• Employed a self-weighted scheme for inter-view correlation and an adaptive weighting scheme to integrate view-specific information into a common cluster indicator matrix.

Main Results:

• AWMVPL effectively considers both intra-view and inter-view information.
• The self-weighted and adaptive weighting schemes improve the learning of proximity matrices.
• Experimental results demonstrate the superiority of AWMVPL over existing methods on various datasets.

Conclusions:

• AWMVPL overcomes key limitations of previous MVPL methods.
• The proposed method achieves superior performance in multiview clustering tasks.
• AWMVPL offers a more robust and effective approach to multiview proximity learning.