AMCFCN: attentive multi-view contrastive fusion clustering net

Area of Science:

• Artificial Intelligence
• Machine Learning
• Data Science

Background:

• Deep learning advances multi-view clustering, aiming for unified data representations.
• Existing methods struggle with noise in view-specific data and information loss during fusion.
• These limitations hinder accurate clustering performance in multi-view datasets.

Purpose of the Study:

• To address noise and information loss in multi-view clustering.
• To develop a novel technique for extracting robust and consistent data representations.
• To improve the accuracy of multi-view clustering outcomes.

Main Methods:

• Introduced a "contrastive attentive strategy" for noise reduction and feature preservation.
• Developed a unified framework (AMCFCN) integrating view-specific encoders, a hybrid attention module, and a fusion module.
• Employed deep clustering for enhanced representation learning.

Main Results:

• The AMCFCN method effectively extracts robust view-specific representations with reduced noise.
• The approach preserves view completeness and extracts consistent representations.
• Experimental results show AMCFCN outperforms seven competitive multi-view clustering methods on four datasets.

Conclusions:

• The proposed contrastive attentive strategy and AMCFCN framework offer significant improvements in multi-view clustering.
• The method successfully balances noise reduction with the preservation of essential view-specific information.
• AMCFCN demonstrates superior performance, highlighting its potential for complex multi-view data analysis.