Anchor Graph Learning with Double Noise Removal for Multi-View Clustering.

Area of Science:

• Machine Learning
• Data Mining
• Computer Science

Background:

• Existing anchor-based multi-view graph clustering methods struggle with noise removal across views.
• This noise leads to inaccurate consensus representations and degraded clustering quality.

Purpose of the Study:

• To propose a novel approach, Anchor Graph Learning with Double Noise Removal (AGLDR), for multi-view clustering.
• To simultaneously learn a consistent anchor graph and remove view-specific noise.

Main Methods:

• Introduced a low-rank constraint on the consistent anchor graph to capture global correlations.
• Minimized F-norm and L2,1 norm of noise terms to eliminate Gaussian and Laplacian noise, respectively.
• Developed a novel algorithm, AGLDR, for effective noise reduction in multi-view clustering.

Main Results:

• AGLDR demonstrates superior performance compared to state-of-the-art methods.
• The proposed method achieves higher clustering accuracy.
• AGLDR exhibits enhanced robustness in multi-view clustering tasks.

Conclusions:

• AGLDR effectively addresses the critical gap in noise reduction for multi-view clustering.
• The algorithm provides a robust and accurate solution for clustering complex multi-view data.
• Experimental results validate the superiority of AGLDR over existing techniques.