Fast Multi-view Discrete Clustering via Spectral Embedding Fusion.

Area of Science:

• Machine Learning
• Data Mining
• Artificial Intelligence

Background:

• Multi-view spectral clustering (MVSC) is valuable for diverse data but struggles with large datasets due to similarity matrix fusion and post-discretization.
• Existing MVSC methods face challenges with noise and two-stage mismatches, reducing clustering effectiveness.

Purpose of the Study:

• To develop a novel fast multi-view discrete clustering (FMVDC) model for efficient and effective large-scale clustering.
• To overcome the limitations of traditional MVSC, including computational complexity and reduced accuracy.

Main Methods:

• Developed the FMVDC model using spectral embedding fusion to directly obtain discrete clusters, bypassing similarity matrix fusion and post-discretization.
• Implemented an anchor-based spectral embedding strategy to reduce computational complexity from cubic to linear.
• Employed a coordinate descent method for efficient optimization of the discrete FMVDC model.

Main Results:

• FMVDC integrates spectral embedding matrices ($n \times c$) to directly output discrete sample categories ($c$ clusters).
• The anchor-based strategy significantly reduces spectral analysis complexity.
• Extensive studies confirm FMVDC's superior performance over state-of-the-art methods, especially on large-scale datasets.

Conclusions:

• FMVDC offers a more efficient and effective approach to multi-view clustering for large-scale applications.
• The proposed model addresses key limitations of traditional MVSC, improving both speed and accuracy.