Late Fusion Multiple Kernel Clustering With Proxy Graph Refinement.

Area of Science:

• Machine Learning
• Data Mining
• Computer Vision

Background:

• Multiple Kernel Clustering (MKC) uses base kernels to enhance clustering.
• Late fusion MKC methods offer computational speed and good performance.
• Existing methods separate kernel partition learning and late fusion, leading to suboptimal results.

Purpose of the Study:

• To propose a novel late fusion multiple kernel clustering framework with proxy graph refinement (LFMKC-PGR).
• To address the suboptimal solutions arising from separated kernel partition learning and fusion processes in existing MKC algorithms.

Main Methods:

• Theoretically revisits the link between late fusion kernel partitions and spectral embedding.
• Constructs a proxy self-expressive graph from kernel base partitions for refinement.
• Develops an alternate algorithm with proven convergence to solve the optimization problem.

Main Results:

• The proposed LFMKC-PGR framework refines kernel partitions and captures partition relationships effectively.
• Experimental results on 12 multi-kernel benchmark datasets demonstrate the algorithm's effectiveness.
• The method shows improved clustering performance compared to existing approaches.

Conclusions:

• The LFMKC-PGR framework successfully integrates kernel partition learning and late fusion.
• The proxy graph refinement mechanism enhances clustering accuracy and efficiency.
• The proposed algorithm offers a promising advancement in multiple kernel clustering.