Efficient and Effective Regularized Incomplete Multi-View Clustering.

Area of Science:

• Machine Learning
• Data Mining
• Computer Science

Background:

• Incomplete Multi-view Clustering (IMVC) combines multiple incomplete datasets to boost clustering.
• Multiple Kernel k-means with Incomplete Kernels (MKKM-IK) is a benchmark but has high computational and storage costs.
• MKKM-IK's kernel imputation method leads to over-complicated optimization and limited performance gains.

Purpose of the Study:

• To propose an Efficient and Effective Incomplete Multi-view Clustering (EE-IMVC) algorithm.
• To address the computational and optimization complexities of existing IMVC methods.
• To improve clustering performance by learning a consensus matrix instead of completing kernel matrices.

Main Methods:

• EE-IMVC imputes incomplete base matrices with a learned consensus clustering matrix.
• Prior knowledge is incorporated to regularize the consensus matrix.
• Two iterative algorithms with linear computational complexity are developed and their convergence proven.
• Generalization bounds of the proposed algorithms are theoretically analyzed.

Main Results:

• The proposed EE-IMVC algorithm significantly reduces computational and storage complexities.
• The algorithms demonstrate improved clustering accuracy compared to state-of-the-art methods.
• Experimental results validate the effectiveness and convergence of the developed algorithms.
• Theoretical analysis confirms the generalization capabilities of EE-IMVC.

Conclusions:

• EE-IMVC offers an efficient and effective solution for incomplete multi-view clustering.
• The method outperforms existing approaches in terms of clustering accuracy and computational efficiency.
• The incorporation of prior knowledge further enhances the robustness of the clustering results.