Multiple Kernel k-Means Clustering by Selecting Representative Kernels.

Area of Science:

• Machine Learning
• Data Mining
• Computational Statistics

Background:

• Kernel k-means clustering extends k-means for non-linearly separable data.
• Multiple kernel learning combines kernels for optimal clustering but often suffers from redundancy.

Purpose of the Study:

• To address kernel redundancy in multiple kernel learning for clustering.
• To develop an efficient and effective multiple kernel k-means clustering algorithm.

Main Methods:

• Proposed a subset selection strategy to identify diverse and representative kernels.
• Integrated subset selection into the multiple kernel k-means framework.
• Developed an alternating minimization method for optimizing kernel coefficients and cluster membership.

Main Results:

• Demonstrated effective selection of diverse kernel subsets.
• Achieved improved clustering performance and efficiency compared to existing methods.
• Validated the approach on benchmark and real-world datasets.

Conclusions:

• The proposed method effectively mitigates kernel redundancy in multiple kernel learning.
• Subset selection enhances the performance and efficiency of kernel k-means clustering.
• The approach offers a superior alternative for complex data clustering tasks.