An Ensemble and Multi-View Clustering Method Based on Kolmogorov Complexity.

Area of Science:

• Data Science
• Machine Learning
• Information Theory

Background:

• Merging clustering partitions is complex, especially with privacy constraints, diverse data features, or distributed data.
• The increasing volume of multi-view data and diverse clustering algorithms exacerbates the challenge of achieving a unified clustering outcome.
• Existing methods for combining clustering results often fall short in complex, real-world scenarios.

Purpose of the Study:

• To propose a novel clustering fusion algorithm for merging multiple clustering partitions into a single, coherent partition.
• To address the challenges posed by privacy-preserving constraints, heterogeneous data features, and distributed data computations.
• To provide a stable and effective solution for unsupervised multi-view learning through data partition merging.

Main Methods:

• Developed a clustering fusion algorithm that integrates existing clustering partitions from multiple sources or views.
• Employed an information theory model based on Kolmogorov complexity for the merging process.
• Utilized a method originally proposed for unsupervised multi-view learning.

Main Results:

• The proposed algorithm demonstrates a stable merging process.
• Achieved competitive results on various real and artificial datasets when compared to state-of-the-art methods.
• Successfully merged clustering partitions from multiple vector space models, sources, or views into a single partition.

Conclusions:

• The developed clustering fusion algorithm offers a robust approach to consolidating diverse clustering results.
• The method is particularly effective in scenarios with privacy concerns and multi-view data.
• This work advances unsupervised learning by providing a reliable technique for merging clustering partitions.