From Ensemble Clustering to Subspace Clustering: Cluster Structure Encoding.

Area of Science:

• Computer Science
• Data Science
• Machine Learning

Background:

• Subspace clustering (SC) identifies clusters in high-dimensional data.
• Ensemble clustering (EC) combines multiple clustering results to improve robustness.
• Integrating EC and SC remains a challenge for enhanced data representation.

Purpose of the Study:

• To propose a novel algorithm that effectively integrates ensemble clustering (EC) into subspace clustering (SC).
• To develop a method that leverages higher-order data relationships for improved clustering performance.
• To introduce a fast predictive coding mechanism for enhanced data representation.

Main Methods:

• Learning a low-rank representation (LRR) from higher-order data relationships induced by ensemble K-means coding.
• Utilizing an encoding function parameterized by neural networks to predict the LRR from partitions.
• Employing an alternating optimization framework for joint learning of LRR, encoding function, and higher-order relationships.

Main Results:

• The proposed algorithm successfully integrates partition information and original features.
• Achieved superior data representations compared to methods using single sources.
• Demonstrated effectiveness on eight benchmark datasets in various clustering tasks.

Conclusions:

• The novel algorithm offers a significant advancement in integrating ensemble and subspace clustering.
• The method provides a robust and effective approach for complex clustering tasks.
• Experimental validation confirms the superiority of the proposed approach over state-of-the-art methods.