Latent Space Learning-Based Ensemble Clustering.

Area of Science:

• Computer Science
• Data Science
• Machine Learning

Background:

• Ensemble clustering combines multiple base clusterings for robust results.
• Current methods often use co-association matrices, but overlook shared latent spaces and discriminative matrix learning.

Purpose of the Study:

• To propose a novel ensemble clustering method, Latent spacE leArning baseD Ensemble Clustering (LEADEC).
• To address limitations in existing methods by exploring latent spaces and high-order relations among connective matrices.

Main Methods:

• Factorizing multiple connective matrices into a consensus latent space and specific matrices.
• Imposing orthogonal constraints for a more discriminative latent space representation.
• Integrating connective matrix learning, high-order relation investigation, and latent space representation within a unified framework.

Main Results:

• LEADEC demonstrates superior performance compared to existing representative methods.
• Experiments on seven benchmark datasets validate the effectiveness of the proposed approach.

Conclusions:

• The proposed LEADEC method effectively leverages latent spaces and high-order relations for improved ensemble clustering.
• LEADEC offers a more robust and accurate approach to data clustering by learning discriminative connective matrices.