Contrastive Ensemble Clustering.

Area of Science:

• Data Science
• Machine Learning
• Artificial Intelligence

Background:

• Ensemble clustering combines multiple base clusterings for improved results.
• Current methods often fail to capture global structure from noisy data.
• Locality preservation in representation matrices is not explicitly addressed.

Purpose of the Study:

• To propose a novel contrastive ensemble clustering (CEC) method.
• To address limitations in capturing global structure and preserving locality in ensemble clustering.
• To leverage latent representation learning for improved clustering.

Main Methods:

• Developed a consensus mapping model for latent representation discovery from noisy observations.
• Introduced a contrastive regularizer to refine latent representations and preserve locality.
• Utilized average or weighted fusion of connective matrices from base clusterings.

Main Results:

• The proposed CEC method demonstrates superior performance on benchmark datasets.
• Successfully captures global structure information from noisy connective matrices.
• Explicitly preserves the locality property of the representation matrix.

Conclusions:

• CEC is a novel approach for ensemble clustering, integrating latent representation learning and contrastive components.
• The method effectively handles noisy data and enhances clustering quality.
• Represents a significant advancement in ensemble clustering techniques.