Unified Discriminative and Coherent Semi-supervised Subspace Clustering

Area of Science:

• Computer Vision
• Machine Learning
• Data Science

Background:

• Large, complex, and high-dimensional datasets pose challenges for traditional clustering methods.
• Subspace clustering aims to partition data into lower-dimensional subspaces.
• Semi-supervised subspace clustering, using limited labeled and abundant unlabeled data, offers a more practical and effective approach.

Purpose of the Study:

• To develop a novel semi-supervised subspace clustering framework.
• To enhance cluster coherence and discrimination by integrating data labels and affinity.
• To improve the performance of subspace clustering in high-dimensional data analysis.

Main Methods:

• A new regularity combining labels and affinity to ensure coherence within subspaces and discrimination between clusters.
• Integration with manifold smoothing and Gaussian fields and harmonic functions methods.
• Development of a unified optimization framework for semi-supervised subspace clustering.

Main Results:

• The proposed model effectively combines affinity and labels to guide each other for improved clustering.
• Demonstrated superior performance compared to existing state-of-the-art semi-supervised subspace clustering methods.
• The method's properties of being discriminative between clusters and coherent within clusters are advantageous.

Conclusions:

• The developed unified framework offers a significant advancement in semi-supervised subspace clustering.
• The integration of labels and affinity provides a robust approach for complex datasets.
• The method's effectiveness is validated through extensive experimental results, showing superiority over current techniques.