Discriminative embedded clustering: a framework for grouping high-dimensional data.

Area of Science:

• Machine Learning
• Data Mining
• Artificial Intelligence

Background:

• High-dimensional data presents significant challenges in machine learning and data mining due to the curse of dimensionality.
• Traditional methods often perform dimensionality reduction and clustering sequentially, which can be suboptimal.

Purpose of the Study:

• To propose a novel framework, discriminative embedded clustering, for joint dimensionality reduction and clustering.
• To address the challenges posed by high-dimensional data in clustering tasks.

Main Methods:

• A novel framework called discriminative embedded clustering is proposed, which iteratively alternates dimensionality reduction and clustering.
• An effective approach for solving the resulting non-convex optimization problem is presented.
• Comprehensive analyses of convergence, parameter determination, and computational complexity are provided.

Main Results:

• The proposed framework allows for viewing traditional approaches and revealing their intrinsic relationships.
• Experimental results on benchmark datasets demonstrate superior performance compared to state-of-the-art clustering and existing joint methods.
• The method effectively handles the curse of dimensionality in high-dimensional data.

Conclusions:

• Discriminative embedded clustering offers an effective solution for clustering high-dimensional data.
• The iterative joint approach outperforms traditional sequential methods and existing joint techniques.
• The framework provides insights into existing methods and facilitates the development of new ones.