Re-weighted Discriminatively Embedded K-Means for Multi-view Clustering.

Area of Science:

• Computer Science
• Data Science
• Machine Learning

Background:

• Multi-view data, common in applications like image analysis, presents challenges due to its high dimensionality and diverse feature representations.
• Efficiently clustering multi-view data from distinct perspectives remains a significant challenge in data analysis.

Purpose of the Study:

• To propose a novel multi-view clustering framework, Re-weighted Discriminatively Embedded KMeans (RDEKM), designed for high-dimensional data.
• To develop a robust method that mitigates the influence of outliers and performs dimension reduction during clustering.

Main Methods:

• The proposed Re-weighted Discriminatively Embedded KMeans (RDEKM) framework utilizes a multi-view least-absolute residual model.
• An unsupervised optimization scheme employing Iterative Re-weighted Least Squares is used to solve the least-absolute residual problem.
• Adaptive control of multiple weights is achieved based on low-dimensional subspaces and a common clustering indicator matrix.

Main Results:

• RDEKM demonstrated substantial improvements in clustering accuracy compared to state-of-the-art multi-view clustering methods.
• The method effectively handles high-dimensional multi-view data, showing superior performance on widely used benchmark datasets.
• Theoretical analysis, including optimality and convergence, supports the proposed framework's effectiveness.

Conclusions:

• The Re-weighted Discriminatively Embedded KMeans (RDEKM) framework offers a superior approach to multi-view clustering.
• The method's robustness to outliers and dimension reduction capabilities make it highly effective for real-world applications.
• RDEKM significantly enhances clustering accuracy, validating its superiority over existing techniques.