Double structure scaled simplex representation for multi-view subspace clustering.

Area of Science:

• Data Science
• Machine Learning
• Artificial Intelligence

Background:

• The proliferation of big data necessitates advanced multi-view clustering (MVC) algorithms.
• Existing MVC methods often struggle with noisy affinity matrices, hindering accurate representation of latent data structures.
• This limitation leads to suboptimal clustering performance in complex datasets.

Purpose of the Study:

• To propose a novel Double Structure Scaled Simplex Representation (DSSSR) method for enhanced multi-view clustering.
• To address the limitations of imprecise affinity matrices in current MVC techniques.
• To improve the accuracy and robustness of clustering algorithms in big data environments.

Main Methods:

• Concatenation of multi-view data into a unified representation.
• Application of Scaled Simplex Representation (SSR) twice: first on concatenated data, then on the resulting affinity matrix, to refine it.
• Integration of the two-step SSR into a unified optimization framework with column sum constraints (0
• Development of an efficient optimization algorithm using the augmented Lagrangian method (ALM).

Main Results:

• The DSSSR method generates a cleaner and more accurate affinity matrix compared to traditional approaches.
• The unified optimization framework effectively refines the affinity matrix.
• Experimental results demonstrate superior clustering performance against state-of-the-art algorithms on benchmark datasets.

Conclusions:

• The proposed DSSSR method offers a significant advancement in multi-view clustering.
• The dual-step SSR approach effectively overcomes the issue of noisy affinity matrices.
• DSSSR provides a robust and efficient solution for clustering complex, multi-source big data.