Sequential Subspace Clustering via Temporal Smoothness for Sequential Data Segmentation.

Area of Science:

• Computer Science
• Data Mining
• Machine Learning

Background:

• Sequential data analysis presents challenges in capturing temporal dynamics.
• Existing subspace clustering methods often overlook the inherent temporal relationships in sequential data.

Purpose of the Study:

• To develop a novel sequential subspace clustering method that effectively models temporal smoothness.
• To introduce a local temporal regularization term that leverages temporal predictability.

Main Methods:

• A novel local temporal regularization term based on temporal predictability is designed.
• An efficient optimization algorithm using alternate convex search is employed to learn the coding matrix and dictionary.
• The proposed method minimizes short-term variance to recover temporal smoothness relationships.

Main Results:

• The proposed method demonstrates superior performance in sequential subspace clustering tasks.
• Extensive experiments on three datasets validate the effectiveness against state-of-the-art methods.
• The local temporal regularization is shown to be crucial for effective sequential subspace clustering.

Conclusions:

• The developed temporal smoothness sequential subspace clustering method is effective for analyzing sequential data.
• The novel local temporal regularization significantly enhances the ability to capture temporal dynamics.
• This approach offers a promising direction for future research in sequential data analysis.