LABIN: Balanced Min Cut for Large-Scale Data.

Area of Science:

• Machine Learning
• Data Mining
• Computer Science

Background:

• Spectral clustering algorithms face scalability challenges with large datasets due to high computational complexity.
• Existing methods like Self-Balanced Min-Cut (SBMC) are not optimized for massive data.
• Efficient clustering is crucial for analyzing large-scale datasets in various scientific domains.

Purpose of the Study:

• To propose a novel spectral clustering method that is scalable to large-scale data.
• To address the computational limitations of traditional spectral clustering algorithms.
• To introduce an efficient and effective solution for clustering massive datasets.

Main Methods:

• A new model is proposed, extending the Self-Balanced Min-Cut (SBMC) model.
• An anchor-based strategy is incorporated into the extended SBMC model.
• A fast spectral rotation with linear time complexity is developed to solve the new model.

Main Results:

• The proposed Large-Scale Balanced Min Cut (LABIN) method demonstrates superior performance on large-scale datasets.
• LABIN significantly outperforms state-of-the-art methods, including SBMC, in experimental evaluations.
• The method achieves efficient clustering with reduced computational complexity.

Conclusions:

• LABIN offers a scalable and computationally efficient spectral clustering solution for large datasets.
• The novel model and fast spectral rotation effectively address the limitations of previous algorithms.
• This advancement facilitates the analysis of massive data where traditional methods fail.