Anchor-based fast balanced multi-view clustering

Area of Science:

• Data Science
• Machine Learning
• Artificial Intelligence

Background:

• Multi-view clustering integrates diverse data perspectives, with graph-based methods excelling on non-convex datasets.
• Existing graph-based methods suffer from high computational complexity, lack of balanced cluster sizes, and biased post-processing discretization.
• Addressing these limitations is crucial for practical applications like resource allocation and load balancing.

Purpose of the Study:

• To propose Fast Balanced Multi-View Clustering (FBMVC), an efficient and balanced graph-based multi-view clustering framework.
• To overcome the computational complexity, imbalanced cluster size, and post-processing bias challenges in current graph-based multi-view clustering.

Main Methods:

• FBMVC utilizes anchor and label transmission strategies to reduce computational costs.
• An adaptive view-weighting scheme optimizes weights for effective multi-view integration.
• The framework implicitly ensures balanced cluster distribution and obtains discrete labels within the optimization process.

Main Results:

• FBMVC significantly reduces computational complexity compared to existing methods.
• The proposed method achieves balanced clustering distributions.
• Experiments on six benchmark datasets show FBMVC outperforms state-of-the-art methods in clustering performance and efficiency.

Conclusions:

• FBMVC offers a fast, balanced, and effective solution for graph-based multi-view clustering.
• The integrated approach eliminates the need for separate post-processing steps, reducing bias.
• FBMVC demonstrates superior performance and efficiency, making it suitable for practical applications.