Communication-Efficient Federated Multi-View Clustering.

Area of Science:

• Machine Learning
• Data Science
• Artificial Intelligence

Background:

• Federated multi-view clustering (FMVC) enables privacy-preserving data grouping across distributed clients.
• Existing FMVC methods suffer from high communication overhead and insufficient utilization of data similarities for large-scale datasets.

Purpose of the Study:

• To propose a communication-efficient federated multi-view clustering framework.
• To address the limitations of existing methods regarding communication costs and data similarity utilization.

Main Methods:

• Developed a framework approximating data representation using shared pseudo-labels and centroid matrices.
• Incorporated a linear kernel function to effectively consider pairwise data similarities without explicit computation.
• Achieved linear complexity concerning the number of samples for optimization.

Main Results:

• Demonstrated significant improvements over existing federated multi-view clustering methods.
• Achieved an average accuracy improvement of 26.84% and up to 98.4% communication overhead reduction.
• Outperformed centralized multi-view clustering approaches in both performance and computational efficiency, with substantial speedups.

Conclusions:

• The proposed communication-efficient federated multi-view clustering framework effectively reduces communication overhead and enhances computational efficiency.
• The method successfully leverages data similarities and achieves superior clustering performance compared to existing federated and centralized approaches.
• This framework offers a promising solution for large-scale, privacy-preserving multi-view clustering tasks.