A Lightweight Privacy-Enhanced Federated Clustering Algorithm for Edge Computing.

Area of Science:

• Edge Computing
• Data Mining
• Machine Learning

Background:

• Distributed data in edge computing is dispersed, heterogeneous, and privacy-sensitive.
• Federated clustering faces challenges like high communication overhead, non-IID data, and privacy risks.

Purpose of the Study:

• To propose a privacy-enhanced federated k-means clustering algorithm for edge computing.
• To address challenges of non-IID data, communication overhead, and privacy leakage.

Main Methods:

• Leveraging locality-sensitive hashing (LSH) for privacy-preserving encryption of cluster centers.
• Implementing a single-round client-to-server communication protocol.
• Performing secondary weighted k-means clustering in an encrypted space on the server.

Main Results:

• The algorithm effectively mitigates the non-IID data problem while preserving privacy.
• Achieves global clustering in a single communication round, reducing overhead.
• Demonstrates robust clustering performance on MNIST and CIFAR-10 datasets.

Conclusions:

• The proposed algorithm offers an efficient, adaptable, and privacy-preserving solution for distributed data mining in edge environments.
• It enhances practicality in communication-constrained settings without relying on a trusted server.