<ul data-eligibleForWebStory="true"><li>Federated learning has become popular for sharing sensitive data without sharing raw data.</li><li>Existing federated clustering methods are limited in handling complex data partitioning scenarios.</li><li>Data collaboration clustering (DC-Clustering) is a new federated clustering method proposed to address complex data partitioning scenarios.</li><li>DC-Clustering ensures privacy preservation by only sharing intermediate representations among institutions, not raw data.</li><li>The method supports k-means and spectral clustering, achieving results with one communication round to the central server.</li><li>Experiments with synthetic and benchmark datasets show that DC-Clustering performs comparably to centralized clustering.</li><li>DC-Clustering fills a gap in federated learning research by enabling effective knowledge discovery from distributed heterogeneous data.</li><li>Its features include privacy preservation, communication efficiency, and flexibility, making it valuable for privacy-sensitive domains like healthcare and finance.</li></ul>

A new type of federated clustering: A non-model-sharing approach

Discover more