SoK: Efficient Privacy-preserving Clustering

Abstract Clustering is a popular unsupervised machine learning technique that groups similar input elements into clusters. It used in many areas ranging from business analysis to health care. In of these applications, sensitive information clustered should not be leaked. Moreover, nowadays it often required combine data multiple sources increase the quality as well outsource complex computation powerful cloud servers. This calls for efficient privacy-preserving clustering. this work, we systematically analyze state-of-the-art We implement and benchmark today’s four most fully private clustering protocols by Cheon et al. (SAC’19), Meng (ArXiv’19), Mohassel (PETS’20), Bozdemir (ASIACCS’21) with respect communication, computation, quality. compare them, assess their limitations practical use real-world conclude open challenges.