Privacy-Preserving Decentralised Singular Value Decomposition

With the proliferation of data and emerging data-driven applications, how to perform data analytical operations while respecting privacy concerns has become a very interesting research topic. With the advancement of communication and computing technologies, e.g. the FoG computing concept and its associated Edge computing technologies, it is now appealing to deploy decentralized data-driven applications. Following this trend, in this paper, we investigate privacy-preserving singular value decomposition (SVD) solutions tailored for these new computing environments. We first analyse a privacy-preserving SVD solution by Chen et al., which is based on the Paillier encryption scheme and some heuristic randomization method. We show that (1) their solution leaks statistical information to an individual player in the system; (2) their solution leaks much more information when more than one players collude. Based on the analysis, we present a new solution, which distributes the SVD results into two different players in a privacy-preserving manner. In comparison, our solution minimizes the information leakage to both individual player and colluded ones, via randomization and threshold homomorphic encryption techniques.