Dynamic Privacy-preserving Collaborative Schemes for Average Computation

In this paper, we consider the privacy-preserving problem in collaborative computing. Based on a two-step average computation framework, we propose three privacy-aware schemes, all of which achieve different levels of privacy protections depending on data servers' trust degrees. Further, by carefully designing noises injected to the distributed computing process, we obtain dynamic privacy-preserving schemes, whose privacy preserving levels are measured by Kullback-Leibler differential privacy. In addition, we prove that the proposed schemes achieve convergence in different senses. Numerical experiments are finally conducted to verify the obtained privacy properties and convergence guarantees. Copyright (C) 2020 The Authors.