SLIM-View: Sampling and Private Publishing of Multidimensional Databases
Ala Eddine Laouir, Abdessamad Imine · 2024
Despite the enormous data processing capacity available in big data frameworks, obtaining appropriate and private responses to large-scale queries without revealing sensitive information is still a challenging problem. In this paper, we address the problem of combining offline sampling techniques for space efficiency in multidimensional databases and Differential Privacy (DP) to protect sensitive data. We present our framework SLIM-View, which uses a novel sampling technique relying on a bi-objective optimization to decide the best sample size and the exponential mechanism to select the best sample while ensuring privacy. Our extensive experiments demonstrate that SLIM-View outperforms existing approaches by orders of magnitude in terms of utility and scalability while ensuring the same level of privacy.