Experimental Analysis of Privacy-Preserving Statistics Computation

The recent investigation of privacy-preserving data mining and other kinds of privacy-preserving distributed computation has been motivated by the growing concern about the privacy of individuals when their data is stored, aggregated, and mined for information. Building on the study of selective private function evaluation and the efforts towards practical algorithms for privacy-preserving data mining solutions, we analyze and implement solutions to an important primitive, that of computing statistics of selected data in a remote database in a privacy-preserving manner. We examine solutions in different scenarios ranging from a high speed communications medium, such as a LAN or high-speed Internet connection, to a decelerated communications medium to account for worst-case communication delays such as might be provided in a wireless multihop setting.Our experimental results show that in the absence of special-purpose hardware accelerators or practical optimizations, the computational complexity is the performance bottleneck of these solutions rather than the communication complexity. We also evaluate several practical optimizations to amortize the computation time and to improve the practical efficiency.