Research Article
Multi-party High-Dimensional Related Data Publishing via Probabilistic Principal Component Analysis and Differential Privacy
@INPROCEEDINGS{10.1007/978-3-030-96791-8_9, author={Zhen Gu and Guoyin Zhang and Chen Yang}, title={Multi-party High-Dimensional Related Data Publishing via Probabilistic Principal Component Analysis and Differential Privacy}, proceedings={Security and Privacy in New Computing Environments. 4th EAI International Conference, SPNCE 2021, Virtual Event, December 10-11, 2021, Proceedings}, proceedings_a={SPNCE}, year={2022}, month={3}, keywords={Data publishing Differential privacy High-dimensional Multi-party Probability model}, doi={10.1007/978-3-030-96791-8_9} }- Zhen Gu
Guoyin Zhang
Chen Yang
Year: 2022
Multi-party High-Dimensional Related Data Publishing via Probabilistic Principal Component Analysis and Differential Privacy
SPNCE
Springer
DOI: 10.1007/978-3-030-96791-8_9
Abstract
In this paper, we study the problem of multi-party horizontal split high-dimensional related data publishing that satisfies differential privacy. The dataset held by each party contains sensitive personal information, directly aggregating and publishing the local dataset from multiple parties will leak personal privacy. Usually, high-dimensional data are correlated, adding noise directly to the data will cause repeated noise addition and reduce the utility of the released data. To solve this problem, we proposed a method that horizontally split data publishing via probabilistic principal component analysis and differential privacy, the data owners add noise to low-dimensional data to reduce noise intake, and collaborate with a semi-trusted curator to reduce the dimensionality, finally, the data owners use the generative model of probabilistic principal component analysis to generate a synthetic dataset for publishing. The experimental results show that the synthetic dataset can maintain more efficient under the guarantee of differential privacy.
