Differentially Private High-Dimensional Data Publication via Markov Network

Abstract

Differentially private data publication has recently received considerable attention. However, it faces some challenges in differentially private high-dimensional data publication, such as the complex attribute relationships, the high computational complexity and data sparsity. Therefore, we propose PrivMN, a novel method to publish high-dimensional data with differential privacy guarantee. We first use the Markov model to represent the mutual relationships between attributes to solve the problem that the direction of relationship between variables cannot be determined in practical application. We then take advantage of approximate inference to calculate the joint distribution of high-dimensional data under differential privacy to figure out the computational and spatial complexity of accurate reasoning. Extensive experiments on real datasets demonstrate that our solution makes the published high-dimensional synthetic datasets more efficient under the guarantee of differential privacy.