Differentially private hierarchical tree with high efficiency

Abstract

Hierarchical histogram publication can improve the accuracy of range queries by adding the same noise level to different layers, and histogram compression can also improve the accuracy of range queries with considering the inherent redundancy of real-life histograms. However, the existing hierarchical histogram publication schemes have high sensitivity without considering the inherent redundancy of real-life histograms. In addition, the existing histogram compression schemes compress the whole histogram with high sensitivity. To address these issues, we present a histogram publication scheme, which combines histogram compression and hierarchical histogram to increase the accuracy of range queries. In particular, we propose a compression method base on coarse division and dynamic budget allocation to get an efficient compressed histogram, which can decrease the sensitivity of compression method and increase the allocated privacy budget for each division in the same privacy level. The accuracy of released histogram by our proposed scheme outperforms the existing schemes. The comparison of simulation results among two latest schemes, a classic hierarchical tree scheme and our scheme show that our scheme outperforms the existing schemes.