一种基于隐树模型的满足差分隐私的高维数据发布算法

Latent Tree Model Based Differentially Private High-dimension Data Publishing Algorithm

针对满足差分隐私的高维数据发布问题,我们提出一种基于隐树模型的满足差分隐私的高维数据发布算法.该算法由隐变量生成、隐树结构学习、隐树参数学习和数据生成四个阶段组成.特别地,在该算法中,为了在对显变量进行分组并生成隐变量的过程中保护隐私,我们提出一种满足差分隐私的隐变量生成方法.此外,为了在构建隐树的过程中保护隐私,我们提出了一种满足差分隐私的隐树模型结构学习方法.分析结果表明本文提出的算法满足ε-差分隐私.实验结果表明,与现有算法相比,所提出的算法可以获得更好的数据效用.

To solve the differentially private high-dimensional data publishing problem,we present a latent tree model based differentially private data publishing algorithm.This algorithm consists of four phases：latent variable generation,latent tree structure learning,latent tree parameter learning and data generation.In particular,in this algorithm,to privately group the observable variables and generate the latent variables,we propose a differentially private latent variable generation method.Moreover,to privately construct the structure of the latent tree,we propose a differentially private structure learning method for the latent tree construction.Formal analysis shows that the proposed algorithm satisfies ε-differential privacy.Experimental results demonstrate that the proposed algorithm can obtain better data utility than existing algorithms.