分布式隐私保护–Logistic回归

Distributed logistic regression with differential privacy

本文关注敏感数据的隐私保护问题,开展满足差分隐私的分布式Logistic回归模型研究.通过对算法输出结果加扰动,实现分布式差分隐私.进一步,为了防止计算机信息交互过程中可能产生的隐私泄露,针对算法迭代过程加扰动的方式提出了基于Alternating Direction Method of Multipliers(ADMM)算法的分布式Logistic变量扰动算法,并给出算法的理论界估计.实验表明,所提算法可有效地处理分布式存储数据并保护其隐私.

In this paper,we focus on the privacy protection of sensitive data and develop a distributed logistic regression that satisfies differential privacy.Distributed differential privacy is achieved by perturbing the distributed algorithm output.Further,to prevent privacy leakage occurring during the computer interaction process,we propose a distributed logistic variable perturbation algorithm based on an alternating direction method of multipliers(ADMM)algorithm.Further,the theoretical bounds of the algorithms are provided.Experiments show that the proposed algorithms can effectively analyze distributed storage data and protect their privacy.