摘要
差分隐私保护模型中,非交互式数据发布是一个研究的热点。提出了一个基于朴素贝叶斯的差分隐私合成数据集发布算法。该算法首先采用朴素贝叶斯的条件独立假设来计算原数据集的联合分布,然后采用指数机制生成发布的数据集。仿真实验表明,随着隐私预算的增加,使用合成数据集训练得到的分类器在测试数据集时分类正确率逐渐提高,并且趋于稳定。
Non-interactive data releasing has been a hotspot in differential privacy preservation model.A synthesis dataset releasing algorithm based on navie bayes was proposed.This algorithm computes the joint distribution of the original dataset based on the hypothesis of conditional independences in navie bayes firstly,then employs exponential mechanismto generate the synthesis dataset.The experiment results show that the accuracy of classifiers trained by the synthesis dataset improves and tends to be stable with privacy budget increasing.
出处
《计算机科学》
CSCD
北大核心
2015年第1期236-238,共3页
Computer Science
关键词
差分隐私
朴素贝叶斯
数据发布
指数机制
Differential privacy
Navie bayes
Data release
Exponential mechanism