摘要
提出了一种能够保护数据隐私的协作式分类机制,即C2MP2(collaborative classification mechanismfor privacy-preserving),该算法利用2类样本各自的均值和协方差作为整体信息,将整体信息共享给对方,参与分类的双方,分别使用各自的隐私数据和对方的整体信息训练获得2个可以保护隐私的分类器,并由2个分类器协作得到最终的分类器.其线性模型的训练过程不仅可以保护双方数据元的隐私,还可以保护数据元的数量信息不泄露.针对测试过程的隐私保护,设计了可以保护待测样本的隐私和分类规则不泄露的安全算法.在C2MP2线性模型的基础上,分析了C2MP2和MPM(mini maxprobability machine),SVM(support vector machine)以及M4(maxi-min margin machine)在处理隐私数据方面的区别和联系.进一步使用核方法通过内积矩阵实现隐私保护的同时提高C2MP2的非线性识别能力,并通过模拟数据和标准数据集上实验检验了C2MP2线性模型和核化模型的有效性.
Privacy-preserving is becoming an increasingly important task in the Web-enabled world.Specifically we propose a novel two-party privacy-preserving classification solution called collaborative classification mechanism for Privacy-preserving（C2MP2）that is inspired from mean value and covariance matrix globally stating data location and direction,and the fact that sharing those global information with others will not disclose ones own privacy.This model collaboratively trains the decision boundary from two hyper-planes individually constructed by ones own privacy information and counter-party＇s global information.As a major contribution,we show that C2MP2 can protect both data-entries and number of entries.We describe the C2MP2 model definition,provide the geometrical interpretation,and present theoretical justifications.To guarantee the security of testing procedure,we then develop a testing algorithm based on homomorphic encryption scheme.Moreover,we show that C2MP2 can be transformed into existing minimax probability machine（MPM）,support vector machine（SVM）and maxi-min margin machine（M4）model when privacy data satisfies certain conditions.We also extend C2MP2 to a nonlinear classifier by exploiting kernel trick without privacy disclosure.Furthermore,we perform a series of evaluations on both toy data sets and real-world benchmark data sets.Comparison with MPM and SVM demonstrates the advantages of our new model in protecting privacy.
出处
《计算机研究与发展》
EI
CSCD
北大核心
2011年第6期1018-1028,共11页
Journal of Computer Research and Development
基金
国家自然科学基金重大研究计划基金项目(90820002)
中央高校基本科研业务费专项基金项目(JUDCF09034)
关键词
分类
隐私保护
协作学习
安全双方计算
支持向量机
classification
privacy-preserving
collaborative learning
secure two-party computation
SVM
