针对传统隐私保护机器学习方案抵抗对抗攻击能力较弱的特点,提出一种基于格雷码置乱和分块混沌置乱的医学影像加密方案(Gray+block chaotic scrambling optimized for medical image encryption,GBCS),并应用于隐私保护的分类挖掘。首...针对传统隐私保护机器学习方案抵抗对抗攻击能力较弱的特点,提出一种基于格雷码置乱和分块混沌置乱的医学影像加密方案(Gray+block chaotic scrambling optimized for medical image encryption,GBCS),并应用于隐私保护的分类挖掘。首先对图像进行位平面切割;然后,对图像不同位平面进行格雷码置乱后再进行分块,在分块的基础上分别进行混沌加密;最后通过深度网络对加密后的图像进行分类学习。通过在公开乳腺癌和青光眼数据集上进行交叉验证仿真实验,对GBCS的隐私保护与分类性能进行量化分析,并从图像直方图、信息熵和对抗攻击能力等指标考虑其安全性。实验结果表明医学图像在GBCS加密前后的性能差距在可接受范围内,方案能更好地平衡性能与隐私保护的矛盾,能有效抵御对抗样本的攻击,验证了本文方法的有效性。展开更多
CART(Classification And Regression Tree,分类回归树)是一种准确率和效率都较高的数据挖掘算法,它支持离散型和连续型的数据分类,但无法适用于对加密的隐私云数据进行分类.因此提出PPCART(Privacy-preserving CART,隐私保护的分类回归...CART(Classification And Regression Tree,分类回归树)是一种准确率和效率都较高的数据挖掘算法,它支持离散型和连续型的数据分类,但无法适用于对加密的隐私云数据进行分类.因此提出PPCART(Privacy-preserving CART,隐私保护的分类回归树),该算法利用同态加密特性对CART算法做了相应的改善,使之在保持CART原有准确率和相对较好执行效率的情况下能分类加密云数据,避免了在半诚实模型下的分类过程中真实数据的泄露.经过安全分析和实验测试表明,PPCART可显著提高传统CART算法的安全性,且具有接近于它的执行时间.展开更多
Privacy-preserving data publishing (PPDP) is one of the hot issues in the field of the network security. The existing PPDP technique cannot deal with generality attacks, which explicitly contain the sensitivity atta...Privacy-preserving data publishing (PPDP) is one of the hot issues in the field of the network security. The existing PPDP technique cannot deal with generality attacks, which explicitly contain the sensitivity attack and the similarity attack. This paper proposes a novel model, (w,γ, k)-anonymity, to avoid generality attacks on both cases of numeric and categorical attributes. We show that the optimal (w, γ, k)-anonymity problem is NP-hard and conduct the Top-down Local recoding (TDL) algorithm to implement the model. Our experiments validate the improvement of our model with real data.展开更多
文摘针对传统隐私保护机器学习方案抵抗对抗攻击能力较弱的特点,提出一种基于格雷码置乱和分块混沌置乱的医学影像加密方案(Gray+block chaotic scrambling optimized for medical image encryption,GBCS),并应用于隐私保护的分类挖掘。首先对图像进行位平面切割;然后,对图像不同位平面进行格雷码置乱后再进行分块,在分块的基础上分别进行混沌加密;最后通过深度网络对加密后的图像进行分类学习。通过在公开乳腺癌和青光眼数据集上进行交叉验证仿真实验,对GBCS的隐私保护与分类性能进行量化分析,并从图像直方图、信息熵和对抗攻击能力等指标考虑其安全性。实验结果表明医学图像在GBCS加密前后的性能差距在可接受范围内,方案能更好地平衡性能与隐私保护的矛盾,能有效抵御对抗样本的攻击,验证了本文方法的有效性。
文摘CART(Classification And Regression Tree,分类回归树)是一种准确率和效率都较高的数据挖掘算法,它支持离散型和连续型的数据分类,但无法适用于对加密的隐私云数据进行分类.因此提出PPCART(Privacy-preserving CART,隐私保护的分类回归树),该算法利用同态加密特性对CART算法做了相应的改善,使之在保持CART原有准确率和相对较好执行效率的情况下能分类加密云数据,避免了在半诚实模型下的分类过程中真实数据的泄露.经过安全分析和实验测试表明,PPCART可显著提高传统CART算法的安全性,且具有接近于它的执行时间.
基金supported in part by Research Fund for the Doctoral Program of Higher Education of China(No.20120009110007)Program for Innovative Research Team in University of Ministry of Education of China (No.IRT201206)+3 种基金Program for New Century Excellent Talents in University(NCET-110565)the Fundamental Research Funds for the Central Universities(No.2012JBZ010)the Open Project Program of Beijing Key Laboratory of Trusted Computing at Beijing University of TechnologyBeijing Higher Education Young Elite Teacher Project(No. YETP0542)
文摘Privacy-preserving data publishing (PPDP) is one of the hot issues in the field of the network security. The existing PPDP technique cannot deal with generality attacks, which explicitly contain the sensitivity attack and the similarity attack. This paper proposes a novel model, (w,γ, k)-anonymity, to avoid generality attacks on both cases of numeric and categorical attributes. We show that the optimal (w, γ, k)-anonymity problem is NP-hard and conduct the Top-down Local recoding (TDL) algorithm to implement the model. Our experiments validate the improvement of our model with real data.