基于迁移学习的敏感数据隐私保护方法

被引:8
作者
付玉香 [1 ]
秦永彬 [1 ,2 ]
申国伟 [1 ,2 ]
机构
[1] 贵州大学计算机科学与技术学院
[2] 贵州大学贵州省公共大数据重点实验室
基金
国家自然科学基金重大研究计划;
关键词
差分隐私; 迁移学习; 模型攻击; 敏感数据; 隐私保护;
D O I
10.16337/j.1004-9037.2019.03.006
中图分类号
TP309 [安全保密]; TP181 [自动推理、机器学习];
学科分类号
081201 ; 0839 ; 1402 ;
摘要
机器学习涉及一些隐含的敏感数据,当受到模型查询或模型检验等模型攻击时,可能会泄露用户隐私信息。针对上述问题,本文提出一种敏感数据隐私保护"师徒"模型PATE?T,为机器学习模型的训练数据提供强健的隐私保证。该方法以"黑盒"方式组合了由不相交敏感数据集训练得到的多个"师父"模型,这些模型直接依赖于敏感训练数据。"徒弟"由"师父"集合迁移学习得到,不能直接访问"师父"或基础参数,"徒弟"所在数据域与敏感训练数据域不同但相关。在差分隐私方面,攻击者可以查询"徒弟",也可以检查其内部工作,但无法获取训练数据的隐私信息。实验表明,在数据集MNIST和SVHN上,本文提出的隐私保护模型达到了隐私/实用准确性的权衡,性能优越。
引用
收藏
页码:422 / 431
页数:10
相关论文
共 30 条
[1]  
Reading digits in natural images with unsupervised feature learning. NETZER Y,WANG T,COATES A,et al. NIPS Workshop on Deep Learning and Unsupervised Feature Learning . 2011
[2]  
Calibrating Noise to Sensitivity in Private Data Analysis. Cynthia Dwork,Frank McSherry,Kobbi Nissim,Adam Smith. Theory of Cryptography Conference . 2006
[3]  
The Algorithmic Foundations of Differential Privacy. Dwork C,Roth A. Theoretical Computer Science . 2014
[4]  
Model inversion attacks that exploit confidence information and basic countermeasures. MATT F,SOMESH J,THOMAS R. ACM SIGSAC Conference on Computer and Communications Security . 2015
[5]  
Associative Domain Adaptation. Haeusser P,Frerix T,Mordvintsev A,et al. IEEE International Conference on Computer Vision . 2017
[6]  
A theory of learning from different domains[J] . Shai Ben-David,John Blitzer,Koby Crammer,Alex Kulesza,Fernando Pereira,Jennifer Wortman Vaughan. &nbspMachine Learning . 2010 (1-2)
[7]  
Deep Learning with Differential Privacy. Abadi M,Chu A,Goodfellow I,et al. ACM Sigsac Conference on Computer and Communications Security . 2016
[8]  
Membership Inference Attacks Against Machine Learning Models. R.Shokri,M.Stronati,C.Song,V.Shmatikov. 2017 IEEE Symposium on Security and Privacy (SP) . 2017
[9]  
Privacy-Preserving Deep Learning. Shokri R,Shmatikov V. Proceedings of the 22nd ACM SIGSAC Conference on Computer and Communications Security . 2015
[10]  
Ensemble methods in Machine Learning. Dietterich T G. First International Workshop on Multiple Classifier systems . 2000