Differentially Private Bayesian Programming

被引:22
作者
Barthe, Gilles [1 ]
Farina, Gian Pietro [2 ]
Gaboardi, Marco [2 ]
Arias, Emilio Jesus Gallego [3 ]
Gordon, Andy [4 ]
Hsu, Justin [5 ]
Strub, Pierre-Yves [1 ]
机构
[1] IMDEA Software, Madrid, Spain
[2] Univ Buffalo, Buffalo, NY USA
[3] CRI Mines ParisTech, Fontainebleau, France
[4] Microsoft Res, Cambridge, England
[5] Univ Penn, Philadelphia, PA 19104 USA
来源
CCS'16: PROCEEDINGS OF THE 2016 ACM SIGSAC CONFERENCE ON COMPUTER AND COMMUNICATIONS SECURITY | 2016年
关键词
D O I
10.1145/2976749.2978371
中图分类号
TP [自动化技术、计算机技术];
学科分类号
0812 ;
摘要
We present PrivInfer, an expressive framework for writing and verifying differentially private Bayesian machine learning algorithms. Programs in PrivInfer are written in a rich functional probabilistic programming language with constructs for performing Bayesian inference. Then, differential privacy of programs is established using a relational refinement type system, in which refinements on probability types are indexed by a metric on distributions. Our framework leverages recent developments in Bayesian inference, probabilistic programming languages, and in relational refinement types. We demonstrate the expressiveness of PrivInfer by verifying privacy for several examples of private Bayesian inference.
引用
收藏
页码:68 / 79
页数:12
相关论文
共 40 条
[1]  
Adams R., 2015, CORR
[2]  
[Anonymous], LICS
[3]  
[Anonymous], 2008, P 24 C UNC ART INT
[4]  
Barthe G., 2016, LICS
[5]  
Barthe G., 2014, CSF
[6]  
Barthe G., 2016, ACM SIGLOG NEWS
[7]  
Barthe G., 2013, ICALP
[8]  
Barthe G., 2015, POPL
[9]  
Barthe G., 2013, CSF
[10]   Probabilistic relational reasoning for differential privacy [J].
Barthe, Gilles ;
Koepf, Boris ;
Olmedo, Federico ;
Zanella Beguelin, Santiago .
ACM SIGPLAN NOTICES, 2012, 47 (01) :97-109