On the Computational Complexities of Three Privacy Measures for - PowerPoint PPT Presentation

On the Computational Complexities of Three Privacy Measures for Large Networks Under Active Attack Bhaskar DasGupta Department of Computer Science University of Illinois at Chicago Chicago, IL 60607, USA dasgupta@cs.uic.edu http://www.cs.uic.edu/~dasgupta Bases on joint work with T. Chatterjee, N. Mobasheri, V. Srinivasan and I. Yero Supported by NSF grant IIS-1160995 2/14/2017 1 UIC

Network Privacy Under Active Attack nodes infected by malicious attackers (privacy loving) Users malicious attackers are interested in sensitive attributes such as • node degrees • inter-node distances • connecitivity of network 2/14/2017 UIC 2

identifying the “relevant attribute” (for this talk) distance vector from attacked nodes 2/14/2017 UIC 3

2/14/2017 UIC 4

Related Prior Concepts • Metric dimension (also called landmarks) Distance vectors must be mutually non-identical [Harary & Melter; 1976] [Khuller, Raghavachari & Rosenfeld; 1996] [Hauptmann, Schmied & Viehmann; 2012] Similar in flavor to general set cover problem • Strong metric dimension Constrained distance vectors [Oellermann & Peters-Fransen; 2012] [DasGupta & Mobasheri; 2017] Similar in flavor to the node cover problem 2/14/2017 UIC 5

Other known privacy computational models and concepts • Multi-party communication context – [Yao, 1979], [Kushilevitz, 1992] • Geometric notions of privacy – [Feigenbaum, Jaggard, Schapira, 2010], [Comi, DasGupta, Schapira, Srinivasan, 2012] • Information-theoretic – [Bar-Yehuda, Chor, Kushilevitz, Orlitsky, 1993] • Differential privacy (database retrieval context) – [Dwork, 2006] • Anonymization approach (like this talk) – [Backstrom, Dwork, Kleinberg, 2007] 2/14/2017 UIC 6

2/14/2017 UIC 7

2/14/2017 UIC 8

n is number of nodes 2/14/2017 UIC 9

2/14/2017 UIC 10

n is number of nodes 2/14/2017 UIC 11

UIC to be discussed next 2/14/2017 12

n is number of nodes k=1 2/14/2017 UIC 13

2/14/2017 UIC 14

2/14/2017 UIC 15