Allowing Bounded Leakage in Secure Computation: A New Application of Differential Privacy

Presented as part of the 2017 HCSS conference.

Abstract:

Secure computation allows two or more parties to perform arbitrary computations on encrypted data.  While this was purely of theoretical interest 10 years ago, today the techniques are quite practical, and the application space of secure computation is rapidly growing.  As researchers and users attempt to apply these techniques to larger data sets, a new set of challenges arise.  In our work, we explore a new trade-off between efficiency and privacy, allowing some bounded amount of leakage to be observed by the computing servers, in the form of access patterns to memory.  However, unlike much of the prior work that has made a similar tradeoff, we give provable guarantees about what is revealed, demonstrating that what is leaked in the process of computing preserves the differential privacy of the users that have contributed their data.  In this talk we will give some background on both secure computation and differential privacy, before presenting our new results that combine the techniques from these two fields. 

__

Dov Gordon is an Assistant Professor of computer science at George Mason University.  His research is in cryptography, focusing mainly on techniques for computing on encrypted data.  He received his PhD from the University of Maryland in 2010, and was a recipient of the Computing Innovations Postdoc Fellowship, which he carried to Columbia University.  He joined George Mason after spending three years at Vencore Labs as a research scientist.