Verifying Computations with (Private) State

Nov 11, 2015

Download: MP4 Video Size: 138.7MB  
Watch on YouTube

Abstract

Is it possible for Alice to compute a result and for Bob to be convinced of its correctness without having to reexecute the computation? What if the computation is performed over sensitive data that Bob is not allowed to see due to privacy concerns? Recent work on proof-based verifiable computation has brought these goals much closer to practicality. In this talk, I will present two implemented systems that incorporate verifiable computation in order to build realistic applications. The first, Pantry, enables a user to outsource a general-purpose computation to a potentially faulty cloud provider and yet verify that the computation was performed correctly. Unlike prior efforts, Pantry allows verifiable computations to operate on remotely-stored data, opening the way to a wide variety of uses such as MapReduce jobs and database queries.

The second system, VerDP, aims to resolve the conflict in many research studies between the verifiability of the results and the privacy of the study participants. VerDP accepts queries over sensitive data that are written in a domain-specific language and processes them only if a) it can certify that the result will not compromise individuals' privacy, and if b) it can prove the integrity of the result to the public. Experimental evaluation shows that VerDP can successfully process several types of useful queries, and that the cost of generating and verifying the proofs is practical.