|
joint work with Seny Kamara |
|
Results: We construct an encryption scheme with homomorphic properties that allow computation of certain classes of functions on encrypted data, which can be also efficiently executed in parallel. In the setting of delegated computation we show how to use our encryption scheme to support various operations for the MapReduce framework that enable evaluation of functions in NC0 with locality 1 to polylog(k) for a security parameter k. In addition to the input privacy our scheme for parallel outsourced computation supports also privacy of the evaluated function. In a separate work we show how to use techniques from multiparty computation and secret sharing to protect against collocation attacks in the setting of multi-tenant cloud enviroments. The threat scenario for this setting assumes trust in the cloud provider (does not provide privacy with respect to the cloud) and aims to protect against other malicious clients who can attempt to mount attacks against the users collocated on the same physcical machine in cloud environment. |
|
|
Resources: Soon to come |
Motivation: In the setting of outsourced computation
it is often the
case that the computational resources behind the service consist of a
cluster of machines. This is especially true
when the inputs for the computation are massive databases such as
the one describing
graphs of large social network, large corpora for training of machine
learning algorithms, etc. In this case the computation is beyond the capabilities of a single machine and needs to be performed on a cluster of machines. Thus a secure computation protocol that can take maximal advantage of computational resources in such an environment should easily lend itself to parallelization.