Communication and Round Balanced Oblivious FSM Evaluation

Abstract

Privacy is a major issue in the age of the Internet. Many advances are being made in Cryptography regarding performing computations over private data, both in homomorphic encryption, multi-party computation, and applications that put these to use. Herein we present a multiparty protocol for the private evaluation of a finite state machine. We motivate this by noting that many features can be extracted from text using the finite state transducer, an easy extension of the general FSM. For example, this protocol could be used as the feature extraction phase of an end-to-end private machine learning algorithm over text inputs. Our protocol(s) build on those previously developed by offering a different balance between communication, computation and rounds. Notably, we offer a 2-round protocol with fairly low communication. The previous constant round protocol had higher communication, and the previous low communication protocol had rounds proportional to the input size. A very computation efficient version is provided if a third party is available who is not trusted beyond non-collusion. And a more computationally intensive version removes the need for this helper.