A minimal model for secure computation (extended abstract)

We consider a minimal scenario for secure computation: Parties A and B have private inputs x and y and a shared random string r. A and B are each allowed to send a single message to a third party C, from which C is to learn the value of ~ (z, y) for some function ~, but nothing else. We show that this model is surprisingly powerful: every function f can be securely computed in this fashion. If the messages are required to be of polynomial size, then we exhibit an efficient protocol for any function f computable in nondeterministic logspace. Using a computational notion of security, we exhibit efficient protocols for any polynomial-time computable function f, assuming the existence of oneway functions. The above results generalize to the case where there are more than two parties with private inputs. The minimalistic nature of our model makes it easy to transform positive results achieved in our model to other more general models of secure computation. It also gives hope for lower-bound proofs. We give an alternative characterization of our model in terms of graph embeddings, and use this to show that for most Boolean functions on O, In x O, In, the need to hide just one of the input bits from C requires a communication overhead of n bits.