Efficient Maliciously Secure Two Party Computation for Mixed Programs

We propose a new approach for practical secure two-party computation (2PC) achieving security in the presence of malicious adversaries. Given a program to compute, the idea is to identify subcomputations that depend on only one or neither of the parties’ private inputs. Such computations can be secured at significantly lower cost, using different protocol paradigms for each case. We then show how to securely connect these subprotocols together, and with standard 2PC yielding our new approach for 2PC for mixed programs. Our empirical evaluations confirm that the mixed-2PC approach outperforms state-of-the-art monolithic 2PC protocols for most computations.