Maliciously Secure Computation, Theory and Practice

Abstract

Data analytics fuels countless innovations and reveals unparalleled insights, and these benefits only grow the more data is amassed. This has resulted in the size of datasets and the compute needed to manage them becoming too resource-intensive for even large companies to handle alone, fueling the rise of cloud computing and outsourced data management. A central problem with this outsourcing is security. How can parties ensure that an untrusted cloud is accurately running the prescribed protocol? More generally, how can two parties collaborate to run a computation over joint inputs, where both inputs remain private while still delivering the correct output? This thesis focuses on answering these questions by constructing secure computation protocols with low communication & computation overhead. The protocols in this thesis include several concretely efficient constructions of private information retrieval, a functional commitment scheme for all functions, and a general two-party secure computation scheme that comes within polylogarithmic factors of the optimal communication and computation complexity. In addition to their efficiency, all protocols presented in this thesis guarantee protection against worst-case, malicious adversaries.