COSIC seminar "Optimized Preprocessing for Maliciously..." (Marc Rivinius, University of Stuttgart) скачать в хорошем качестве

COSIC seminar "Optimized Preprocessing for Maliciously..." (Marc Rivinius, University of Stuttgart)

COSIC seminar – Optimized Preprocessing for Maliciously Secure MPC: Matrix Multiplications, Convolutions, and More – Marc Rivinius (University of Stuttgart) Preprocessing for MPC is a well established technique to separate protocols into a fast online phase and a slower input-independent offline phase. Specialized preprocessing for certain operations not only greatly speeds up the online phase but also the offline phase, e.g., we can perform preprocessing specific to matrix multiplications instead of breaking everything down into individual scalar multiplications for orders of magnitude less communication and faster computation. In this talk, we present our recent contributions to improving preprocessing for actively secure MPC. (1) We show how to avoid the overhead of the previous “sacrificing” technique that overproduces preprocessing material to later check for consistency. (2) We improve the state-of-the-art for generating preprocessing material for matrix multiplications. (3) We introduce the first actively secure protocol to directly preprocess convolutions instead of emulating them with matrix multiplications.