Scalable Mixed-Mode MPC
Radhika, Kang Yang, Jonathan Katz, Xiao Wang
IEEE Symposium on Security and Privacy 2024 · Day 1 · Continental Ballroom 6
Multiparty Computation (MPC) is a cryptographic primitive that enables several parties to jointly compute a function on their private inputs without revealing anything beyond the function's output. While traditionally MPC protocols assume either an arithmetic or a Boolean circuit representation for the function, many real-world applications, such as biometric matching, inherently require a combination of both. This necessitates efficient conversion protocols between arithmetic and Boolean shares, a process that has historically been a significant bottleneck in mixed-mode MPC, consuming up to 99% of computation time and communication bandwidth in prior state-of-the-art solutions.
AI review
This research obliterates a major bottleneck in mixed-mode MPC, transforming an O(N^2) communication problem into O(N) for dishonest majority. The novel MP-PTL protocol and clever use of homomorphic encryption packing finally make large-scale privacy-preserving computations, like biometric matching, practical for up to 128 parties, with groundbreaking speed and communication efficiency. This is a critical advancement for secure data processing.