Abstract

Homomorphic encryption (HE) and multi-party computation (MPC) protocols can be slow in practice. However, they can be accelerated by various accelerators such as multi-core CPUs, GPUs (Graphics Processing Units), FPGAs (Field-Programmable Gate Arrays), and ASICs (Application-Specific Integrated Circuits) to make them practical for artificial intelligence (AI) applications. We discuss the bottlenecks in these protocols and various ways of mitigating them. We then discuss recent works that mitigate bottlenecks with accelerators and enable use in AI applications.