Abstract

Homomorphic encryption (HE) is one technique that enables privacy-preserving computation. Operations (or circuit evaluations) are performed directly on homomorphically encrypted data. The result can then be decrypted. Since its introduction 46 years ago, homomorphic encryption has evolved significantly with a plethora of schemes to choose from, each with their own advantages and disadvantages. Following the invention of fully homomorphic encryption (FHE) in 2009, there has been a revolution in the field. FHE theoretically allows for evaluation of arbitrary circuits of unbounded depth. HE can be used with a variety of models, ranging from simple support vector machines (SVMs) and random forests to computationally expensive deep neural networks.