Machine Learning on Encrypted Data with Homomorphic Encryption

Abstract

Homomorphic encryption is a form of encryption which allows computation on data in its encrypted form. For example, if we encrypt numbers a and b by computing Enc(a) and Enc(b), then computing Enc(a) + Enc(b) will decrypt back to a + b. This is especially powerful, since we can extend this idea to create many exciting new technologies, one of which is machine learning on encrypted data. My poster will give a brief introduction on how homomorphic encryption works and how I learned how to select parameters in order to obtain maximum efficiency for the scheme. My poster will also include an introduction to CKKS, a specific homomorphic encryption scheme which I implemented.