Privacy Preserving Classification via Machine Learning Model Inference on Homomorphic Encrypted Medical Data
Kristof T’Jonck, Chandrakanth R. Kancharla, Bozheng Pang, Hans Hallez, Jeroen Boydens · 2022
This paper studies the use of homomorphic encryption to preserve privacy when using machine learning classifiers. The paper compares different parameters and explores drawbacks in terms of accuracy, speed, and packet size when using encrypted data versus unencrypted data by using a client-server use case. These comparisons were performed during multiple tests on different datasets with different sizes and complexity. These tests show it is possible to do machine learning with homomorphic encrypted data without losing accuracy. However, the increased processing time, data size and communication time have to be considered.