Machine Learning over Encrypted Data with Fully Homomorphic Encryption
Jordan Frery & Benoit Chevallier-Mames
Video
Abstract:
Privacy enhancing technologies (PETs) have been proposed as a way to protect the privacy of data while still allowing for data analysis. This is in particular interesting for online services that handle sensitive data, such as health data, biometrics, credit scores and other personal information. In our presentation, we focus on Fully Homomorphic Encryption (FHE), a powerful tool that allows for arbitrary computations to be performed on encrypted data. FHE has received lots of attention in the past few years and has reached realistic execution times and correctness.
More precisely, we explain how we apply FHE to linear, tree-based and neural-network models. For trees (decision trees, random forests, and gradient boosted trees), we get state-of-the-art solutions over encrypted tabular data. We also describe how we handle deep learning, to already achieve promising results on some vision tasks.
Our techniques are implemented within our open-source Concrete-ML library. We show a selected set of use-cases, and demonstrate that our FHE version is very close to the unprotected version in terms of accuracy.