Low Latency Privacy Preserving Inference

Alon Brutzkus, Oren Elisha, Ran Gilad-Bachrach

Abstract: When applying machine learning to sensitive data one has to balance between accuracy, information leakage, and computational-complexity. Recent studies have shown that Homomorphic Encryption (HE) can be used for protecting against information leakage while applying neural networks. However, this comes with the cost of limiting the kind of neural networks that can be used (and hence the accuracy) and with latency of the order of several minutes even for relatively simple networks. In this study we improve on previous results both in the kind of networks that can be applied and in terms of the latency. Most of the improvement is achieved by novel ways to represent the data to make better use of the capabilities of the encryption scheme.

Keywords: privacy, classification, homomorphic encryption, neural networks

TL;DR: This work presents methods, combining neural-networks and encryptions, to make predictions while preserving the privacy of the data owner with low latency

Code: [ microsoft/CryptoNets](https://github.com/microsoft/CryptoNets)

Data: [CIFAR-10](https://paperswithcode.com/dataset/cifar-10)

Community Implementations: [ 1 code implementation](https://www.catalyzex.com/paper/low-latency-privacy-preserving-inference/code)