Paper 2023/503
Neural Network Quantisation for Faster Homomorphic Encryption
Abstract
Homomorphic encryption (HE) enables calculating on encrypted data, which makes it possible to perform privacy- preserving neural network inference. One disadvantage of this technique is that it is several orders of magnitudes slower than calculation on unencrypted data. Neural networks are commonly trained using floating-point, while most homomorphic encryption libraries calculate on integers, thus requiring a quantisation of the neural network. A straightforward approach would be to quantise to large integer sizes (e.g., 32 bit) to avoid large quantisation errors. In this work, we reduce the integer sizes of the networks, using quantisation-aware training, to allow more efficient computations. For the targeted MNIST architecture proposed by Badawi et al., we reduce the integer sizes by 33% without significant loss of accuracy, while for the CIFAR architecture, we can reduce the integer sizes by 43%. Implementing the resulting networks under the BFV homomorphic encryption scheme using SEAL, we could reduce the execution time of an MNIST neural network by 80% and by 40% for a CIFAR neural network.
Metadata
- Available format(s)
-
PDF
- Category
- Applications
- Publication info
- Published elsewhere. Major revision. 2023 IEEE 29th International Symposium on On-Line Testing and Robust System Design
- DOI
- 10.1109/IOLTS59296.2023.10224890
- Keywords
- convolutional neural networksquantisationprivacy-preserving machine learningfully homomorphic encryption
- Contact author(s)
-
wouter legiest @ esat kuleuven be
furkan turan @ esat kuleuven be
michiel vanbeirendonck @ esat kuleuven be
janpieter danvers @ esat kuleuven be
ingrid verbauwhede @ esat kuleuven be - History
- 2023-08-29: last of 3 revisions
- 2023-04-07: received
- See all versions
- Short URL
- https://ia.cr/2023/503
- License
-
CC BY
BibTeX
@misc{cryptoeprint:2023/503,
author = {Wouter Legiest and Furkan Turan and Michiel Van Beirendonck and Jan-Pieter D'Anvers and Ingrid Verbauwhede},
title = {Neural Network Quantisation for Faster Homomorphic Encryption},
howpublished = {Cryptology {ePrint} Archive, Paper 2023/503},
year = {2023},
doi = {10.1109/IOLTS59296.2023.10224890},
url = {https://eprint.iacr.org/2023/503}
}