High-Resolution Convolutional Neural Networks on Homomorphically Encrypted Data via Sharding Ciphertexts

TL;DR

This paper advances homomorphic encryption techniques to enable high-resolution CNN inference on encrypted data, achieving significant speedups and high accuracy on ImageNet and CIFAR-10 datasets.

Contribution

It introduces methods for evaluating larger, multi-channel images homomorphically, simplifies the image format, and improves inference efficiency and accuracy.

Findings

Encrypted ImageNet inference with 80.2% top-1 accuracy.

Homomorphic evaluation on CIFAR-10 with 98.3% accuracy.

Achieved 4.6-6.5x speedup in encrypted inference.

Abstract

Recently, Deep Convolutional Neural Networks (DCNNs) including the ResNet-20 architecture have been privately evaluated on encrypted, low-resolution data with the Residue-Number-System Cheon-Kim-Kim-Song (RNS-CKKS) homomorphic encryption scheme. We extend methods for evaluating DCNNs on images with larger dimensions and many channels, beyond what can be stored in single ciphertexts. Additionally, we simplify and improve the efficiency of the recently introduced multiplexed image format, demonstrating that homomorphic evaluation can work with standard, row-major matrix packing and results in encrypted inference time speedups by $4.6-6.5\times$. We also show how existing DCNN models can be regularized during the training process to further improve efficiency and accuracy. These techniques are applied to homomorphically evaluate a DCNN with high accuracy on the high-resolution ImageNet dataset, achieving $80.2\%$ top-1 accuracy. We also achieve an accuracy of homomorphically evaluated CNNs on the CIFAR-10 dataset of $98.3\%$.