Fully Homomorphically Encrypted Deep Learning as a Service

Paul Mayfield,Georgios Leontidis,George Onoufriou

doi:10.3390/make3040041

Paul Mayfield, Georgios Leontidis + Show 1 more

Open Access

https://doi.org/10.3390/make3040041

Copy DOI

Abstract

Fully Homomorphic Encryption (FHE) is a relatively recent advancement in the field of privacy-preserving technologies. FHE allows for the arbitrary depth computation of both addition and multiplication, and thus the application of abelian/polynomial equations, like those found in deep learning algorithms. This project investigates how FHE with deep learning can be used at scale toward accurate sequence prediction, with a relatively low time complexity, the problems that such a system incurs, and mitigations/solutions for such problems. In addition, we discuss how this could have an impact on the future of data privacy and how it can enable data sharing across various actors in the agri-food supply chain, hence allowing the development of machine learning-based systems. Finally, we find that although FHE incurs a high spatial complexity cost, the run time is within expected reasonable bounds, while allowing for absolutely private predictions to be made, in our case for milk yield prediction with a Mean Absolute Percentage Error (MAPE) of 12.4% and an accuracy of 87.6% on average.

Highlights

Homomorphic Encryption (FHE) is a structure-preserving encryption transformation [1] first appearing in 2009 [2], and having several generational advancements since to improve its efficiency, usability and speed
For the sake of consistency, all results presented were obtained on the same local area network (LAN) to prevent the effect of otherwise uncontrollable conditions and traffic over the wider area network (WAN)
This means the only significant difference between plaintext and cyphertext processing is the cost of processing it, I.E computational and spacial cost. This table shows how the network architecture despite being quite shallow is still capable of learning this problem while achieving an MAE of 0.124 no less, which means since the data is normalised between 0–1 that our Mean Absolute Percentage Error (MAPE) is 12.4%, or put another way our neural network can accurately infer the milk yield while encrypted with an accuracy of 87.6% on average

Summary

Introduction

Homomorphic Encryption (FHE) is a structure-preserving encryption transformation [1] first appearing in 2009 [2], and having several generational advancements since to improve its efficiency, usability and speed. FHE promises to enable encrypted computation of cyphertexts directly without needing the associated private keys required for decryption. The ability to process cyphertexts directly makes it possible to provide encrypted computations in particular with deep learning what we term Encrypted Deep. Deep learning lends itself to FHE since many neural networks are already formed of polynomials that are innately abelian compatible. Abelian compatibility is important when using FHE as we are limited to only abelian operations such as addition and multiplication, there are particular issues with activation functions and loss functions that are frequently incompatible by default.

Objectives

Methods

Results

Conclusion