Polynomial Approximations of ReLU for Secure CNN Inference in Homomorphic Encryption Environments
IEICE Transactions on Fundamentals of Electronics, Communications and Computer Sciences · 01.07.2025 · E108.A (7) · 977-980
DOI: 10.1587/transfun.2024eal2090 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEICE Transactions on Fundamentals of Electronics, Communications and Computer Sciences, erschienen am 01.07.2025. Verfasst von Pierpaolo AGAMENNONE. Die Arbeit wurde bisher 3-mal zitiert (Crossref) und nennt 10 Literaturangaben.
Abstract (Originalsprache)
In this paper, we introduce a novel approach to improve secure neural network inference by addressing the challenges posed by ho-momorphicencryption,specificallywithinthecontextoftheCKKSscheme.Amajorlimitationinhomomorphicencryptionistheinabilitytoefficiently handlenon-linearactivationfunctions,suchasReLU,duetotheirnon-polynomialnature.Weproposeaninnovative7th-degreepolynomialap-proximationoftheReLUfunction,generatedusingtheRemezalgorithm,whichcloselymimicsReLU’sbehaviorwhilebeingfullycompatiblewith encryptedoperations.Tofurtheroptimizeperformance,weintroducedy-namicdomainextensiontechniques,whichallowforefficientscalingof inputsduringpolynomialevaluation,significantlyreducingcomputationaloverhead.OurmethodisvalidatedusingtheMNISTdataset,demonstrating secureinferenceonencrypteddatawith97.93%accuracy,whileachievingnear-plaintextperformance.Thisworkrepresentsasignificantstepforward inthepracticalapplicationofhomomorphicencryptionforneuralnetworkinference,providingamoreefficientandaccurateapproachtoapproximat-ingnon-linearfunctionsunderencryption.
Zitieren
APA
AGAMENNONE, P. (2025). Polynomial Approximations of ReLU for Secure CNN Inference in Homomorphic Encryption Environments. IEICE Transactions on Fundamentals of Electronics, Communications and Computer Sciences, E108.A(7), 977-980. https://doi.org/10.1587/transfun.2024eal2090
Vancouver
AGAMENNONE P. Polynomial Approximations of ReLU for Secure CNN Inference in Homomorphic Encryption Environments. IEICE Transactions on Fundamentals of Electronics, Communications and Computer Sciences. 2025;E108.A(7):977-980. doi:10.1587/transfun.2024eal2090
BibTeX
@article{AGAMENNONE2025,
author = {AGAMENNONE, Pierpaolo},
title = {Polynomial Approximations of ReLU for Secure CNN Inference in Homomorphic Encryption Environments},
journal = {IEICE Transactions on Fundamentals of Electronics, Communications and Computer Sciences},
year = {2025},
volume = {E108.A},
number = {7},
pages = {977--980},
doi = {10.1587/transfun.2024eal2090}
}
EUVIMED