Show simple item record

dc.contributor.authorLloret Talavera, Guillermo
dc.contributor.authorJorda, Marc
dc.contributor.authorServat, Harald
dc.contributor.authorBoemer, Fabian
dc.contributor.authorChauhan, Chetan
dc.contributor.authorTomishima, Shigeki
dc.contributor.authorShah, Nilesh N.
dc.contributor.authorPeña, Antonio
dc.contributor.otherBarcelona Supercomputing Center
dc.date.accessioned2021-05-17T10:46:03Z
dc.date.available2021-05-17T10:46:03Z
dc.date.issued2021
dc.identifier.citationLloret Talavera, G. [et al.]. Enabling homomorphically encrypted inference for large DNN models. "IEEE Transactions on Computers", 2021, 9417740.
dc.identifier.issn0018-9340
dc.identifier.urihttp://hdl.handle.net/2117/345697
dc.description.abstractThe proliferation of machine learning services in the last few years has raised data privacy concerns. Homomorphic encryption (HE) enables inference using encrypted data but it incurs 100x-10,000x memory and runtime overheads. Secure deep neural network (DNN) inference using HE is currently limited by computing and memory resources, with frameworks requiring hundreds of gigabytes of DRAM to evaluate small models. To overcome these limitations, in this paper we explore the feasibility of leveraging hybrid memory systems comprised of DRAM and persistent memory. In particular, we explore the recently-released Intel Optane PMem technology and the Intel HE-Transformer nGraph to run large neural networks such as MobileNetV2 (in its largest variant) and ResNet-50 for the first time in the literature. We present an in-depth analysis of the efficiency of the executions with different hardware and software configurations. Our results conclude that DNN inference using HE incurs on friendly access patterns for this memory configuration, yielding efficient executions.
dc.description.sponsorshipWe would like to thank Jesus Labarta from BSC and Steve Scargall from Intel for their insightful and productive comments.
dc.format.extent11 p.
dc.language.isoeng
dc.publisherInstitute of Electrical and Electronics Engineers
dc.subjectÀrees temàtiques de la UPC::Informàtica::Enginyeria del software
dc.subject.lcshEncryption of data (Computer science)
dc.subject.lcshRandom access memory.
dc.subject.lcshMemory management (Computer science)
dc.subject.lcshNeural networks
dc.subject.otherPrivacy-Preserving Machine Learning
dc.subject.otherHomomorphic Encryption
dc.subject.otherDeep Learning
dc.titleEnabling homomorphically encrypted inference for large DNN models
dc.typeArticle
dc.subject.lemacProgramari
dc.identifier.doi10.1109/TC.2021.3076123
dc.description.peerreviewedPeer Reviewed
dc.relation.publisherversionhttps://ieeexplore.ieee.org/abstract/document/9417740
dc.rights.accessOpen Access
dc.description.versionPostprint (author's final draft)
local.citation.other9417740
local.citation.publicationNameIEEE Transactions on Computers


Files in this item

Thumbnail

This item appears in the following Collection(s)

Show simple item record

All rights reserved. This work is protected by the corresponding intellectual and industrial property rights. Without prejudice to any existing legal exemptions, reproduction, distribution, public communication or transformation of this work are prohibited without permission of the copyright holder