Harnessing ML privacy by design through crossbar array non-idealities

Md Shohidul Islam, Sankha B. Dutta, Andres Marquez, Ihsen Alouani, Khaled N. Khasawneh

Research output: Chapter in Book/Report/Conference proceedingConference contribution

6 Downloads (Pure)

Abstract

Deep Neural Networks (DNNs), handling compute- and data-intensive tasks, often utilize accelerators like Resistive- switching Random-access Memory (RRAM) crossbar for energy- efficient in-memory computation. Despite RRAM's inherent non- idealities causing deviations in DNN output, this study transforms the weakness into strength. By leveraging RRAM non-idealities, the research enhances privacy protection against Membership Inference Attacks (MIAs), which reveal private information from training data. RRAM non-idealities disrupt MIA features, increasing model robustness and revealing a privacy-accuracy tradeoff. Empirical results with four MIAs and DNNs trained on different datasets demonstrate significant privacy leakage reduction with a minor accuracy drop (e.g., up to 2.8% for ResNet-18 with CIFAR-100).
Original languageEnglish
Title of host publicationDesign, Automation & Test in Europe Conference & Exhibition (DATE 2024): proceedings
PublisherInstitute of Electrical and Electronics Engineers Inc.
Number of pages2
ISBN (Electronic)9783981926385
ISBN (Print)9798350348606
Publication statusPublished - 10 Jun 2024
EventDesign, Automation and Test in Europe Conference 2024 - Valencia, Spain
Duration: 25 Mar 202427 Mar 2024
https://www.date-conference.com/

Publication series

NameDesign, Automation & Test in Europe Conference & Exhibition: proceedings
ISSN (Print)1530-1591
ISSN (Electronic)1558-1101

Conference

ConferenceDesign, Automation and Test in Europe Conference 2024
Abbreviated titleDATE 2024
Country/TerritorySpain
City Valencia
Period25/03/202427/03/2024
Internet address

Keywords

  • harnessing ML privacy
  • by design
  • crossbar array non-idealities

Fingerprint

Dive into the research topics of 'Harnessing ML privacy by design through crossbar array non-idealities'. Together they form a unique fingerprint.

Cite this