Paper 2017/1190

EPIC: Efficient Private Image Classification (or: Learning from the Masters)

Eleftheria Makri, Dragos Rotaru, Nigel P. Smart, and Frederik Vercauteren

Abstract

Outsourcing an image classification task raises privacy concerns, both from the image provider's perspective, who wishes to keep their images confidential, and from the classification algorithm provider's perspective, who wishes to protect the intellectual property of their classifier. We propose EPIC, an efficient private image classification system based on support vector machine (SVM) learning, which is secure against malicious adversaries. The novelty of EPIC is that it builds upon transfer learning techniques known from the Machine Learning (ML) literature and minimizes the load on the privacy-preserving part. Our solution is based on Secure Multiparty Computation (MPC), it is 34 times faster than Gazelle (USENIX 2018) --the state-of-the-art in private image classification-- and it improves the total communication cost by 50 times, while achieving a 7\% higher accuracy on CIFAR-10 dataset. When benchmarked for performance, while maintaining the same CIFAR-10 accuracy as Gazelle, EPIC is 700 times faster and the communication cost is reduced by 500 times.

Note: small changes

Metadata
Available format(s)
PDF
Publication info
Published elsewhere. Minor revision.CT-RSA 2019
Keywords
multiparty computationmachine learningtransfer learning
Contact author(s)
eleftheria makri @ esat kuleuven be
dragos rotaru @ esat kuleuven be
nigel smart @ kuleuven be
frederik vercauteren @ kuleuven be
History
2018-12-02: last of 8 revisions
2017-12-12: received
See all versions
Short URL
https://ia.cr/2017/1190
License
Creative Commons Attribution
CC BY

BibTeX

@misc{cryptoeprint:2017/1190,
      author = {Eleftheria Makri and Dragos Rotaru and Nigel P.  Smart and Frederik Vercauteren},
      title = {EPIC: Efficient Private Image Classification (or: Learning from the Masters)},
      howpublished = {Cryptology ePrint Archive, Paper 2017/1190},
      year = {2017},
      note = {\url{https://eprint.iacr.org/2017/1190}},
      url = {https://eprint.iacr.org/2017/1190}
}
Note: In order to protect the privacy of readers, eprint.iacr.org does not use cookies or embedded third party content.