Paper 2020/1014

GANRED: GAN-based Reverse Engineering of DNNs via Cache Side-Channel

Yuntao Liu and Ankur Srivastava

Abstract

In recent years, deep neural networks (DNN) have become an important type of intellectual property due to their high performance on various classification tasks. As a result, DNN stealing attacks have emerged. Many attack surfaces have been exploited, among which cache timing side-channel attacks are hugely problematic because they do not need physical probing or direct interaction with the victim to estimate the DNN model. However, existing cache-side-channel-based DNN reverse engineering attacks rely on analyzing the binary code of the DNN library that must be shared between the attacker and the victim in the main memory. In reality, the DNN library code is often inaccessible because 1) the code is proprietary, or 2) memory sharing has been disabled by the operating system. In our work, we propose GANRED, an attack approach based on the generative adversarial nets (GAN) framework which utilizes cache timing side-channel information to accurately recover the structure of DNNs without memory sharing or code access. The benefit of GANRED is four-fold. 1) There is no need for DNN library code analysis. 2) No shared main memory segment between the victim and the attacker is needed. 3) Our attack locates the exact structure of the victim model, unlike existing attacks which only narrow down the structure search space. 4) Our attack efficiently scales to deeper DNNs, exhibiting only linear growth in the number of layers in the victim DNN.

Metadata
Available format(s)
PDF
Category
Applications
Publication info
Preprint. MINOR revision.
Keywords
Deep Neural NetworksReverse EngineeringCache Side-ChannelPrime+Probe
Contact author(s)
ytliu @ umd edu
History
2020-08-22: received
Short URL
https://ia.cr/2020/1014
License
Creative Commons Attribution
CC BY

BibTeX

@misc{cryptoeprint:2020/1014,
      author = {Yuntao Liu and Ankur Srivastava},
      title = {{GANRED}: {GAN}-based Reverse Engineering of {DNNs} via Cache Side-Channel},
      howpublished = {Cryptology {ePrint} Archive, Paper 2020/1014},
      year = {2020},
      url = {https://eprint.iacr.org/2020/1014}
}
Note: In order to protect the privacy of readers, eprint.iacr.org does not use cookies or embedded third party content.