eduzhai > Applied Sciences > Engineering >

Hardware Accelerator for Adversarial Attacks on Deep Learning Neural Networks

  • king
  • (0) Download
  • 20210506
  • Save

... pages left unread,continue reading

Document pages: 8 pages

Abstract: Recent studies identify that Deep learning Neural Networks (DNNs) arevulnerable to subtle perturbations, which are not perceptible to human visualsystem but can fool the DNN models and lead to wrong outputs. A class ofadversarial attack network algorithms has been proposed to generate robustphysical perturbations under different circumstances. These algorithms are thefirst efforts to move forward secure deep learning by providing an avenue totrain future defense networks, however, the intrinsic complexity of themprevents their broader usage.In this paper, we propose the first hardware accelerator for adversarialattacks based on memristor crossbar arrays. Our design significantly improvesthe throughput of a visual adversarial perturbation system, which can furtherimprove the robustness and security of future deep learning systems. Based onthe algorithm uniqueness, we propose four implementations for the adversarialattack accelerator ($A^3$) to improve the throughput, energy efficiency, andcomputational efficiency.

Please select stars to rate!


0 comments Sign in to leave a comment.

    Data loading, please wait...