A Framework for Accelerating the Adversarial Training of Deep Neural Networks through Accuracy Gradient
This is the source code of AccelAT framework, an acceleration technique to speed up adversarial training of DNN (deep neural networks). For more details refer to the main paper published on IEEE journal in 2022: DOI - 10.1109/ACCESS.2022.3213734.
The main file is accelat.py
, to use correctly the code you have to install the packages of accelat.yml
file in your environment.
If you want to obtain the results directly on your phone you have to create a telegram bot and compile the telegram_bot.txt
file before starting.
To cite this work please use:
F. Nikfam, A. Marchisio, M. Martina and M. Shafique, "AccelAT: A Framework for Accelerating the Adversarial Training of Deep Neural Networks Through Accuracy Gradient," in IEEE Access, vol. 10, pp. 108997-109007, 2022, doi: 10.1109/ACCESS.2022.3213734.
@ARTICLE{9915577,
author={Nikfam, Farzad and Marchisio, Alberto and Martina, Maurizio and Shafique, Muhammad},
journal={IEEE Access},
title={AccelAT: A Framework for Accelerating the Adversarial Training of Deep Neural Networks Through Accuracy Gradient},
year={2022},
volume={10},
number={},
pages={108997-109007},
doi={10.1109/ACCESS.2022.3213734}}