Closed aamir-mustafa closed 5 years ago
Thank you for the submission. From a quick glance at your code, I could not find an implementation of the robustml interface. Could you please implement the model, and then we can add your paper to the website?
Also, we require claims to be phrased in terms of the perturbation bound, not based on the particular attack being used (many papers talk about why this is a reasonable thing to do, see e.g. this for one explanation).
Thanks, I have uploaded the python script, "robust_ml.py" to my Git Repository.
Merged and deployed in 1757daf213c761291e1c4d443caa3cee3a38f75e.
Name: Adversarial Defense by Restricting the Hidden Space of Deep Neural Networks
Authors: Aamir Mustafa, Salman Khan, Munawar Hayat, Roland Goecke, Jianbing Shen, Ling Shao
Paper: https://arxiv.org/abs/1904.00887
Code: https://github.com/aamir-mustafa/pcl-adversarial-defense
Venue: Under Submission
Does the code implement the robust-ml API and include pre-trained models: yes
Dataset: CIFAR-10
Threat model: robust_model.pth.tar , L_inf (Epsilon = 8/255)
Natural accuracy: 90.62%
Claims: L_inf (Epsilon = 8/255), 32.32% against PGD