Complete implementation of CW attack about detecting-adversarial-samples HOT 4 OPEN

Hi @a7b23,

I did not have time to convert my CW attack implementation from a mess of Jupyter Notebooks into proper code for this repository. I suggest using the Cleverhans repository, which has a very nice, clean implementation of the CW L2 attack. See "CarliniWagnerL2" class here: https://github.com/tensorflow/cleverhans/blob/master/cleverhans/attacks.py

from detecting-adversarial-samples.

If anyone would like to contribute to this repository and add the Carlini & Wagner L2 attack here, it would be greatly appreciated.

from detecting-adversarial-samples.

Hi Reuben, nice work here!
My code for ICLR 2018 paper "Characterizing adversarial subspaces using local intrinsic dimensionality", has included the BU, KD measures along with the CW attack (and other attacks). I used your code for BU and KD, and have provided a reference link to this repository. https://github.com/xingjunm/lid_adversarial_subspace_detection

from detecting-adversarial-samples.

@xingjunm thank you for using our code and for referencing the repository. At the moment, I don't have time to update this repository with your CW attack code, but I appreciate the pointer and may use it in the future. If you'd like to submit a PR, I'd be happy to review.

from detecting-adversarial-samples.