The code for the paper: https://arxiv.org/pdf/1802.00168.pdf
Switch branches/tags
Nothing to show
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
Fool_StandardDNN.py
Fool_WNLLDNN.py
NIPS_DeepWNLL_Poster.pdf
README.md
TrainStandardDNN.py
TrainWNLLDNN.py
WNLL.py
resnet.py
utils.py
visualize_data_FooledbyStandardDNN.py
visualize_data_FooledbyWNLLDNN.py

README.md

DNN-DataDependentActivation

Code for the paper: https://arxiv.org/pdf/1802.00168.pdf

External dependency: pyflann (https://github.com/primetang/pyflann)

Place the pyflann library in your current directory

Usage

Step 1. Train the deep neural nets with softmax and WNLL activation functions

python TrainStandardDNN.py python TrainWNLLDNN.py

Step 2. Attack the trained deep neural nets

python Fool_StandardDNN.py -method fgsm -epsilon 0.02

python Fool_WNLLDNN.py -method fgsm -epsilon 0.02

The method and epsilon are adjustable, where we support fgsm, ifgsm, cwl2 attacks