Paddle/adversarial

Advbox

Advbox is a Python toolbox to create adversarial examples that fool neural networks. It requires Python and paddle.

How to use

1. train a model and save it's parameters. (like fluid_mnist.py)
2. load the parameters which is trained in step1, then reconstruct the model.(like mnist_tutorial_fgsm.py)
3. use advbox to generate the adversarial sample.