Generative Adverserial Networks for MNIST

This is an example of generating adversarial examples to exploit the deep MNIST Convolution network. It is inspired from Intriguing Properties of Neural Networks, Explaining and Harnessing Adversarial Examples and Breaking Convnets.

Setup Instructions

git clone https://github.com/divyam3897/adversarial-examples.git

Make sure tha you have Jupyter Notebook installed. You can install Anaconda (which installs Python, Jupyter Notebook, and a bunch of other useful computing libraries) or use pip.

To install Anaconda.
If you want to install using pip, update pip with the following code (Replace pip with pip3 if using Python 3).

On Linux/Mac OS:

pip install -r requirements.txt

You should be able to run the following.

pip install jupyter

jupyter notebook

For more resources on Jupyter Notebooks, check out the following:

The image below shows 10 samples of 2 on the left, perturbation in the middle and the adversarial examples on the right