In this repository, you can find two jupyter notebooks where the adverserial samples are generated step by step using FGSM attack and training two different neural network architectures(LeNet-5 and VGG-Net) with them. They are explained in detail in the notebooks:

1. Generating Adversarial Samples Using FGSM Attack
2. Training of the models using German Sign Classification dataset

In the presentation the results are compared and analyzed.