PyTorch implementation following algorithms:

• Fast Gradient Sign Method (FGSM) [1]
• Basic Iterative Method (BIM) [2]
• DeepFool [3]

• Python 3.5.2
• PyTorch 0.4.0
• torchvision 0.2.1
• NumPy 1.14.3

$ python main.py

• MNIST

Label: 0	Label: 1	Label: 2	Label: 3	Label: 4	Label: 5	Label: 6	Label: 7	Label: 8	Label: 9

Label: 2	Label: 8	Label: 1	Label: 2	Label: 9	Label: 3	Label: 5	Label: 2	Label: 1	Label: 7

Label: 7	Label: 8	Label: 3	Label: 2	Label: 9	Label: 3	Label: 5	Label: 2	Label: 1	Label: 7

Label: 9	Label: 8	Label: 3	Label: 8	Label: 9	Label: 3	Label: 5	Label: 8	Label: 3	Label: 7

[1] Ian J. Goodfellow, Jonathon Shlens, Christian Szegedy. Explaining and Harnessing Adversarial Examples. ICLR, 2015

[2] Alexey Kurakin, Ian J. Goodfellow, Samy Bengio. Adversarial Examples in the Physical World. arXiv, 2016

[3] Seyed-Mohsen Moosavi-Dezfooli, Alhussein Fawzi, Pascal Frossard. DeepFool: A Simple and Accurate Method to Fool Deep Neural Networks. CVPR, 2016

Kuan-Hao Huang / @ej0cl6