Skip to content

Latest commit

 

History

History
12 lines (8 loc) · 595 Bytes

README.md

File metadata and controls

12 lines (8 loc) · 595 Bytes

Fast Gradient Sign Method

This is the implementation (in PyTorch) of the method proposed in the paper: Explaining and Harnessing Adversarial Examples, for generating adversarial examples. The implementation is over the MNIST dataset.

Results

Accuracy of the network w/o adversarial attack on the 10000 test images: 97 %

Accuracy of the network with adversarial attack on the 10000 test images: 14 %

Number of misclassified examples (as compared to clean predictions): 8374/10000

Please check the iPython notebook for visualization of the results.