Hands-On Artificial Intelligence for Cybersecurity

Adversarial examples against FaceNet

An example of an attack that uses adversarial examples to deceive a CNN implementing a facial-recognition model is contained in the CleverHans library (under the examples directory; it is freely available for download at https://github.com/tensorflow/cleverhans/blob/master/examples/facenet_adversarial_faces/facenet_fgsm.py. The example code is released under the MIT license at https://github.com/tensorflow/cleverhans/blob/master/LICENSE).

The example code shows how to perform an adversarial attack against the FaceNet library, using the FGSM method, obtaining an accuracy in excess of 99%.

Here is the code for the adversarial attack example against the facial-recognition model implemented by the FaceNet ...