Hands-On Artificial Intelligence for Cybersecurity

Deep neural network attacks

There are basically two main ways to carry out an attack against DNNs:

• White-box attacks: This type of attack presupposes the model transparency of the DNN's target, which grants the ability to directly verify the sensitivity of the response to the adversarial examples.
• Black-box attacks: Unlike the previous case, the sensitivity check of the adversarial example is implemented indirectly, not having available the configuration details of the targeted neural network; the only information available is the output values ​​returned by the neural networks to the respective inputs sent to them.

Irrespective of the type of attack, the attacker is, in any case, able to exploit some general characteristics concerning ...