Discover gradient-based adversarial attacks in new Academy Module

The

AI Evasion - First Order Attacks

module provides a deep exploration of the mechanics of gradient-based evasion techniques that target neural network classifiers. Through hands-on exercises, you’ll learn how to implement and analyze key evasion techniques, including the Fast Gradient Sign Method (FGSM), Iterative FGSM (I-FGSM), and DeepFool, to develop a better understanding of how neural network classifiers can be manipulated and how to defend against such attacks.

Key learning outcomes include:

How to evaluate model robustness and apply defensive strategies such as adversarial training and input filtering
Mechanics of first-order evasion attacks, including how gradients are used to manipulate model predictions
Applying the concept of input perturbation norms to measure the size and impact of adversarial changes
Assessing the gap between model accuracy and true robustness