AdversarialAttackMethod
Enums · AiDotNet.Enums
An adversarial attack that robustness testing can run against a trained model.
For Beginners
Each value is a different way of searching for a small change to an input that makes the model answer wrongly. Testing against several of them gives a more honest picture of robustness.
Fields
| Field | Summary |
|---|---|
AdversarialAttackMethod AutoAttack |
AutoAttack (Croce and Hein 2020): an ensemble that keeps the strongest result of its component attacks. |
AdversarialAttackMethod CarliniWagner |
The Carlini-Wagner L2 attack (Carlini and Wagner 2017): an optimization for the smallest misclassifying change. |
AdversarialAttackMethod Fgsm |
Fast Gradient Sign Method (Goodfellow et al. |
AdversarialAttackMethod Pgd |
Projected Gradient Descent (Madry et al. |
int value__ |