Adversarial Training

Fact

A way to train AI with trick examples, so it becomes harder to fool.

In Plain Words

Adversarial Training is like dodgeball practice with one sneaky kid. They only throw the weird bounces, so game day feels less scary.

You see it in image recognition and security tools. It helps AI resist attacks, but training takes more work.

Related Concepts

• Adversarial Example
• Data Augmentation
• Regularization

Adversarial Example
Adversarial Training uses Adversarial Examples as practice for tricky inputs.

Data Augmentation
Adversarial Training is a harsher form of Data Augmentation.

Regularization
Adversarial Training acts like Regularization by blocking weak shortcuts.