AI Security & Adversarial Attacks › Adversarial Examples & Attacks

How does Adversarial Examples & Attacks work

Beginner

Answer

Small, carefully calculated changes are added to an input, often invisible to a human eye, that cause the model to misclassify it while a person would not notice anything unusual.
Real-world example Researchers showed that adding tiny, almost invisible changes to a stop sign image could trick a self driving car's AI into misreading it.

Common follow-ups: What is AI Security & Adversarial Attacks, How is Adversarial Examples & Attacks evaluated in practice, What tools are commonly used for Adversarial Examples & Attacks