Test your understanding before moving to Section 1.4.
Section Quiz
Q1. A backdoor poisoning attack requires what percentage of training data to be corrupted to be effective?
Q2. Model memorisation is most concerning when:
Q3. Which question is most important to ask a third-party AI vendor from a security perspective?
Q4. In the BadNets research (Gu et al., 2017), what caused the poisoned road-sign classifier to misbehave?
Q5. What is a "canary input", as used in defending against training data poisoning?
Q6. Which of the following is NOT one of the four dimensions of AI supply chain risk covered in this lesson?
