CyberCertPrepAI Security FundamentalsAdversarial ML & Model AttacksEASYFree question
A university lecturer asks students: which of these is NOT a standard Lp-norm used to bound adversarial perturbation magnitude?
- A.L-infinity norm (maximum absolute change per feature)
- B.L3 norm (sum of cubed absolute values)
- C.L2 norm (Euclidean distance of the perturbation vector)
- D.L0 norm (number of modified features)
Why B is correct
Standard adversarial ML literature uses L-infinity (maximum per-feature change), L2 (Euclidean distance), and L0 (number of changed features) norms to bound perturbations. The L3 norm (sum of cubed absolute values) is not a standard adversarial perturbation metric and is not used in established attack or defense methods.
Know someone studying for AI Security Fundamentals? Send them this one.