フリー問題

問題文

An attacker who can append rows to the corpus used for the next scheduled retraining inserts crafted records, so that the future model will misread one specific pattern chosen by the attacker. Which attack is that?

選択肢

1. The attack being described is adversarial evasion against the model.
2. The attack being described is training poisoning against the model.
3. The attack being described is gradient reconstruction against the model.
4. The attack being described is membership inference against the model.