Abstract: Backdoor attacks embed an attacker-chosen pattern into inputs to cause model misclassification. This security threat to machine learning has been a long concern. There are a number of ...