الگوی درِ پشتی

تعریف در منبع (انگلیسی)

A transformation or insertion applied to a data sample that triggers an adversary-specified behaviour in a model subject to a backdoor poisoning attack.