1 paper · 1 filter
Zhenyu Yu
Backdoor unlearning aims to remove a malicious trigger behavior from a deployed model while preserving clean utility. We study the update-free inference-time setting, where model p…