1 paper
Qiyuan Wang, Yao Li, Raymond K. W. Wong
Despite recent progress in backdoor attacks, existing methods remain susceptible to post-training defenses that erase the backdoor through fine-tuning or pruning. We revisit the co…