Defending against Insertion-based Textual Backdoor Attacks via Attribution

Open in new window