On the Difficulty of Defending Contrastive Learning against Backdoor Attacks

Open in new window