Zhenhong Zhou

large language model (6) backdoor attack (3) safety alignment (3) jailbreak attack (2) resource consumption (2) privacy risk (1) harmful content (1) preference learning (1) adversarial defense (1) hidden state (1) synthetic datum (1) safety evaluation (1) model safety (1) training datum (1) representation space (1) black-box setting (1) model alignment (1) jailbreak defense (1) communication topology (1) adversarial machine learning (1)