Xiangyu Qi

adversarial attack (3) jailbreak attack (2) safety guardrail (2) large language model (2) deep neural network (2) adversarial robustness (2) adversarial training (1) model safety (1) responsible ai (1) ai safety (1) safety alignment (1) distribution shift (1) first-order logic (1) domain knowledge (1) backdoor attack (1) probabilistic graphical model (1) adversarial defense (1) vision-language model (1) model deployment (1) fine-tuning attack (1)