Xinwei Liu

few-shot learning (1) backdoor attack (1) adversarial attack (1) adversarial perturbation (1) vision-language model (1) safety evaluation (1) trigger generation (1) privacy protection (1) adversarial prompt (1) attack success rate (1) support set (1) reasoning pathway (1) imperceptible perturbation (1) trigger optimization (1) geolocation privacy (1) multimodal large reasoning model (1) trigger feature (1) poisoned support set (1) emotional flattery (1) emotional misalignment (1)