backdoor-learning

This dissertation addresses security and efficiency in AI by analyzing backdoor attacks in language and vision-language models, proposing detection frameworks and novel attack methods, and introducing efficient multimodal models for clinical applications.