As the model structure becomes more complex and the number of model parameters continues to increase, the processing power of the model is further enhanced, and it is widely applied in numerous fields such as images, voice, and text. However, both internally and externally, the model faces multiple security threats. Internally, the increase in the number of model parameters leads to emerging risks such as insufficient explainability and new biases. Externally, large models face various threats that affect model performance and violate privacy data. Our research goal is to design effective methods to identify the vulnerabilities of the model, and to develop efficient methods to protect the model from the aforementioned threats.

The incredible potential of large models in Artificial Intelligence Generated Content (AIGC), including cutting-edge technologies like Large Language Models (LLMs) and Diffusion Models, have revolutionized a wide range of applications, spanning natural language processing, content generation, creative arts, and more. However, large model size, and high memory and computational requirements present formidable challenges. We aim to tackle these hurdles head-on and make these advanced AI technologies more practical, democratizing access to these future-changing technologies for everyone.

The rapid development of Artificial Intelligence (AI) has led to concerns about the misuse of advanced technologies, particularly in the emergence of DeepFakes. DeepFakes use deep learning techniques to create highly convincing fake media, posing risks such as spreading misinformation and manipulating public opinion. As DeepFake technology becomes more sophisticated, the primary objective of our research is to develop effective strategies to defense (e.g., DeepFake Passive Detection, DeepFake Proactive Defense) and countermeasures to evade (e.g., DeepFake Evasion Attack, Realistic Creation).