In order to strengthen the ties and academic exchanges between the City University of Macau (CityU) and the Minzu University of China (MUC), and to explore the security and privacy protection of big models and other cutting-edge directions, Professor Zhu Tianqing from the Faculty of Data Science of the City University of Macau (CityU) led a team of faculty members to visit the Hanian International College  of the Minzu University of China on the 17th of December, 2024 and to give a series of four lectures on this topic. The seminars aimed to help undergraduate students to enhance their research capabilities and to understand the latest progress and future development of big model security and privacy protection. 

  

Professor Zhu Tianqing gave a speech entitled ‘Unique Security and Privacy Threats of Large Language Model’. Professor Zhu pointed out that with the continuous progress of artificial intelligence technology, large language models have achieved remarkable results in the field of natural language processing. However, these models also face unique privacy and security concerns in scenarios such as pre-training, fine-tuning, retrieval of augmented generative systems, deployment and LLM-based agents. He emphasised the importance of studying these threat models, and proposed targeted countermeasures that provide useful references for the development of related fields. 

  

Assistant Professor Geng-Shen Wu gave a detailed presentation on ‘Examples of Deep Learning Applications in Medical Image Processing’. He shared his team's latest research results in the field of MRI accelerated reconstruction and medical image segmentation. The team has successfully achieved accelerated reconstruction of MRI images through an anisotropic diffusion-based generative adversarial network and a multi-scale cold diffusion model based on K-space. Meanwhile, the new segmentation network and a series of image processing modules proposed by them have significantly improved the accuracy and efficiency of medical image segmentation. Professor Wu also discussed cutting-edge medical imaging technologies such as real-time image fusion, and looked forward to the potential and development direction of further integration between deep learning and medical imaging. 

  

Assistant Professor Wong Chi Fong gave a talk entitled ‘Topological Data Analysis and Its Financial Applications’. He introduced the application of topological data analysis (TDA), an emerging mathematical tool, in the financial field. By calculating the topological nature of stock market data, deep-seated features and signals can be extracted that are not easily detected by general machine learning models. According to Professor Wong, this approach can help to better understand the nature of the market, improve investment returns and manage risk. 

  

Assistant Professor Cui Sanshuai gave a presentation on ‘Spoofing Detection and Counter Attacks in Audiovisual Multimedia’. He pointed out that with the rapid development of multimedia technology and mobile Internet, it has become easier and more covert to spoof audiovisual multimedia messages. This poses a serious threat to the fields of copyright, identification and forensics. Professor Cui introduced the latest research results of his team in audiovisual multimedia spoofing detection, and discussed the security of audiovisual multimedia content from the perspective of counter samples. He emphasised the importance of strengthening research in this area for copyright protection, information security and justice.