Privacybeschermend en Gedistribueerd Machinaal Leren

Artificial Intelligence has made significant breakthroughs in pattern recognition, generative modeling, and other areas of data science. These technologies rely on large amounts of data as learning examples. An essential aspect of data is the privacy associated with the data producers. When learning from data examples, AI models can inadvertently expose private information, thereby raising significant concerns. My research focuses on the theoretical understanding of such issues and practical solutions to address these challenges. In particular, my work involves: 1) Distributed Learning: Allowing data owners to hold their data locally while contributing to AI model training, thereby adding a layer of privacy protection. 2) Differential Privacy: Providing a mathematical definition of privacy and inspiring general privacy protection mechanisms. 3) Privacy Attacks: Evaluating the privacy risks associated with AI technologies.