Curriculum Computer Science and Innovation for Societal Challenges, XXXVII series  Grant sponsor CSC (China Scholarship Council) Supervisor s Mauro Conti Co-Supervisor s Anna Spagnolli Contact jiaxin.li@studenti.unipd.it

Project description

Training data play an important role in the performance and utility of machine learning models. But sometimes the training data will be inferred by the attackers from the prediction API of the trained model with the help of additional information, which leads to a privacy leakage in the training data-sensitive scenario. So my research focuses on the defense strategies, interpretability, and possible attacking methods about the Membership Inference Attack with the help of differential privacy and some analysis means in the field of privacy and security. In the next step, I will focus on another type of attack in machine learning security.