Attacks on Privacy in Federated Learning Scenarios



Jonas Geiping (University of Maryland)

Jonas Geiping received his PhD degree in 2021 from the University of Siegen, Germany and is now a postdoctoral research at the University of Maryland, College Park, United States. His research interests include optimization in machine learning and the implications of optimization in machine learning for privacy and security applications.



Short Abstract: This talk will briefly introduce the paradigm of federated learning which in which machine learning models are trained collaboratively over a group of users, ideally without sharing private data. However, the actual security of this scheme depends on a variety of factors and possible threat models. We will discuss several scenarios in which attacks are possible that can breach the privacy of federated learning.