Collaborative learning, and in particular Federated Learning (FL) is a Machine Learning approach in which multiple clients collaboratively train a Neural Network (NN) model on their private data without the need to share the data. With the increasing large-scale application of FL systems in real-world settings, e.g., for IoT Malware detection or mobile risk management, a number of security, privacy, and functional challenges are posed in the design and implementation of the underlying algorithms and systems.
Besides giving an introduction to the security perspective of Machine Learning, this interactive course will focus on security and privacy attacks and defenses in FL systems. In particular, the course will cover attach such as model/data poisoning and backdoor attacks that allow the adversary to control the outcome of the learning system. The course will cover the introduction of the different components of federated learning systems both on clients and server (aggregation algorithms) and provides hands-on tasks of different difficulty including the state-of-the-art backdoor attacks and defenses.