Our group's full-day tutorial introduced participants to the critical aspects of security and privacy in AI, with a particular focus on Federated Learning. The first part of the morning session, led by Prof. Sadeghi, provided an overview of the challenges and risks associated with Large Language Models (LLMs) and DeepFakes. He then explained the motivation behind Federated Learning as a solution to some of these challenges.

Following this, Phillip Rieger delved into the unique challenges posed by the distributed nature of Federated Learning, especially when dealing with heterogeneous data across different clients. He discussed how these variations can impact the detection of manipulated model updates, highlighting the complexities involved in maintaining security in such systems.

In the afternoon, the tutorial transitioned to a practical, hands-on session. Participants engaged in 14 exercises designed to provide practical experience with Federated Learning and its security aspects.

Under the supervision of Alessandro Pegoraro and Phillip Rieger, participants implemented state-of-the-art attacks and defenses for both targeted and untargeted poisoning attacks. These exercises offered valuable practical insights and experience, enhancing the participants' understanding of the security challenges in Federated Learning.

The active engagement and the insights gained by the participants made the tutorial a highly rewarding experience for both the instructors and the attendees.

A huge thank you to Stjepan Picek and Lejla Batina for organizing this fantastic event. Their efforts created a platform for many interesting talks and fruitful discussions, making this a memorable and impactful experience for all attendees. We are looking forward to more such enlightening events in the future!