Xavier Lessage

The course aims to introduce students to the understanding of different Federated Learning concepts with a focus on security vulnerabilities and cyber security challenges. The course will introduce the Federated Learning and compare it to other Machine Learning approaches. The main concepts and process of Federated Learning will then be presented. The model aggregation phase will then be presented along with the security threats. The concept of differential privacy and its relevance for federated Learning explained. Homomorphic encryption techniques will then be introduced for securing the federated Learning process. The course will then present open-source frameworks for federated learning that will be used in the practical work. The practical work aims at applying Federated Learning concepts with a practical exercise from the medical/hospital domain (classification of medical images (malignant or benign lesions)). The practical work will cover the steps required to train a neural network (CNN) with a Federated learning architecture. The practical work will involve adapting the model to meet cybersecurity challenges and performing cybersecurity tests.