Description
Date depot: 6 octobre 2022
Titre: Cybersecure Federated Learning
Directrice de thèse:
Maria POTOP-BUTUCARU (LIP6)
Directeur de thèse:
Sébastien TIXEUIL (LIP6)
Domaine scientifique: Sciences et technologies de l'information et de la communication
Thématique CNRS : Systèmes et réseaux
Resumé: The ambition of the research project is to enhance the concept of FML (Federated Machine Learning) and
MADRL (Multi-agent deep reinforcement learning) to a new concept of CF-DRL (CyberSecure Distributed
Reinforced Learning). Follow the FML principles, CF-DRL will utilize local data for model training and
aggregated iteratively at the cloud. However, since FML is associated with frequent information exchange
and communication overheads, the thesis will propose and evaluate the effectiveness of different
optimization methods. Moreover, to achieve real-time and low overhead offloading decisions, we will
experiment with a novel two-timescale DRL approach, consisting of a fast-timescale and a slow-timescale
training process, minimize the total offloading delay and network resource usage by jointly optimizing
computation offloading, resource allocation and service caching placement. Moreover, we will leverage
secure computation techniques and aggregation primitives to privately and securely combine local ML
training, in order to update a global model.
Doctorant.e: Legheraba Mohamed