Description
Date depot: 28 mars 2023
Titre: Privacy Preserving Federated Learning for the Internet of Things
Directeur de thèse:
Farid NAIT-ABDESSELAM (LIPADE)
Domaine scientifique: Sciences et technologies de l'information et de la communication
Thématique CNRS : Systèmes et réseaux
Resumé: Machine learning models benefit greatly from training datasets that are large and diverse. However, it is often difficult for an individual organization to collect sufficiently large and diverse data. Additionally, data sensitivity and government regulations such as GDPR, HIPAA, and CCPA limit how organizations can share data with other entities. This forces organizations with sensitive datasets to develop models that are only optimal locally. Federated Learning (FL) facilitates robust machine learning by enabling the development of global models without sharing private and sensitive data. However, there are two major challenges associated with deploying FL systems: privacy challenges and training/performance challenges. Privacy concerns pertain to attacks that reveal sensitive information about local client data. Training/performance challenges include high communication costs, heterogeneity of data between clients, and lack of personalization techniques. All these concerns must be addressed to make Federated Learning practical, scalable, and useful in this era of internet of things. In this thesis, new techniques will be developed to address and mitigate these challenges.
Doctorant.e: Wei Wenjing