Doctoral student / Etudiant doctorant, LISTIC, Université Savoie Mont Blanc
PERSONAL INFORMATION
Email : mamour.ndiaye@univ-smb.fr
Office : A221
Adress : 5 chemin de Bellevue, Annecy-le-Vieux, CS 80439, 74944 ANNECY CEDEX
Research team : ReGaRD
THESIS
Subject :Privacy-preserving federated learning incorporating domain adaptation without compromising performance.
Supervisors : Kavé SALAMATIAN, Faiza LOUKIL (ReGaRD)
Doctoral School : Sciences, Ingénierie, Environnement (SIE)
Start of the thesis : Octobre 2025
Abstract:
The objective of this Ph.D. research is to design a novel method that enhances privacy risk
mitigation in federated learning while maintaining high model performance in the face of
dynamic and heterogeneous client data. To this end, the research will explore integrating
domain adaptation techniques into the FL pipeline. The goal is to adapt the global model
dynamically to new or evolving client data distributions without accessing the raw data. This
approach will simultaneously improve generalization capabilities and privacy preservation.
A key aspect of this Ph.D. research will be evaluating the proposed method against widely
adopted baselines in terms of performance. Moreover, the research will simulate common
attack scenarios, such as membership inference and training data reconstruction, to evaluate
the effectiveness of the approach in mitigating privacy risks compared to conventional
privacy-preserving techniques.