Privacy preserving and Robust Federated Learning

About the project

The project aims to address two interconnected challenges, privacy and security, within the context of Federated Learning (FL) by proposing novel algorithms. The primary focus is on exploring the potential of compression techniques in FL training to create a computation-efficient private and secure FL system. Unlike previous studies, the project sets a distinct goal: to propose compression strategy that strikes the best trade-off among privacy, robustness with low computational complexity, and model performance. The project's novelty lies in its exploration of an uncharted direction, seeking a thorough understanding of how compression can augment the computation efficiency, privacy, and security of FL systems. The ultimate goal is to contribute valuable insights to the field and advance the development of computation-efficient private and secure FL systems, with potential applications in healthcare and beyond.

Principal investigator
Project's partners
  • Nirupam Gupta, Univeristy of Copenhagen
  • Giovanni Neglia, Inria d’UCA
  • Hao Chen, Hong Kong University of Science and Technology
Duration
  • November 2024 - November 2025
Total amount
  • 69700 euros