SUPR
Secure and Byzantine-Robust Federated Learning
Dnr:

NAISS 2025/5-125

Type:

NAISS Medium Compute

Principal Investigator:

Salman Toor

Affiliation:

Uppsala universitet

Start Date:

2025-03-28

End Date:

2025-10-01

Primary Classification:

10201: Computer Sciences

Secondary Classification:

10105: Computational Mathematics

Webpage:

Allocation

Abstract

This project focuses on developing secure and robust aggregation techniques in Federated Learning (FL), particularly in adversarial settings where malicious clients attempt to manipulate global model updates. Our research investigates Byzantine-robust aggregation rules, adversarial defenses, and privacy-preserving mechanisms to enhance the security of FL.