SUPR
Byzantine-Robust Federated Learning
Dnr:

NAISS 2025/22-260

Type:

NAISS Small Compute

Principal Investigator:

Javad Parsa

Affiliation:

Uppsala universitet

Start Date:

2025-02-17

End Date:

2026-03-01

Primary Classification:

10201: Computer Sciences

Allocation

Abstract

Federated Learning (FL) is a machine learning approach that enables multiple clients to collaboratively train a model while keeping their data decentralized rather than centrally stored. A key characteristic of FL is data heterogeneity—since client data remains decentralized, the data samples across clients are often non-independent and non-identically distributed. FL has numerous applications, particularly in self-driving cars, healthcare, medical AI, robotics, and biometrics. However, one of its major challenges is ensuring robustness against malicious clients. During the learning process, some clients may be compromised by adversarial attacks, leading to data corruption and model manipulation. This project aims to develop robust FL techniques that can detect and mitigate malicious clients during training, ensuring they are identified and excluded from the learning process to maintain model integrity.