Title:	Resilient federated learning (FL) based autonomous driving and FL Initialisation for security
DNr:	Berzelius-2026-130
Project Type:	LiU Berzelius
Principal Investigator:	Panagiotis Papadimitratos <papadim@kth.se>
Affiliation:	Kungliga Tekniska högskolan
Duration:	2026-04-07 – 2026-11-01
Classification:	10202
Homepage:	https://www.kth.se/profile/papadim
Keywords:

Abstract

Federated Learning (FL) has emerged as a promising solution for privacy-preserving autonomous driving, specifically camera-based Road Condition Classification (RCC) systems. Cooperative deep learning model training harnesses distributed sensing, computing, and communication resources on board vehicles without sharing sensitive image data. However, the collaborative nature of FL-RCC frameworks introduces new vulnerabilities. In particular, Targeted Label Flipping Attacks (TLFAs) allow malicious clients (vehicles) to deliberately alter their training data labels, thereby compromising the learned model inference performance. Based on our current results, we will explore additional types of poisoning attacks and develop new defensive schemes. We will execute extensive experiments across various RCC tasks, evaluation metrics, baselines, and deep learning models to demonstrate the effectiveness of our scheme in mitigating the attack impact. In addition, we will investigate techniques that can strengthen FL against attacks in a broader sense, leveraging for example techniques that initlize FL in a different way towards achieving stronger resilience.