Federated Learning in Vehicular Networks Faces New Poisoning Attack and Defense Strategy

This paper introduces a game-theoretic framework to analyze the interaction between clean-label poisoning attacks and defenses in federated learning (FL) within vehicular networks. The study models anomaly detection using Basic Safety Messages (BSMs) and proposes a feature-space clean-label attack that subtly perturbs benign samples to degrade classification accuracy. A feature-consistency relabeling defense is presented to detect and correct these suspicious samples, aiming to maintain high accuracy while mitigating the attack's effectiveness. The research demonstrates that the attack can scale with compromised nodes, but the proposed defense effectively limits its success and shifts the equilibrium towards more attack-averse strategies. AI

IMPACT Introduces new security considerations and defense mechanisms for federated learning in critical infrastructure like vehicular networks.

RANK_REASON Academic paper detailing a novel attack and defense mechanism for federated learning. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• Basic Safety Message
• Clean-label Poisoning Attack
• Feature-consistency relabeling defense
• Feature-space clean-label attack
• federated learning
• Game-Theoretic Frameworks for Demand Response in Electricity Markets
• vehicular communication system

• paper
• safety

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →