Federated learning faces new hybrid Byzantine attacks targeting network pruning

Researchers have developed a novel hybrid Byzantine attack for federated learning that combines a sparse manipulation strategy with a slow-accumulating poisoning method. This approach aims to maximize disruption to the global model while remaining imperceptible to common detection mechanisms. The attack selectively targets sensitive parameters and gradually poisons updates over multiple rounds, demonstrating effectiveness against eight state-of-the-art defense strategies. AI

Summary written by gemini-2.5-flash-lite from 1 source. How we write summaries →

IMPACT Introduces a novel attack vector that could necessitate new defenses in federated learning systems.

RANK_REASON Academic paper detailing a new attack method for federated learning. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• Federated Learning
• Byzantine attack
• neural network
• sparse NNs
• Kerem Özfatura

• paper
• safety