Concepts de base
Identifizierung und Ausnutzung von Backdoor-kritischen Schichten für effektive Backdoor-Angriffe in Federated Learning.
Stats
"Extensive experiments show that our BC layer-aware backdoor attacks can successfully backdoor FL under seven SOTA defenses with only 10% malicious clients and outperform latest backdoor attack methods."
"We propose Layer Substitution Analysis, a novel method that recognizes backdoor-critical layers, which naturally fits into FL attackers’ context."
"Our evaluation on a wide range of models and datasets shows that the proposed layer-wise backdoor attack methods outperform existing backdoor attacks, such as DBA (Xie et al., 2019), on both main task accuracy and backdoor success rate under SOTA defense methods."
Citations
"Angriffe auf die BC-Schichten ermöglichen effektive Backdoor-Angriffe mit geringer Entdeckungswahrscheinlichkeit."
"Unsere BC-Schicht-bewussten Backdoor-Angriffe können FL erfolgreich unter sieben SOTA-Verteidigungen mit nur 10% bösartigen Clients backdooren und übertreffen die neuesten Backdoor-Angriffsmethoden."