Low-ASR Backdoors: Exploiting Attack Success Rate Reduction and Attacker-Defender Asymmetry
What it is
This paper demonstrates that backdoor defenses relying on high Attack Success Rates (ASR) are fundamentally flawed by showing that attackers can deliberately lower ASR without removing the backdoor. It introduces a reverse-training framework that weakens trigger-target associations to create stealthy, low-ASR backdoors that evade state-of-the-art defenses.
The problem it solves
Existing backdoor defenses fail because they assume successful backdoors must have high ASRs, a variable attackers can easily manipulate to evade detection.
What you could use it for
- Evaluating the robustness of existing backdoor detection and defense mechanisms
- Developing new defense strategies that do not rely on high-ASR assumptions
- Understanding attacker-defender asymmetries in machine learning security
Read the source ↗Back to the library →
Summary written by an AI model from the source; check the source before you rely on it.