Command Palette
Search for a command to run...
Attack Success Rate ASR (Attack)
Date
Tags
Attack Success Rate (ASR, attacker's perspective) is a core metric in adversarial attack research, defined as the proportion of adversarial examples that successfully cause the model to output incorrect predictions. The current entry adopts the attacker's perspective, so it is usually reported as a percentage or a 0–1 score, where higher values are better, indicating that the attack method is more effective. From a defense or robustness evaluation perspective, the same value is better when lower. When using it, one must clearly report the perspective, perturbation budget, success criterion (e.g., whether targeted attacks are allowed), and sample set; otherwise, cross-paper comparisons are of limited meaning.
Build AI with AI
From idea to launch — accelerate your AI development with free AI co-coding, out-of-the-box environment and best price of GPUs.