基于掩码语言模型的文本对抗样本检测（英文版）

英文标题：Masked Language Model Based Textual Adversarial Example Detection中文摘要：提出了基于掩蔽语言模型的检测方法（MLMD），用于区分正常示例和对抗攻击示例，通过探索被掩蔽语言模型引起的流形变化产生明显可区分的信号，并且在各种基准文本数据集、机器学习模型和最先进的对抗攻击上都表现出强大的性能。英文摘要：Adversarial attacks are a serious threat to the reliable deployment ofmachine learning models in safety-critical applications. They can misguidecurrent models to predict incorrectly by slightly modifying the inputs.Recently, substantial work has shown that adversarial examples tend to deviatefrom the underlying