Detect Textual Adversarial Distribution on DNN

What is it about?

In this paper, we provide a new framework, namely TADD for detecting textual adversarial samples by leveraging the interpretability of DNNs. In particular, we distinguish between the adversarial distribution and the benign distribution for the decision boundary of the victim models. Our method applies to NLP tasks and does not require re-training victim models and prior knowledge of adversarial attack methods.

Featured Image

Photo by BoliviaInteligente on Unsplash

Photo by BoliviaInteligente on Unsplash

Why is it important?

We distinguish the adversarial distribution and benign distribution for the decision boundary of the victim models, leading us to discover the new adversarial examples in the future that are generated by an unseen attack method.