Adversarial Texts

AbstractAdversarial attacks against natural language processing systems, which perform seemingly innocuous modifications to inputs, can induce arbitrary mistakes to the target models. Though raised great concerns, such adversarial attacks can …...

There are two cases describing how a classi- fier processes input text, namely, misclassi- fication and correct classification. In terms of misclassified texts, a classifier handles the texts with both …...