Document worth reading: “A survey on Adversarial Attacks and Defenses in Text”

Deep neural networks (DNNs) have confirmed an inherent vulnerability to adversarial examples which might be maliciously crafted on precise examples by attackers, aiming at making aim DNNs misbehave. The threats of adversarial examples are extensively existed in image, voice, speech, and textual content material recognition and classification. Inspired by the sooner work, researches on adversarial assaults and defenses in textual content material space develop shortly. To the perfect of our data, this textual content presents an entire evaluation on adversarial examples in textual content material. We analyze the advantages and shortcomings of newest adversarial examples period methods and elaborate the effectivity and limitations on countermeasures. Finally, we focus on the challenges in adversarial texts and current a evaluation course of this aspect. A survey on Adversarial Attacks and Defenses in Text