Exploring Security Vulnerabilities of Deep Learning Models by Adversarial Attacks

Figure 3

Accuracy changes with the number of replaced words (Yelp review dataset).