Exploring Security Vulnerabilities of Deep Learning Models by Adversarial Attacks

Figure 4

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