Binary Black-Box Adversarial Attacks with Evolutionary Learning against IoT Malware Detection

Table 4

Performance evaluation on MalConv with and without adversarial training.

Defense	Test dataset	Accuracy
No defense	×	98.4%
	√	80.2%
Adversarial training	√	90.5%

× indicates no adversarial samples and √ denotes having adversarial ones in the test dataset.