Abstract—The rapid advancement of artificial intelligence (AI) and deep learning (DL) approaches, it's more important than ever to assure the security and robustness of the algorithms being used. The vulnerability of DL algorithms to hostile samples has recently been extensively recognised as a security concern. The faked samples can cause numerous DL model misbehaviors while being seen as harmless by humans. Adversarial attacks have been successfully implemented in real-world circumstances, demonstrating their utility. As a result, adversarial attack and defensive strategies have gotten a lot of interest from the machine learning and security sectors, and turned into a hot area of study.The theoretical foundations, methods, and applications of adversarial attack strategies are originally introduced in this study. Following that, we examine a number of outstanding topics and concerns in the hopes of spurring more study in this important area.
Keywords—Machine learning, Deep neural network, FGSM, Adversarial attack, Adversarial defense,CNN
|
DOI:
10.17148/IJARCCE.2021.107100
