Systematic Literature Review: Evaluating Effects of Adversarial Attacks and Attack Generation Methods

Advancement in Artificial Intelligence (AI) aims to train the Machine Learning (ML) Models in such a way that they would be able to take decisions spontaneously, however on the other side attackers attempt to manipulate the results generated by these models which makes the application of these models difficult in security-critical areas including classification of medical images, autonomous system installed in vehicles, street lights, malware detection and identification of a person as criminal or innocent. Advancement in research proved the vulnerability of these classifiers due to adversarial attacks that can alter their result in training as well as in testing phase of the model. Causative attacks are the training phase attacks whereas the attacks done at testing phase are the exploratory attacks. This systematic literature Review (SLR) is conducted to gain in-depth knowledge of adversarial attacks, which is the most effective type of exploratory attack, parameters on which these attacks are based, along with the most conventional methods of generating adversarial attacks.