Abstract:

This paper presents a review of adversarial machine learning in the area of war on algorithms. It collects knowledge in this field, shows the practical applications of attacks and defenses with the use of adversarial machine learning (AML), presents the issues and current advancement of research in this area. The knowledge is summed up into several AML taxonomies and presents methods of defenses and attacks onto intelligent systems depends on the machine learning model phase in which an attack is performed. In addition, a number of experiments were performed to present examples and effectiveness of attacks and defenses against image recognition model.