DEEP TRAINING IN METHODS OF PROTECTION AGAINST ATTACKS

In recent years, machine learning algorithms, or rather deep have beenwidely used in many fields, including cybersecurity. However, systems are vulnerableto attacks by attackers, and this limits the use of learning, especially nonstationaryenvironments with hostile actions, such as cybersecurity field, where real attackersexist (for example, malware developers). With rapid development artificial intelligence (AI)and (GO) methods, it is important to ensure safety reliability implementedalgorithms. Recently, vulnerability algorithms conflicting patternshas been widely recognized. Fabricated samples for analysis can lead various violations thebehavior models, while people will consider them safe use. The successful implementationof enemy physical situations scenarios worldonce again proves their practicality. As a result, methods adversarial attack defense areattracting increasing attention from security communities havebecome hot topic research years not only Russia, but also other countries.Sberbank, Yandex, T1 Group, Atlas Medical Center others developing competitivesolutions, on international market. Unfortunately, list 10 largest ITcompanies, direction Big Data, particular, protection against representedonly Group company, market growth potential huge. paper, theoreticalfoundations, application arepresented. Then number papers described, covering awide range area. This article explores summarizes anddefenses, which represent most up-to-date field meet latest requirementsfor information security.