CASPIAN JOURNAL
MANAGEMENT AND HIGH TECHNOLOGIES
A METHOD FOR PROTECTING A MACHINE LEARNING SYSTEM FROM MALICIOUS PROGRAMS
 |
Read | Petrenko Vyacheslav I., Tebueva Fariza B., Anzorov Artur R., Struchkov Igor V. A METHOD FOR PROTECTING A MACHINE LEARNING SYSTEM FROM MALICIOUS PROGRAMS // Caspian journal : management and high technologies. — 2022. — №1. — pp. 113-127. |
Petrenko Vyacheslav I. - Candidate Sci. (Engineering), Acting Director of the Institute for Digital Development, Head of the Department of Organization and Technology of Information Security, North-Caucasian Federal University, vipetrenko@ncfu.ru
Tebueva Fariza B. - Doct. Sci. of (Physics and Mathematics), Head of the Department of Computer Security, North-Caucasian Federal University, ftebueva@ncfu.ru
Anzorov Artur R. - student, North-Caucasian Federal University, artanzrv@gmail.com
Struchkov Igor V. - postgraduate student, North-Caucasian Federal University, selentar@bk.ru
The article is devoted to the problem of protecting a machine learning system from malware. An analysis of possible vulnerabilities of machine learning systems has been carried out, a classification of the most dangerous attacks with a description of classes, including the method of impact and consequences of using these attacks in a machine learning system, has been given. To counter a number of attacks, a method is proposed for protecting a machine learning system from malware based on the Neural-Cleanse and Jpeg-Compression algorithms.
Key words: machine learning, neural networks, information security, Neural-Cleanse, Jpeg-Compression, poisoning attacks, evasion attacks, model extraction attacks