Abstract: The security of machine learning, also referred to as Adversarial Machine Learning (AML) has come to the forefront in machine learning and is not well understood in the application to the cyber security area. AML has been largely applied to image classification but has been limited in application to the cyber security area. One of the most fundamental components of machine learning, is the features. The disparate features of the cyber security area vary and are different than in image classification. To understand the features of the cyber security area, traffic classification is selected as a use case to focus on. Additionally, we present an example of cyber security AML of a network scanning classifier. A background on AML attack types, Adversarial Knowledge, and Image Classification features is given first. Next a discussion of the Cyber security traffic analysis features and AML of the cyber security area is given. We propose the disparate features of the cyber security area, augmented with ensemble learning could lead to a defense against AML. Future research is proposed for experimentation of AML with a subset of the cyber features discussed and the development of a defense against AML.
Keywords: Adversarial Machine Learning, Cyber Security, Traffic Analysis, Features, Machine Learning
Download this article: JISAR - V12 N1 Page 26.pdf
Recommended Citation: De Lucia, M. J., Cotton, C. (2019). Adversarial Machine Learning for Cyber Security. Journal of Information Systems Applied Research, 12(1) pp 26-35. http://jisar.org/2019-12/ ISSN: 1946-1836. (A preliminary version appears in The Proceedings of CONISAR 2018)