A Survey of Adversarial Machine Learning in Cyber Warfare

Abstract

The changing nature of warfare has seen a paradigm shift from the conventional to asymmetric, contactless warfare such as information and cyber warfare. Excessive dependence on information and communication technologies, cloud infrastructures, big data analytics, data-mining and automation in decision making poses grave threats to business and economy in adversarial environments. Adversarial machine learning is a fast growing area of research which studies the design of Machine Learning algorithms that are robust in adversarial environments. This paper presents a comprehensive survey of this emerging area and the various techniques of adversary modelling. We explore the threat models for Machine Learning systems and describe the various techniques to attack and defend them. We present privacy issues in these models and describe a cyber-warfare test-bed to test the effectiveness of the various attack-defence strategies and conclude with some open problems in this area of research.

Robustness Evaluations of Sustainable Machine Learning Models against Data Poisoning Attacks in the Internet of Things

“…As these algorithms are used increasingly in business, military, and other real-world use-cases, the requirements for security of these systems and privacy of their data become proportionally more important. Within cyber-security systems, machine learning algorithms are considered to be the weakest link because their nature of constant evolution, and the ability of sophisticated adversaries to manipulate their behaviours to exploit models [6]. There are several ways in which an adversary can violate a machine learning model [23].…”

“…The field of adversarial machine learning (AML) was coined to study how adversarial techniques can potentially exploit ML algorithms, and develop robust strategies to defend systems against the exposure these algorithms generate [5]. AML has been studied out in a number of fields, including image classification [6] and intrusion detection [7,8]. There have been very few studies in IoT systems [3].…”

Adversarial Machine Learning Applied to Intrusion and Malware Scenarios: A Systematic Review

“…Gradient masking comprises a group of defensive techniques which assume that ''if the model is non-differentiable or if the model's gradient is zero at data points, then gradient based attacks are ineffective'' [23]. One form of gradient masking is gradient hiding, which consists on using non differentiable models to perform classification, such as decision trees.…”

“…Models are many times trained with assumptions in mind for convenience or ease of computation, such as feature independence and linear separability of the data, but these types of assumptions can often open possibilities for adversarial attacks [23].…”

“…We also found other papers that apply adversarial concepts to intrusion and malware scenarios, but we specifically focus on adversarial machine learning. Several adversarial machine learning surveys and books were used to find multiple adversarial attack and defense strategies [12], [15], [18], [21], [23], [24], [26], [50]. After exploring these studies, we concluded that adversarial attack techniques were proven effective in both malware and intrusion scenarios, with a high diversity of techniques being tested.…”