Adversarial Attacks on Cognitive Self-Organizing Networks: The Challenge and the Way Forward

Abstract

Future communications and data networks are expected to be largely cognitive self-organizing networks (CSON). Such networks will have the essential property of cognitive selforganization, which can be achieved using machine learning techniques (e.g., deep learning). Despite the potential of these techniques, these techniques in their current form are vulnerable to adversarial attacks that can cause cascaded damages with detrimental consequences for the whole network. In this paper, we explore the effect of adversarial attacks on CSON. Our experiments highlight the level of threat that CSON have to deal with in order to meet the challenges of next-generation networks and point out promising directions for future work.

AI-Driven Zero Touch Network and Service Management in 5G and Beyond: Challenges and Research Directions

“…It aims at assessing the security robustness of ML algorithms against attacks and designing appropriate countermeasures. While AML has attracted much interest in vision field, only very few contributions (e.g., [15], [16]) have addressed ML security in the context of service and network management. Usama et al [15] highlight the importance of tackling adversarial attacks against cognitive selforganizing networks.…”

“…While AML has attracted much interest in vision field, only very few contributions (e.g., [15], [16]) have addressed ML security in the context of service and network management. Usama et al [15] highlight the importance of tackling adversarial attacks against cognitive selforganizing networks. As a proof of concept, white-box evasion attacks against Convolutional Neural Network have been designed to show how a malware classifier can be evaded.…”

ZSM Security: Threat Surface and Best Practices

“…The privacy attacks aim to obtain private information about the system, its users or data by reverse-engineering the learning algorithm. The attacks against ML can be also divided into two categories based on the attacker's knowledge [12], namely: (i) White-box attacks, which assume that the attacker has complete knowledge about the training data, the algorithm and its hyper-parameters. (ii) Black-box attacks, which assume that the attacker has no knowledge about the algorithm and its hyper-parameters.…”

“…To mitigate model inversion and model extraction attacks, various solutions have been proposed, ranging from restricting information provided by ML APIs, adding noise to the ML predictions, to adding noise to execution time of the ML model. While AML has attracted much interest in computer vision field, only very few contributions (e.g., [12], [11]) have addressed ML security in the context of service and network management. Usama et al [12] highlight the importance of tackling adversarial attacks against cognitive self-organizing networks.…”

The Adversarial Machine Learning Conundrum: Can the Insecurity of ML Become the Achilles' Heel of Cognitive Networks?

“…This method of generating adversarial ML attacks is called the fast gradient sign method (FGSM). Kurakin et al [9] proposed the basic iterative method (BIM) attack which improves the FGSM attack by introducing an iterative small perturbation optimization method for generating adversarial examples. Papernot et al [3] proposed a targeted saliency map based attack, where saliency map is used in an iterative manner to find the most significant features of the input that when fractionally perturbed cause DNNs to misclassify, this adversarial attack is known as Jacobian saliency map based attack (JSMA).…”