IaC cloud testbed for secure ML based management of IoT services
conference paper
The use of Machine Learning in many control and management processes is rising fast. This is especially true for the management of cloud native applications and for anomaly detection in cyber-security mechanisms. However, it has been shown that ML is an attack surface on its own, especially when the attack is performed by another ML system. Noting this issue, we develop a testbed infrastructure with Infrastructure-as-Code principles, designed to support research in the area of cloud services vulnerability to massive attacks from unsecure Internet of Things botnets with ML devised strategies. The infrastructure is modular and can support different malware to create botnets and use different algorithms in both the defensive and offensive sides, provided they use the available data or extend the data collection. We also present some results from a scenario where the defense uses an autoencoder for detecting anomalies in the traffic patterns of IoT devices and the attacker uses a Contextual MultiArmed Bandit algorithm to penetrate this defense and induce resource waste on the cloud service behind it. (C) 2023 IEEE.
Topics
AutoencoderAloud computingAontextual multi-armed banditFlexible resource allocationIoTMLAnomaly detectionBotnetCybersecurityDistributed database systemsInternet of thingsLearning systemsMalwareNetwork securityWeb servicesAuto encodersCloud servicesCloud-computingContextual multi-armed banditFlexible resourcesMultiarmed bandits (MABs)Resources allocationTestbeds
TNO Identifier
985768
Publisher
IEEE
Source title
6th Conference on Cloud and Internet of Things, CIoT 2023
Pages
239-246
Files
To receive the publication files, please send an e-mail request to TNO Repository.