IoT -based adversarial attack's effect on cloud data platform services in a smart building context

Abstract

IoT sensors and sensor networks are widely employed in businesses. The common problem is a remarkable number of IoT device transactions are unencrypted. Lack of correctly implemented and robust defense leaves the organization's IoT devices vulnerable to numerous cyber threats, such as adversarial and man-in-the-middle attacks or malware infections. A perpetrator can utilize adversarial examples when attacking machine learning (ML) models, such as convolutional neural networks (CNN) or deep neural networks (DNN) used, e.g., in DaaS cloud data platform service of smart buildings. DaaS cloud data platform's function in this study is to connect data from multiple IoT sensors, databases, private on-premises cloud services, public or hybrid cloud services into a metadata database. This study focuses on reviewing adversarial attack threats towards artificial intelligence systems in the smart building's context where the DaaS cloud data platform services under various information propagation chain structures utilizing ML models and reviews. Adversarial examples can be malicious inputs to ML models providing erroneous model outputs while appearing to be unmodified in human eyes. This kind of attack can knock out the classifier, prevent ML model from generalizing well, and from learning high-level representation, but instead to learn superficial dataset regularity. The purpose of this study is to investigate, detect, and prevent cyber-attack vectors, such as adversarial attacks towards DaaS cloud data platform.