Machine Learning and the Trusted Internet of Things (IoT)

Abstract

Machine learning employs computational methods from advanced analytics that use statistical algorithms to find patterns in datasets. The Internet of Things (IoT) represent the seamless merging of the real and digital world, with new devices being created that store and pass around data. These machine learning models are updated and refined by continually feeding in new user data (as features) and their feedback (as labels) from the IoT devices. Trust is a decision-making process and is always considered a binary decision. More specifically, trust is treated as a performance requirement in the Software Development Life Cycle (SDLC) that refers to a system’s specific capabilities. Machine learning techniques are being applied to the IoT environment to facilitate performance and efficiency with the concept of edge computing. However, current machine learning models may threaten IoT environments’ security, privacy, and trust. Fortunately, future machine learning models, utilizing trust, may mitigate or even offset these current issues. There are two levels. First, the static trust computation at this level focuses on static attributes associated with a device. Second, the dynamic trust computation uses an initial trust level gained by a device or obtained from a recommender along with keeping track of all the interactions that are happening in between the devices. There is a need for well-defined trust models for IoT applications, where the trust score is a performance metric based on functional properties relevant to the collaboration context.