Inhaltszusammenfassung

Edge-AI refers to the execution of AI tasks, such as inference and limited model training on edge devices located near data sources, reducing latency and network congestion. The SIoT applies social networking principles to IoT systems, enabling smart devices to autonomously form relationships for service discovery and collaboration. Integrating Edge-AI into SIoT enables decentralized, intelligent decision-making and real-time data processing. This paper explores the broad field SIoT to...Edge-AI refers to the execution of AI tasks, such as inference and limited model training on edge devices located near data sources, reducing latency and network congestion. The SIoT applies social networking principles to IoT systems, enabling smart devices to autonomously form relationships for service discovery and collaboration. Integrating Edge-AI into SIoT enables decentralized, intelligent decision-making and real-time data processing. This paper explores the broad field SIoT to derive relevant studies that constitute the research field of Edge-AI in SIoT. SIoT-related papers from multiple academic databases over the past decade are systematically collected, classified and synthesized into a concept matrix, according to research trajectories in SIoT. Despite the enormous research already conducted in SIoT, research needs remain for Edge-AI enabled SIoT, particularly in developing novel use cases and real-world applications, but also creating new approaches for training AI models on edge devices. These gaps indicate opportunities for advancing SIoT through specialized Edge-AI concepts. This study concludes by proposing research needs for future investigation and can be seen as a roadmap for researchers to pursue advancements in Edge-AI enabled SIoT applications. Furthermore, we complemented the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA)-based search with an inductive coding protocol: titles, abstracts, and, where required, full texts were open coded using salient keywords (e.g., federated learning, vehicular edge, trust, human-in-the-loop), refined via comparison to derive the twelve streams and populate the concept matrix; two researchers coded a sample, and reconciled disagreements.» weiterlesen» einklappen

Autoren