Abstract

The rise of edge computing has transformed the ways in which the data is consumed, analyzed and utilized with the possibilities of decision making at the point when they come into the creation. However, the traditional machine learning approaches are not suitable in the modern smart applications like smart cities, health care, and IoT due to the dynamic, distributed and resource limited nature of the edges. As a result, existing machine learning techniques fall short of handling these issues, and Graph Neural Networks (GNNs), are the solution to these challenges due to their ability to model and learn from graph-structured data.

The present article offers a detailed examination of how GNNs can be incorporated into edge computing to enhance the development of prediction techniques. It describes the architecture of GNNs and focuses on the benefits offered by these networks in learning and analyzing connected data and deeply embedded patterns that escape other architectures. We expand the discussion on factors influencing GNNs deployment on edge devices to include computational capabilities, latency, and security threats. To solve these problems, several solutions are put forward including the model optimization, lightweight algorithms, and federated learning.

Moreover, to validate the novelty of edges-GNN integration and its capabilities, real-world cases are also discussed. Applications are, for example, in intelligent transportation systems in smart cities, in force IoT and in diagnostics in medicine. However, the integration of GNNs and edge computing is a way of opening up more opportunities for the development of efficient, scalable and privacy preserving analytic prediction systems.

Finally, the authors present the outlook for further research, in which they consider the potential application of self-supervised learning, the interaction with new technologies like quantum computing or the creation of collaborative platforms for the definition of a clear edge AI system. This work also established the significance a of narrowing the divide between advanced AI techniques and edge computing in opening the possibility for revolutionary advancements in predictive analytics across fields.