Hybrid Edge–Cloud Intelligence: A Deep Learning Architecture forReal-Time Decision Optimization

Abstract

The growing need of real-time intelligent decision-making in the contemporary computing environments has shown the weakness of strictly cloud-based artificial intelligence systems, especially in cases of latency sensitivity and the resource constraints. Although edge computing can be used to perform inferences with low latency near data, it does not have the computing power demanded to train and optimize deep learning models. The research paper suggests a hybrid architecture, which combines deep learning models at the distributed computing architectures to produce real time, scalable and adaptable decision-making. The suggested architecture effectively separates the inference and learning activities between edge nodes and cloud servers, allowing the edge to make time-sensitive decisions, and allows the cloud server to train and synchronize models and optimize them globally. A system model and workflow are introduced, and a simulated analysis is given on the basis of performance metrics, i.e., latency, bandwidth use, accuracy and energy efficiency. The experimental findings prove that the hybrid architecture can use a much lower inference and bandwidth cost than cloud-only strategies without sacrificing the predictive accuracy of the hybrid model. The results verify that the concept of hybrid edgeous cloud intelligences is a powerful tool in the real-time decision-making in recent intelligent systems.