Accelerating Neural Network Model Deployment with Transfer Learning Techniques Using Cloud-Edge-Smart IoT Architecture

The distribution and updating of neural network models are efficiently achieved through the use of the collaborative functionalities offered by cloud computing, edge servers, and Internet of Things (IoT) devices. Our study involved conducting tests and simulations to demonstrate the efficacy and effectiveness of the proposed design. Overfitting and inadequate training data are common challenges in traditional machine learning methodologies. Transfer learning is a technique that mitigates these challenges by capitalising on pretrained models and utilising their knowledge to train new models. The rapid implementation of model modifications was made possible by the use of a collaborative edge computing platform. This platform permitted the integration of edge Internet of Things (IoT) devices with the latest advancements in artificial intelligence (AI), eliminating the need for extensive data transfer to the cloud. In addition to examining the system’s performance in different scenarios, our study also investigated its performance in scenarios characterised by varying quantities of edge IoT nodes. Our research proposes a Cloud-Edge-Smart IoT architecture in conjunction with transfer learning methods as a viable and efficient approach to expedite the deployment of neural network models. The approach employed in our study aims to enhance the efficiency of AI applications by using the benefits of cloud computing, edge servers, and IoT devices. This methodology results in reduced data transfer demands, accelerated deployment rates, and enhanced service quality. This research contributes to the progress of edge computing and Internet of Things (IoT) technologies, hence opening up novel opportunities for the implementation of intelligent and real-time applications across many sectors.