A Cloud-Edge-Smart IoT Architecture for Speeding Up the Deployment of Neural Network Models with Transfer Learning Techniques

Existing edge computing architectures do not support the updating of neural network models, nor are they optimized for storing, updating, and transmitting different neural network models to a large number of IoT devices. In this paper, a cloud-edge smart IoT architecture for speeding up the deployment of neural network models with transfer learning techniques is proposed. A new model deployment and update mechanism based on the share weight characteristic of transfer learning is proposed to address the model deployment issues associated with the significant number of IoT devices. The proposed mechanism compares the feature weight and parameter difference between the old and new models whenever a new model is trained. With the proposed mechanism, the neural network model can be updated on IoT devices with just a small quantity of data sent. Utilizing the proposed collaborative edge computing platform, we demonstrate a significant reduction in network bandwidth transmission and an improved deployment speed of neural network models. Subsequently, the service quality of smart IoT applications can be enhanced.