A Hot/Cold Task Partition for Energy-Efficient Neural Network Deployment on Heterogeneous Edge Device

In recent years, neural network has made great achievements in the fields of image classification and object detection. At the same time, the rapid development of edge devices has also led many scholars to study the deployment of neural networks on resource-limited edge devices. However, the traditional neural network did not consider the specific application scenarios, which makes the data processing very inefficient. In this paper, we propose a method to deploy neural networks on heterogeneous edge devices. This method fully considers the specific applications scenarios of neural network, and proposes the idea of task partition. The whole dataset is divided into two classes: hot class and cold class, and then corresponding network models are trained respectively to handle the data. In order to speed up the execution efficiency, we propose a simple algorithm, which can balance the front-end and back-end load according to the processing capacity of edge devices. In addition, the time delay and energy consumption are considered in this paper. The experiments show that our method can save 36.4% energy consumption compared with the traditional method.