Artificial Intelligence Service Architecture for Edge Device

Edge computing is getting more focused recently due to high demand of Artificial Intelligence application, for example, detection of wearing masks from a video stream. In edge computing, the AI applications are placed near data source to improve quality of service, and there are several researches to bring AI service onto edge device such as TensorFlow Serving. However, existing researches focus on providing accessibility of the trained model itself and require additional preprocessing and postprocessing of data to build an end-to-end service. In this paper, an AI Service Architecture for an Edge Device is proposed to provide accessibility to the AI service itself. The proposed architecture provides AI as a service, which means it includes pre-processing and postprocessing, as well as the model itself. Since it includes all the methods which consists an AI service, the proposed architecture provides more intuitive ways to bring an AI method to edge device. Moreover, it defines interfaces to configure and access the AI service, which makes it suitable apply microservice architecture.