Enabling Intelligence on Edge Through an Artificial Intelligence as a Service Architecture

Artificial Intelligence (AI) has changed different applications and markets, making them even more engaging and user-centered in Smart Cities, Smart Farms, and e-health. One such sector that has benefited from these capabilities is precision agriculture, which has seen increased productivity and more efficient and intelligent management through AI and cloud computing. By bringing computing resources closer to clients, edge devices are able to embed AI solutions near the client, reducing the time it takes to transfer data to High-Performance Computing (HPC) centers. The latest approaches explore lightweight or distributed Convolutional Neural Networks (CNNs) methods for training and prediction tasks in edge devices using medical, satellite, or precision agriculture images. In this paper, we hypothesized that prediction tasks in precision agriculture can be more easily and seamlessly assigned to edge devices as long as pre-trained AI models are embedded in these devices through Artificial Intelligence as a Service (AIaaS) Architecture. Hence, we propose and evaluate a general-purpose image classification system for edge devices that uses an AI Model Store for edge intelligence empowerment. We conducted a case study on precision agriculture to functionally assess cognitive service delivery performance, including prediction time, memory, and CPU usage, and found a suitable model embodiment on edge devices compared with HPC.