ECOMA - Edge-Cloud Collaborative Framework for Multi-Task Applications.

As Internet of Things (IoT) becomes evermore immersive and embedded in the environments where people live, work and entertain, we expect the edge devices connected through IoT to exponentially gain in computing capability, and device intelligence will transform how we design the IoT architecture as well as the optimal manner in which the IoT applications are enabled. We investigate in this paper the issue of IoT architecture to take advantage of edge intelligence. Most IoT systems typically run multiple applications for different tasks simultaneously. This requires edge infrastructure to be able to switch between applications at the least cost in order to save computation. To address these problems, we propose a cloud-edge collaborative framework which is composed of one universal block (UB, a few initial layers of deep model) located at the edge and multiple parallel task-specific blocks (TSBs) located in the cloud. For each task, a deep model is constructed by the UB and a TSB where the UB is responsible for the general feature extraction and the TSBs are responsible for the subsequent task-specific processes. The proposed framework not only has the common advantages for edge-cloud collaborative structure (such as reducing data transport, protecting user privacy), but also has two important properties: (1) It only installs one UB on the edge, so as to reduce memory storage and computing resources, and make management more convenient; (2) There is no requirement to update the UB frequently, which is useful in practice.