Fig. 2

Example of selecting a communication path & carrying out neural network collaborative computing on the devices of a selected path. It shows an example of the collaborative computing method of neural network in an edge network. The end user has two paths to the cloud. However, as the first path (on left in black) is shorter and has larger available bandwidth when comparing with the second one (on right in red), it may achieve shorter overall communication delay. And we assume that the overall computing delay with the properly designed three neural sub-models deployed on the first path is smaller than the five neural sub-models deployed on the second path, then the first path is selected to process the data, and it illustrates the deployment of the neural network sub-models along the first path. Meanwhile, we can observe that this computing method has the characteristics of pipeline computing. When a batch request is made by the end device, the edge devices on the path can perform parallel computing as pipelines