DynaComm: Accelerating Distributed CNN Training between Edges and Clouds through Dynamic Communication Scheduling
This work addresses the time-consuming nature of distributed training in edge networks, which is an incremental improvement for edge computing applications.
The paper tackles the problem of slow distributed CNN training between edge devices and clouds by introducing DynaComm, a dynamic communication scheduler that decomposes transmissions into segments to overlap communications and computations, achieving optimal layer-wise scheduling without affecting model accuracy.
To reduce uploading bandwidth and address privacy concerns, deep learning at the network edge has been an emerging topic. Typically, edge devices collaboratively train a shared model using real-time generated data through the Parameter Server framework. Although all the edge devices can share the computing workloads, the distributed training processes over edge networks are still time-consuming due to the parameters and gradients transmission procedures between parameter servers and edge devices. Focusing on accelerating distributed Convolutional Neural Networks (CNNs) training at the network edge, we present DynaComm, a novel scheduler that dynamically decomposes each transmission procedure into several segments to achieve optimal layer-wise communications and computations overlapping during run-time. Through experiments, we verify that DynaComm manages to achieve optimal layer-wise scheduling for all cases compared to competing strategies while the model accuracy remains untouched.