Model Distributed Learning at Edge Networks

The traditional approach to distributed deep neural network (DNN) inference in edge computing systems is data-distributed inference. In this paradigm, each worker has a pre-trained DNN model. Using the DNN model, the worker processes the data that is offloaded to itself. In this work, we develop an Adaptive and Resilient Model-Distributed Inference (AR-MDI) algorithm based on our optimal model allocation formulation. Future more, we investigate MDI with multiple sources, when more than one device has data. We design a multi-source MDI (MS-MDI), which optimizes task scheduling decisions across multiple source devices and workers. We also focus on model distributed training and we design ResPipe, a novel resilient model-distributed DNN training mechanism against delayed/failed workers. We also design mcPipe to reduce the memory cost of model-distributed learning.