D2D-Enabled Data Sharing for Distributed Machine Learning at Wireless Network Edge

Abstract--Mobile edge learning is an emerging technique that enables distributed edge devices to collaborate in trainin g shared machine learning (ML) models by exploiting their local data samples and communication/computation resources. T o deal with the stragglers dilemma issue faced in this technique, this p aper proposes a new device-to-device (D2D)-enabled data sharin g approach, in which different edge devices share their data samples among each other over D2D communication links, in order to properly adjust their computation loads for increa sing the training speed. Under this setup, we optimize the radio resource allocation for both D2D-enabled data sharing and distributed training, with the objective of minimizing the total training delay under fixed numbers of local and global iterat ions (for training). Numerical results show that the proposed D2 D-enabled data sharing design significantly reduces the train ing delay, and also enhances the training accuracy when the data samples are non-independent and identically distributed ( non-IID) among edge devices. Mobile edge learning has recently attracted growing research interests from both academia and industry to enable various new artificial intelligence (AI) applications such as augmented reality (AR), industrial automation, and autonomous driving [1].