A Survey on Low Power SBCs for Optimized Implementation of Deep Learning Models

Abstract

SBCs are low computing devices that contain all of the necessary components of a computer on a single board.  SBCs can meet the goals of cost-effectiveness, minimal power usage, design and implementation flexibility and improved communication speed. Deep neural networks can benefit from the structure provided by SBCs. Deep neural network performance on SBCs can be optimized by employing efficient methods that are intended to lower the model's computational requirements. Deep neural networks on SBCs require hardware optimization as well. This may involve accelerating deep neural network processing with specialized hardware including Graphics Processing Units (GPUs) or Field Programmable Gate Arrays (FPGAs). Deep learning edge devices are becoming more and more important in a number of applications, from autonomous automobiles to industrial automation to healthcare, since they provide real-time decision-making and analysis at the edge of the network. An overview of various single board computers available on the market as well as major research efforts involved in the deployment of deep learning at the edge of computer networks is presented in this survey.