M2C: Energy efficient mobile cloud system for deep learning

Abstract

With the number increasing of applications and services that are available on mobile devices, mobile cloud computing has drawn a substantial amount of attention by academia and industry in the past several years. When facing the most exciting machine learning applications such as deep learning, the computing requirement is intensive. For the purpose of improving energy efficiency of mobile device and enhancing the performance of applications through reducing execution time, M2C offloads computation of its machine learning application to the cloud side. We propose the prototype of M2C with the mobile side on Android, iPad and with the cloud side on the open source cloud: Spark, a part of the Berkeley Data Analytics Stack with NVIDA GPU. M2C's distinct set of varying computational tools and mobile nodes allows for thorough implementing distributed machine learning algorithm and innovative wireless protocols with energy efficiency, verifying the theoretical research and bringing the user extremely fast experience.