Informatics collaboration with Huawei on Efficient Neural Networks for mobile devices
Christophe Dubach from Institute for Computing Systems Architecture will work with Huawei Technologies on a project that aims to conserve battery life on mobile devices.
Machine learning applications are used extensively in areas such as computer vision, machine translation or self-driving cars. This has become possible thanks to the widespread use of parallel commodity hardware such as GPUs (Graphics Processing Units). However, writing efficient parallel implementations for these applications remains a challenge even for expert programmers. This expensive and time-consuming process has to be repeated every time new hardware emerges or when the machine learning model changes.
In this project, a collaboration with Huawei Technologies, the researchers propose to target mobile devices, which are becoming one of the fastest growing environments for machine learning applications. This project will extend existing Lift compiler, developed in Edinburgh, to deal with such applications and transparently generate efficient code without any programmer intervention.
Our end goal is to exploit effectively mobile GPUs, leading to higher energy-efficiency, one of the most important factors for conserving battery life on mobile devices.