A collaborative CPU-GPU approach for deep learning on mobile devices

摘要

As mobile devices become more prevalent, users tend to reassess their expectations regardingthe personalization ofmobile services. The data collected by amobile device's sensors provide anopportunity to gain insight into the user's profile. Recently, deep learning has gained momentumand has become the method of choice for solving machine learning problems. Interestingly,training a deep neural network on a mobile device is often mistakenly regarded as cumbersome.For instance, several deep learning frameworks only provide a CPU-based implementation forprediction tasks on a mobile device. In contrast to servers, a mobile computing environmentimposes many domain-specific constraints that invite us to review the general computingapproach used in a deep learning framework implementation. In this paper, we propose adeep learning framework that has been specifically designed for mobile device platforms. Ourapproach relies on the collaboration of the multicore CPU and the integrated GPU to acceleratedeep learning computation onmobile devices.Ourwork exploits the sharedmemory architectureof mobile devices to promote CPU-GPU collaboration without any data copying. We analyzeour approach with regard to three factors: performance/portability trade-off, power efficiency,and memory management.