Work

Accelerating Deep Neural Networks on Low Power Heterogeneous Architectures

Year: 2018
Type: conference-paper
Abstract: Deep learning applications are able to recognise images and \nspeech with great accuracy, and their use is now everywhere in our daily \nlives. However, developing deep learning architectures such as deep neural networks in embedded systems is a challenging task because of the \ndemanding computational resources and power consumption. Hence, sophisticated algorithms and methods that exploit the hardware of the \nembedded systems need to be investigated. This paper is our first step \ntowards examining methods and optimisations for deep neural networks \nthat can leverage the hardware architecture of low power embedded devices. In particular, in this work we accelerate the inference time of the \nVGG-16 neural network on the ODROID-XU4 board. More specifically, \na serial version of VGG-16 is parallelised for both the CPU and GPU \npresent on the board using OpenMP and OpenCL. We also investigate \nseveral optimisation techniques that exploit the specific hardware architecture of the ODROID board and can accelerate the inference further. \nOne of these optimisations uses the CLBlast library specifically tuned \nfor the ARM Mali-T628 GPU present on the board. Overall, we improve \nthe inference time of the initial serial version of the code by 2.8X using \nOpenMP, and by 9.4X using the most optimised version of OpenCL.
Language: English
Open Access status: green
OpenAlex ID: W2786457529
FWCI: 2.764
Cites: 29
Cited by: 46
Related to: 10