Project Summary: Deep learning algorithms use extremely large data sets thus are computational extensive. GPGPUs have been used to accelerate these algorithms. With the wider adoption and refinement of heterogeneous computing language such as openCL, acceleration on FPGAs may provide a closer model to higher level programming languages, without the need of using specific hardware description language such as Verilog or VHDL. FPGAs allow much more flexible hardware configuration than GPUs, and better power performance, so are viable alternatives to GPUs.
Specific Requirements of the Project
A 2:2 (Hons) or above degree in Computer Engineering, Electrical and Electronic Engineering or relevant subjects are required. Knowledge with designing using GPU and/or FPGA, and machine learning is preferred.
Project Aims and Objectives
The aim of the project is to evaluate the suitability of using FPGAs to accelerate deep learning algorithms. Especially it will achieve the following objectives: • Review deep learning algorithms and identify suitable candidate for acceleration on FPGA. • Use openCL to implement the algorithm on Altera FPGAs. • Evaluation the timing and power performances against those published in literature, focusing on one type of applications such as computer vison.
Project is open to: Home/EU and overseas
Informal enquiries can be made to Dr. Helen Ji Tel: 0161 2474617 email: [email protected]