FPGA-based Artificial Intelligence
Principal Supervisor: Dr. J. McAllister
+ Project Description
Field Programmable Gate Array (FPGA) offer enormous levels of computation and communications bandwidth to support machine learning and data analysis operations, such as the neural networks which power the latest artificial intelligence systems, with low energy expenditure. They are ideal for deployment in embedded systems such as robots, vehicles or drones. However, Graphics Processing Units (GPUs) are currently much more frequently used due to the difficult FPGA programming process. In addition, the latest of these algorithms combine different kinds of neural processing technology, in different ways as the system evolves. Typical FPGA realisations do not support this kind of flexibility.
This project will develop new ways to overcome this problem. It will use a processor developed at ECIT – the leading of its kind, to develop very large-scale adaptable machine learning, or more specifically neural network, processing platforms. It will demonstrate the effectiveness of this platform on state-of-the-art deep learning algorithms in a representative robotics, computer vision or vehicle setting.
• Develop an understanding of the behaviour, key operations and network structure of state-of-the-art neural networks for machine learning.
• Devise realisations of multicore deep neural networks on FPGA.
• Devise and realise strategies for run-time optimisation of application mapping, scheduling and synchronization.
• Quantify the performance and/or cost benefit of the proposed run-time strategies.
• Create a demonstrator system showing the effectiveness of the resulting platform.
• Present your work in leading international journals and conferences in the area.
+ How to Apply
Applicants should apply electronically through the Queen’s online application portal at: https://dap.qub.ac.uk/portal/
+ Contact Details
|Supervisor Name:||Dr. John McAllister|
Institute of Electronics, Communications and Information Technology (ECIT)
+44 (0)28 9097 1743