Project description
As Deep Learning and Generative AI applications continue to expand, optimizing computational efficiency is becoming increasingly critical, particularly for AI in resource-constrained environments or at the edge.
To address this challenge, semiconductor manufacturers have introduced dedicated Neural Processing Units (NPUs), significantly enhancing performance and energy efficiency. AMD, for example, integrates a CPU, NPU, and GPU into its latest Ryzen processors, unlocking new possibilities for on-device AI acceleration.
This project aims to maximize the potential of AMD’s cutting-edge NPUs in the area of computer vision. This project is partially funded by AMD and the successful candidate will collaborate with AMD researchers.
As part of this research, you will
- Investigate how different AI tasks perform on AMD Ryzen CPU, GPU, and NPU, in terms of inference speed, energy consumption, and performance per watt. Different quantization levels (e.g., int8, fp16) will also be explored.
- Develop intelligent workload allocation techniques to maximize resource utilization.
- Develop optimized software routines of commonly used AI tasks tailored for AMD NPUs.
- Validate these methods on two real-world applications: bone fracture recognition and video-based sign language recognition.
Why apply?
- Conduct high-impact research in AI system acceleration.
- Collaborate with AMD engineers, gaining valuable industry experience.
- Benefit from strong career prospects in both academia and industry.
Candidate requirements
We seek highly motivated candidates with expertise in programming (C++, Python), computer architectures, Deep Learning (PyTorch, TensorFlow), and AI acceleration.
Exceptional international candidates may be eligible for a fee waiver (read the following section carefully).