However, this type of sparse data cannot be consumed directly by deep learning models like CNN or LSTM, as these models expect inputs in the form of dense matrices (or tensors). To address this limitation, a more general class of models called Graph Neural Networks (GNN) is proposed, which can efficiently learn from raw graph-structured data.
This project aims to study the hardware execution of the GNN workload, identify bottlenecks, and improve energy efficiency by developing specialized software/hardware solutions. This will involve implementations of GNN on GPU and FPGA. In the project, the student will learn the GNN algorithm, frameworks like TensorFlow or PyTorch, internal architecture of a GPU, and FPGA designing.
- The project is also open for recently graduated undergraduate students.
- The project is available in the Fall and Spring semester.
- Number of placements available: 2 per semester.
- The start and end date of participation is flexible and will be decided upon discussion with the research project supervisor.
The department of Electrical Engineering (also known as ESAT) of the KU Leuven conducts research at a high international level. It is also responsible for education in the domains of electrical engineering, electronics, and information processing.
ESAT works on several technological innovations in the fields of energy, integrated circuits, information processing, image & speech processing, and telecommunication systems. The department is also co-founder of many spin-off companies. With more than 300 PhD students, 200 master students, and 100 staff members, ESAT is a strong international research and educational department.