Talk Title
Accelerating GNNs Learning: Review and Proposal

Talk Summary
Graph Neural Networks (GNNs) are the most widely used techniques for handling unstructured data and have demonstrated success in various applications, such as drug discovery and recommendation systems. However, the utilization of GNNs on large-scale graphs is constrained due to their substantial memory demands for storing node and edge features at every layer, coupled with resource-intensive recursive computation for aggregating neighbouring nodes, leading to neighbourhood expansion problems. Consequently, applying GNNs to large-scale medical data is particularly difficult, limiting their scalability for such large-scale datasets. In this tutorial, we will first introduce the existing solutions in the literature to achieve efficient memory and computation for GNNs when learning on large-scale graphs. Subsequently, we will present our methodologies for expediting GNN learning via a hyperedge sampling strategy.

Speaker Bio – Jiameng Liu
Jiameng Liu, a third-year Ph.D. student from ShanghaiTech University, is presently engaged in studies at Imperial College London under the guidance of Dr Islem Rekik. His research concentrates on medical image analysis, with a particular emphasis on infant structural MRI data. His work spans infant brain tissue segmentation, tissue topology correction, and brain surface reconstruction. Currently, he is delving into the realm of Graph Neural Networks (GNNs) to accelerate learning processes and their potential application in medical image classification.

Time: 14.00 – 15.00
Date: Tuesday 7 May
Location: Hybrid Event | Online and in I-X Conference Room, Level 5
Translation and Innovation Hub (I-HUB)
Imperial White City Campus
84 Wood Lane
W12 0BZ

Link to join online via Teams.

Any questions, please contact Andreas Joergensen (a.joergensen@imperial.ac.uk).