A Survey on Optimizing the Training Efficiency and Scalability of Graph Neural Networks

PhD Qualifying Examination


Title: "A Survey on Optimizing the Training Efficiency and Scalability 
of Graph Neural Networks"

by

Mr. Zhiyuan LI


Abstract:

Graph Neural Networks (GNNs) have rapidly gained increasing popularity, 
establishing themselves as the primary tools for a wide range of fundamental 
graph-related tasks including node/graph classification and link prediction, 
due to their successful applications in various domains, such as social 
network analysis, biomedical research, recommendation systems and so on. The 
expressiveness power of GNNs lies in the nested neighborhood aggregation 
layers which incorporate both the graph topology structure and feature 
information. However, such an effectiveness gain comes at the cost of 
increased time or space complexity since the neighborhood aggregations 
introduce complex data dependencies and significant computation and 
communication cost. To address the efficiency issue of GNNs and scale them to 
larger real-world graphs, tremendous research endeavors have been devoted to 
accelerating GNN training and improving its scalability. In this survey, we 
examine the challenges they encounter and address. We also present a new 
taxonomy for the optimizations and present their methodologies in categories: 
(1) data-level optimization, (2) model-level optimization, and (3) 
system-level optimization. In the conclusion, we summarize the existing 
research efforts with the trend on how the optimization techniques develop 
and the discussion on future directions in this research topic.


Date:                   Friday, 12 December 2025

Time:                   8:00pm - 10:00pm

Zoom meeting:
https://hkust.zoom.us/j/94463591685?pwd=YfdMQjSQyjkp2IqgbXDB0qdXxhGYW2.1

Committee Members:      Prof. Lei Chen (Supervisor)
                        Prof. Raymond Wong (Chairperson)
                        Prof. Qiong Luo