Graph Learning at Scale: Characterizing and Optimizing Pre-Propagation GNNs

Zichao Yue, Chenhui Deng, Zhiru Zhang

Conference on Machine Learning and Systems 2025 · Day 4 · Session 12: Edge and Cloud Systems

Graph Neural Networks (GNNs) have emerged as a cornerstone in modern machine learning, demonstrating remarkable success in diverse applications ranging from fraud detection to circuit functional reasoning. However, their widespread adoption, especially on large-scale graphs, is often hampered by significant scalability challenges. The "neighbor explosion problem," inherent to traditional **Message Passing GNNs (MPGNNs)**, leads to an exponential increase in memory and computational demands as the network deepens, making training on massive graphs prohibitively expensive. This talk, presented by Zichao Yue alongside collaborators Zhenhui Deng and Professor Zhiru Zhang, delves into **Pre-Propagation GNNs (PPGNNs)**, a promising alternative approach designed to circumvent these scalability bottlenecks.

AI review

Solid systems-level GNN research that identifies a real bottleneck (data loading in pre-propagation GNNs), characterizes it carefully, and applies practical optimizations to close the gap between theoretical and realized efficiency. The core insight — that decoupling propagation from transformation is sound but vanilla implementations eat that advantage through I/O overhead — is genuinely useful. The numbers are impressive, but the article doesn't give me enough implementation specifics to know whether the optimizations are portable or tightly coupled to their test harness. Worth reading for…