Scaffold: Support Graph Theory Based Sparsification for Graph Neural Networks

2026-09-25

AI Summary

This study addresses the high computational overhead of graph neural networks (GNNs) and the tendency of conventional sparsification methods to disrupt communication structures. To this end, it proposes the first scalable unsupervised topological sparsification framework grounded in support graph theory. By introducing a joint path-dilation-congestion criterion alongside preconditioner techniques, the work designs a topology-aware sparsification algorithm that reduces complexity while rigorously preserving structural graph properties. Experimental results demonstrate that retaining only 10%–50% of edges achieves performance comparable to full-graph models, effectively halving memory consumption and significantly accelerating training. This framework offers a theoretically guaranteed yet practical solution for the efficient deployment of large-scale GNNs.

Abstract

Graph neural networks (GNNs) rely on message passing over graph edges, making their computational and memory costs strongly dependent on graph density. Graph sparsification offers a natural way to reduce these costs, but removing edges indiscriminately can distort important communication structure and degrade predictive performance. We introduce Scaffold, a topology-based, unsupervised graph sparsification framework derived from support graph theory preconditioners. Scaffold explicitly controls two complementary structural quantities: dilation, which measures the length of rerouting paths induced by removed edges, and congestion, which measures how strongly these rerouted paths concentrate on the retained support. By jointly controlling dilation and congestion, Scaffold preserves short communication paths while avoiding structural bottlenecks. To our knowledge, Scaffold is the first scalable GNN sparsification framework to use a joint supporting-path dilation-congestion criterion. Across 19 homophilic and heterophilic benchmarks spanning small to large graphs, Scaffold achieves the best aggregate rank among the evaluated sparsification and related methods. Using only 10%-50% of the original edges per sparse support, Scaffold recovers or closely approaches full-graph GNN performance while using less than half the memory of full-graph training and reducing end-to-end training time, including sparsification overhead. We provide an open-source software package at https://github.com/siddhartha047/Scaffold.

Problem

• Graph Neural Networks
• Graph Sparsification
• Computational Cost
• Structural Distortion

Innovation

• Graph Sparsification
• Support Graph Theory
• Dilation-Congestion Criterion
• Graph Neural Networks
• Topology-based Framework