Exact Acceleration of Subgraph Graph Neural Networks by Eliminating Computation Redundancy

TL;DR

Subgraph GNNs enhance expressiveness but suffer from explosive storage and computation due to大量 subgraphs. ENFA addresses this by performing convolutions on the original graph and compact ego nets around pivot nodes, replacing redundant subgraph computations with original-graph embeddings to achieve exact outputs. It formalizes pivot nodes and pivot hops, guarantees identical results after the final layer, and extends to subgraph-message-passing variants. Empirical results across diverse benchmarks show substantial storage reductions (up to 84.5%) and up to 1.66x training speedups, while preserving or improving predictive performance, indicating strong practical impact for scalable, expressive graph learning.

Abstract

Graph neural networks (GNNs) have become a prevalent framework for graph tasks. Many recent studies have proposed the use of graph convolution methods over the numerous subgraphs of each graph, a concept known as subgraph graph neural networks (subgraph GNNs), to enhance GNNs' ability to distinguish non-isomorphic graphs. To maximize the expressiveness, subgraph GNNs often require each subgraph to have equal size to the original graph. Despite their impressive performance, subgraph GNNs face challenges due to the vast number and large size of subgraphs which lead to a surge in training data, resulting in both storage and computational inefficiencies. In response to this problem, this paper introduces Ego-Nets-Fit-All (ENFA), a model that uniformly takes the smaller ego nets as subgraphs, thereby providing greater storage and computational efficiency, while at the same time guarantees identical outputs to the original subgraph GNNs even taking the whole graph as subgraphs. The key is to identify and eliminate the redundant computation among subgraphs. For example, a node $v_i$ may appear in multiple subgraphs but is far away from all of their centers (the unsymmetric part between subgraphs). Therefore, its first few rounds of message passing within each subgraph can be computed once in the original graph instead of being computed multiple times within each subgraph. Such strategy enables our ENFA to accelerate subgraph GNNs in an exact way, unlike previous sampling approaches that often lose the performance. Extensive experiments across various datasets reveal that compared with the conventional subgraph GNNs, ENFA can reduce storage space by 29.0% to 84.5% and improve training efficiency by up to 1.66x.

Figures (10)

• Figure 1: Shared part produces identical outputs.
• Figure 2: Redundant computations of subgraph GNNs.
• Figure 3: An example workflow of ENFA for node-marking policy.
• Figure 4: Experimental Results on Storage Efficiency.
• Figure 5: Experimental Results on Computation Efficiency.

Theorems & Definitions (13)

• Theorem 1
• Theorem 2
• Theorem 1
• Proposition 1
• Lemma 1
• Proof
• Theorem 2
• Proposition 2
• Proof
• Lemma 2