Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

Graph Condensation via Receptive Field Distribution Matching

About

Graph neural networks (GNNs) enable the analysis of graphs using deep learning, with promising results in capturing structured information in graphs. This paper focuses on creating a small graph to represent the original graph, so that GNNs trained on the size-reduced graph can make accurate predictions. We view the original graph as a distribution of receptive fields and aim to synthesize a small graph whose receptive fields share a similar distribution. Thus, we propose Graph Condesation via Receptive Field Distribution Matching (GCDM), which is accomplished by optimizing the synthetic graph through the use of a distribution matching loss quantified by maximum mean discrepancy (MMD). Additionally, we demonstrate that the synthetic graph generated by GCDM is highly generalizable to a variety of models in evaluation phase and that the condensing speed is significantly improved using this framework.

Mengyang Liu, Shanchuan Li, Xinshi Chen, Le Song• 2022

Related benchmarks

TaskDatasetResultRank
Node ClassificationCora
Accuracy77.14
1225
Node ClassificationPhoto semi-supervised (test)
Accuracy81.5
44
Semi-supervised node classificationCora (test)
Accuracy82.5
31
Semi-supervised node classificationarXiv (test)
Accuracy69.5
31
Semi-supervised node classificationComputers (test)
Accuracy84.2
31
Graph CoarseningCora
Running Time (s)142.8
21
Node ClassificationOgbn-arxiv Non-Evolving
Accuracy64.23
20
Graph CoarseningFlickr
Running Time (s)353.5
20
Graph CoarseningREDDIT
Running Time (s)337.1
18
Node ClassificationFlickr (Non-Evolving)
Accuracy46.99
10
Showing 10 of 23 rows

Other info

Follow for update