Papers › Are Powerful Graph Neural Nets Necessary? A Dissection on Graph Classification

Are Powerful Graph Neural Nets Necessary? A Dissection on Graph Classification

11 May 2019arXiv:1905.04579archive 2025-07-28

Ting Chen, Song Bian, Yizhou Sun

Graph Neural Nets (GNNs) have received increasing attentions, partially due to their superior performance in many node and graph classification tasks. However, there is a lack of understanding on what they are learning and how sophisticated the learned graph functions are. In this work, we propose a dissection of GNNs on graph classification into two parts: 1) the graph filtering, where graph-based neighbor aggregations are performed, and 2) the set function, where a set of hidden node features are composed for prediction. To study the importance of both parts, we propose to linearize them separately. We first linearize the graph filtering function, resulting Graph Feature Network (GFN), which is a simple lightweight neural net defined on a \textit{set} of graph augmented features. Further linearization of GFN's set function results in Graph Linear Network (GLN), which is a linear function. Empirically we perform evaluations on common graph classification benchmarks. To our surprise, we find that, despite the simplification, GFN could match or exceed the best accuracies produced by recently proposed GNNs (with a fraction of computation cost), while GLN underperforms significantly. Our results demonstrate the importance of non-linear set function, and suggest that linear graph filtering with non-linear set function is an efficient and powerful scheme for modeling existing graph classification benchmarks.

PaperPDFConference PDFCode

In Syntology Open this paper in Syntology's Atlas, the map of the papers in Syntology's graph and their citations.

Code

chentingpc/gfn officialmentioned in papermentioned on GitHubpytorch report

Repository list and official/mentioned flags are the archive's, frozen 2025-07-28. Reachability, where shown, is from one Syntology probe window (2026-09-16 to 2026-09-18); repositories not probed show nothing. GitHub stars are not tracked.

Code Syntology ran Syntology

Not run by Syntology. Nothing on this page verifies that the listed code works.

Tasks

ClassificationGeneral ClassificationGraph Classification

Results from the paper archive 2025-07-28

TaskDatasetModelMetricValueRank at snapshotLeaderboardReport
Graph Classification COLLAB GFN Accuracy 81.50% #5 of 39 Archive leaderboard report
Graph Classification COLLAB GFN-light Accuracy 81.34% #7 of 39 Archive leaderboard report
Graph Classification D&D GFN Accuracy 78.78% #21 of 53 Archive leaderboard report
Graph Classification D&D GFN-light Accuracy 78.62% #25 of 53 Archive leaderboard report
Graph Classification ENZYMES GFN Accuracy 70.17% #12 of 54 Archive leaderboard report
Graph Classification ENZYMES GFN-light Accuracy 69.50% #14 of 54 Archive leaderboard report
Graph Classification IMDb-B GFN Accuracy 73.00% #35 of 51 Archive leaderboard report
Graph Classification IMDb-B GFN-light Accuracy 73.00% #36 of 51 Archive leaderboard report
Graph Classification IMDb-M GFN Accuracy 51.80% #13 of 36 Archive leaderboard report
Graph Classification IMDb-M GFN-light Accuracy 51.20% #16 of 36 Archive leaderboard report
Graph Classification MUTAG GFN Accuracy 90.84% #18 of 74 Archive leaderboard report
Graph Classification MUTAG GFN-light Accuracy 89.89% #27 of 74 Archive leaderboard report
Graph Classification NCI1 GFN Accuracy 83.65% #26 of 69 Archive leaderboard report
Graph Classification NCI1 GFN-light Accuracy 81.43% #35 of 69 Archive leaderboard report
Graph Classification PROTEINS GFN-light Accuracy 77.44% #32 of 103 Archive leaderboard report
Graph Classification PROTEINS GFN Accuracy 76.46% #45 of 103 Archive leaderboard report
Graph Classification RE-M12K GFN-light Accuracy 49.75% #1 of 6 Archive leaderboard report
Graph Classification RE-M12K GFN Accuracy 49.43% #2 of 6 Archive leaderboard report
Graph Classification RE-M5K GFN-light Accuracy 49.75% #6 of 8 Archive leaderboard report
Graph Classification RE-M5K GFN Accuracy 49.43% #7 of 8 Archive leaderboard report

Ranks are positions in the archive's leaderboards as they stood at the 2025-07-28 snapshot. Results published since then are not among these rows, so a rank here is not a current standing.

Report a problem or propose a change · a person checks every report against the paper or source before anything changes; decisions are listed on /corrections