GNNBenchmark
CSL
Structure
150Graphs
41Nodes
164Edges
10Classes
Coverage
102Results
1Variants
19Papers
- Message Passing on the Edge: Towards Scalable and Expressive GNNs Message Passing on the Edge: Towards Scalable and Expressive GNNs
- Efficient Mixed Precision Quantization in Graph Neural Networks Efficient Mixed Precision Quantization in Graph Neural Networks
- WWW 2025: Graph Self-Supervised Learning with Learnable Structural and Positional Encodings Graph Self-Supervised Learning with Learnable Structural and Positional Encodings WWW 2025
- ICLR 2025: Learning Efficient Positional Encodings with Graph Neural Networks Learning Efficient Positional Encodings with Graph Neural Networks ICLR 2025
- ICML 2025: Beyond Message Passing: Neural Graph Pattern Machine Beyond Message Passing: Neural Graph Pattern Machine ICML 2025
- TMLR 2024: Towards Graph Foundation Models: A Study on the Generalization of Positional and Structural Encodings Towards Graph Foundation Models: A Study on the Generalization of Positional and Structural Encodings TMLR 2024
- ICLR 2024: Revisiting Random Walks for Learning on Graphs Revisiting Random Walks for Learning on Graphs ICLR 2024
- ICML 2024: Comparing Graph Transformers via Positional Encodings Comparing Graph Transformers via Positional Encodings ICML 2024
- Topology-Informed Graph Transformer Topology-Informed Graph Transformer
- ICLR 2024: PF-GNN: Differentiable particle filtering based approximation of universal graph representations PF-GNN: Differentiable particle filtering based approximation of universal graph representations ICLR 2024
- ICLR 2023: Efficient Subgraph GNNs by Learning Effective Selection Policies Efficient Subgraph GNNs by Learning Effective Selection Policies ICLR 2023
- NeurIPS 2023: MAG-GNN: Reinforcement Learning Boosted Graph Neural Network MAG-GNN: Reinforcement Learning Boosted Graph Neural Network NeurIPS 2023
- KDD 2023: An Efficient Subgraph GNN with Provable Substructure Counting Power An Efficient Subgraph GNN with Provable Substructure Counting Power KDD 2023
- JMLR 2023: GNNBenchmark GNNBenchmark JMLR 2023
- ICML 2022: A Generalization of ViT/MLP-Mixer to Graphs A Generalization of ViT/MLP-Mixer to Graphs ICML 2022
- NeurIPS 2021: Reconstruction for Powerful Graph Representations Reconstruction for Powerful Graph Representations NeurIPS 2021
- NeurIPS 2021: Weisfeiler and Lehman Go Cellular: CW Networks Weisfeiler and Lehman Go Cellular: CW Networks NeurIPS 2021
- ICML 2020: Graph Homomorphism Convolution Graph Homomorphism Convolution ICML 2020
- NeurIPS 2019: On the Equivalence between Graph Isomorphism Testing and Function Approximation with GNNs On the Equivalence between Graph Isomorphism Testing and Function Approximation with GNNs NeurIPS 2019
10-class graph classification Circular skip link graphs — tests graph isomorphism discrimination (10 classes). Uses the benchmark-provided synthetic or superpixel node/edge features. Evaluated by Accuracy.
Dataset details
Task
Circular skip link graphs — tests graph isomorphism discrimination (10 classes). The primary catalog metric is Accuracy.
Data
Circular skip link graphs — tests graph isomorphism discrimination (10 classes).
Features
The benchmark release supplies fixed graph tensors and task-specific node or edge inputs. Image graphs use superpixel appearance and position, synthetic node tasks use generated signals, TSP uses coordinates, and CSL is intentionally featureless.
Splits and evaluation
The catalog records these protocols or variants: Standard split. Evaluation uses Accuracy (higher is better). Exact masks or folds must come from the cited release.
Quirks and pitfalls
Use the released train/validation/test files rather than resplitting. These tasks were designed to expose specific architectural limitations, and preprocessing such as changing superpixels, adding positional encodings, or balancing edge labels changes the benchmark.
Sources
SOTA Progression
new SOTA
matches SOTA at that time
other proposed result
other baseline result
Leaderboard
| # | Model | Accuracy ↑ | Paper |
|---|---|---|---|
| 1 | PPGN | 1.0000 | |
| 2 | RW-AgentNet | 1.0000 | |
| 3 | AgentNet | 1.0000 | |
| 4 | CRaWl | 1.0000 | |
| 5 | RWNN-DeBERTa | 1.0000 | |
| 6 | GHC-Cycles | 1.0000
±
0.0000 | |
| 7 | Reconstr. (n-1) | 1.0000
±
0.0000 | |
| 8 | NGNN | 1.0000 | |
| 9 | GNNAK+ | 1.0000 | |
| 10 | SSWL+ | 1.0000 | |
| 11 | RNM | 1.0000 | |
| 12 | I^2GNN | 1.0000 | |
| 13 | MAG-GNN | 1.0000 | |
| 14 | Full | 1.0000
±
0.0000 | |
| 15 | Random T=2 | 1.0000
±
0.0000 | |
| 16 | MPNNs | 1.0000
±
0.0000 | |
| 17 | I^2-GNN | 1.0000 | |
| 18 | ESC-GNN | 1.0000 | |
| 19 | PF-GNN | 1.0000
±
0.0000 | |
| 20 | GCN | 1.0000
±
0.0000 | |
| 21 | APE-GT | 1.0000
±
0.0000 | |
| 22 | DeepSet RPE-GT | 1.0000
±
0.0000 | |
| 23 | RPE-augWL | 1.0000
±
0.0000 | |
| 24 | EGN APE-GT | 1.0000
±
0.0000 | |
| 25 | GPS+LapPE+RWSE (1 layer) | 1.0000
±
0.0000 | |
| 26 | TIGT | 1.0000
±
0.0000 | |
| 27 | CIN | 1.0000
±
0.0000 | |
| 28 | B-PEARL | 1.0000 | |
| 29 | R-PEARL | 1.0000 | |
| 30 | Graph ViT/MLP-Mixer | 1.0000 | |
| 31 | Graph ViT | 1.0000 | |
| 32 | GPM | 1.0000 | |
| 33 | MoNet | 0.9997
±
0.0033 | |
| 34 | GraphSAGE | 0.9993
±
0.0047 | |
| 35 | GAT | 0.9993
±
0.0047 | |
| 36 | GatedGCN | 0.9960
±
0.0108 | |
| 37 | GatedGCN (Rand sign(EigVecs)) | 0.9960
±
0.0108 | |
| 38 | GatedGCN (Abs(EigVecs)) | 0.9943
±
0.0113 | |
| 39 | FP32 | 0.9940
±
0.0130 | |
| 40 | GIN | 0.9933
±
0.0133 | |
| 41 | Positional GIN | 0.9933
±
0.0133 | |
| 42 | StructPosGSSL-SA | 0.9860
±
0.0280 | |
| 43 | Cy2C-GIN-1 | 0.9833
±
0.0333 | |
| 44 | StructPosGSSL-FA | 0.9830
±
0.0250 | |
| 45 | PPGN (3-WL) | 0.9780
±
0.1091 | |
| 46 | 3-WL GNN | 0.9780
±
0.1090 | |
| 47 | 3WLGNN | 0.9780
±
0.1092 | |
| 48 | 3-GCN | 0.9570
±
0.1480 | |
| 49 | 3-GNN | 0.9570 | |
| 50 | MixQ - (λ=-ε) | 0.9500
±
0.0510 |
SOTA performance
set a new SOTA at the time
matched the SOTA at the time