mlsanityData to 7 Oct 2026

Can Classic GNNs Be Strong Baselines for Graph-level Tasks? Simple Architectures Meet Excellence

arXiv 2502.09263 ↗First posted 13 Feb 2025cs.LGcited by 24

Not on the map. Why?

Abstract (arXiv)

Message-passing Graph Neural Networks (GNNs) are often criticized for their limited expressiveness, issues like over-smoothing and over-squashing, and challenges in capturing long-range dependencies. Conversely, Graph Transformers (GTs) are regarded as superior due to their employment of global attention mechanisms, which potentially mitigate these challenges. Literature frequently suggests that GTs outperform GNNs in graph-level tasks, especially for graph classification and regression on small molecular graphs. In this study, we explore the untapped potential of GNNs through an enhanced framework, GNN+, which integrates six widely used techniques: edge feature integration, normalization, dropout, residual connections, feed-forward networks, and positional encoding, to effectively tackle graph-level tasks. We conduct a systematic re-evaluation of three classic GNNs (GCN, GIN, and GatedGCN) enhanced by the GNN+ framework across 14 well-known graph-level datasets. Our results reveal that, contrary to prevailing beliefs, these classic GNNs consistently match or surpass the performance of GTs, securing top-three rankings across all datasets and achieving first place in eight. Furthermore, they demonstrate greater efficiency, running several times faster than GTs on many datasets. This highlights the potential of simple GNN architectures, challenging the notion that complex mechanisms in GTs are essential for superior graph-level performance. Our source code is available at https://github.com/LUOyk1999/GNNPlus.

Key facts. Cited by 24 papers in the map (data through 7 Oct 2026). Builds on 73 of its 91 references that are in the map. First posted on arXiv 13 Feb 2025.

Builds on

73 of 91 references.

References · 73

  1. 2017Attention Is All You Needcited by 59,932
  2. 2015Deep Residual Learning for Image Recognitioncited by 51,809
  3. 2016Semi-Supervised Classification with Graph Convolutional Networkscited by 11,282
  4. 2017Graph Attention Networkscited by 8,854
  5. 2015Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shiftcited by 6,979
  6. 2016Densely Connected Convolutional Networkscited by 6,691
  7. 2017Inductive Representation Learning on Large Graphscited by 6,284
  8. 2016Layer Normalizationcited by 5,579
  9. 2018How Powerful are Graph Neural Networks?cited by 4,355
  10. 2017Neural Message Passing for Quantum Chemistrycited by 3,460
  11. 2019A Comprehensive Survey on Graph Neural Networkscited by 2,906
  12. 2019Fast Graph Representation Learning with PyTorch Geometriccited by 2,414
38 more
  1. 2020Open Graph Benchmark: Datasets for Machine Learning on Graphscited by 2,148
  2. 2018Deeper Insights into Graph Convolutional Networks for Semi-Supervised Learningcited by 1,099
  3. 2018Pitfalls of Graph Neural Network Evaluationcited by 994
  4. 2016Revisiting Semi-Supervised Learning with Graph Embeddingscited by 970
  5. 2019Strategies for Pre-training Graph Neural Networkscited by 964
  6. 2015Convolutional Networks on Graphs for Learning Molecular Fingerprintscited by 751
  7. 2020TUDataset: A collection of benchmark datasets for learning with graphscited by 718
  8. 2022Recipe for a General, Powerful, Scalable Graph Transformercited by 702
  9. 2020Benchmarking Graph Neural Networkscited by 672
  10. 2020A Generalization of Transformer Networks to Graphscited by 653
  11. 2020On the Bottleneck of Graph Neural Networks and its Practical Implicationscited by 624
  12. 2019Measuring and Relieving the Over-smoothing Problem for Graph Neural Networks from the Topological Viewcited by 544
  13. 2019DeepGCNs: Can GCNs Go as Deep as CNNs?cited by 527
  14. 2019Graph Transformer Networkscited by 520
  15. 2021Rethinking Graph Transformers with Spectral Attentioncited by 468
  16. 2017Residual Gated Graph ConvNetscited by 341
  17. 2023A critical look at the evaluation of GNNs under heterophily: Are we really making progress?cited by 332
  18. 2021Graph Neural Networks with Learnable Structural and Positional Representationscited by 326
  19. 2016Learning Convolutional Neural Networks for Graphscited by 314
  20. 2022Long Range Graph Benchmarkcited by 258
  21. 2019A Fair Comparison of Graph Neural Networks for Graph Classificationcited by 244
  22. 2022Representing Long-Range Context for Graph Neural Networks with Global Attentioncited by 193
  23. 2022Structure-Aware Transformer for Graph Representation Learningcited by 191
  24. 2020Graph-Bert: Only Attention is Needed for Learning Graph Representationscited by 176
  25. 2021Are we really making much progress? Revisiting, benchmarking, and refining heterogeneous graph neural networkscited by 176
  26. 2023Exphormer: Sparse Transformers for Graphscited by 166
  27. 2023Graph Inductive Biases in Transformers without Message Passingcited by 150
  28. 2022NAGphormer: A Tokenized Graph Transformer for Node Classification in Large Graphscited by 137
  29. 2024Graph-Mamba: Towards Long-Range Graph Sequence Modeling with Selective State Spacescited by 132
  30. 2023Rethinking the Expressive Power of GNNs via Graph Biconnectivitycited by 119
  31. 2020GraphNorm: A Principled Approach to Accelerating Graph Neural Network Trainingcited by 116
  32. 2021Global Self-Attention as a Replacement for Graph Convolutioncited by 113
  33. 2022Transformer for Graphs: An Overview from Architecture Perspectivecited by 107
  34. 2023Specformer: Spectral Graph Neural Networks Meet Transformerscited by 102
  35. 2022A Generalization of ViT/MLP-Mixer to Graphscited by 93
  36. 2023DIFFormer: Scalable (Graph) Transformers Induced by Energy Constrained Diffusioncited by 85
  37. 2024Graph Mamba: Towards Learning on Graphs with State Space Modelscited by 84
  38. 2023Where Did the Gap Go? Reassessing the Long-Range Graph Benchmarkcited by 83

50 most cited of 73 All 73 (JSON)

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2025: 9; 2026: 15.

24 papers.

Sort:

By area:

  1. 2025Turning Tabular Foundation Models into Graph Foundation Modelscited by 30
  2. 2025GraphPFN: A Prior-Data Fitted Graph Foundation Modelcited by 20
  3. 2026Learning Rate Matters: Vanilla LoRA May Suffice for LLM Fine-tuningcited by 9
  4. 2026On the Necessity of Learnable Sheaf Laplacianscited by 5
  5. 2026A Fair Evaluation of Graph Foundation Models for Node Property Predictioncited by 4
  6. 2026Graph Tokenization for Bridging Graphs and Transformerscited by 3
  7. 2025Early-Exit Graph Neural Networkscited by 2
  8. 2025Unifying and Enhancing Graph Transformers via a Hierarchical Mask Frameworkcited by 2
  9. 2026Fixed Aggregation Features Can Rival GNNscited by 2
  10. 2026Node4All: Learning Node Representation Beyond Datasetscited by 2
  11. 2026Cluster Attention for Graph Machine Learningcited by 1
  12. 2026End-to-End Subgraph Detection with GraphDETRcited by 1
12 more
  1. 2026Benchmarking Sheaf Neural Networks for Inductive Taskscited by 1
  2. 2025Fractal Graph Contrastive Learningcited by 0
  3. 2025Reliable Microservice Tail Latency Prediction via Decoupled Dual-Stream Learning and Gradient Modulationcited by 0
  4. 2025Graph Learning via Logic-Based Weisfeiler-Leman Variants and Tabularizationcited by 0
  5. 2025Inferring the Graph Structure of Images for Graph Neural Networkscited by 0
  6. 2025Training Diverse Graph Experts for Ensembles: A Systematic Empirical Studycited by 0
  7. 2026Enhanced Graph Transformer with Serialized Graph Tokenscited by 0
  8. 2026Local Message-Passing for Discrete Graph Generationcited by 0
  9. 2026Latent-Hysteresis Graph ODEs: Modeling Coupled Topology-Feature Evolution via Continuous Phase Transitionscited by 0
  10. 2026Message Tuning Outshines Graph Prompt Tuning: A Prismatic Space Perspectivecited by 0
  11. 2026Semi-Supervised Learning for Molecular Graphs via Ensemble Consensuscited by 0
  12. 2026PreGS: A Parameter-Transfer-Based Multi-Expert Graph Neural Network for Node Classificationcited by 0