Research map / Graph learning / GNN architectures
On the Connection Between MPNN and Graph Transformer
Chen Cai, Truong Son Hy, Rose Yu, Yusu Wang
This paper tried a common change: Add positional encoding
add positional encoding to DeepSets-based MPNN+VN · ogbg-ppa / ogbg-molhiv / ogbg-molpcba / ogbg-code2 · accuracy / AUROC / AP / F1: ppa 0.8055 (NoPE) vs 0.8027 (PE); molhiv 0.7676 vs 0.7687; molpcba 0.2823 vs 0.2848; code2 0.1727 vs 0.1719 No clear effect
What 646 accepted papers report for this change
- 525 improved
- 58 no clear effect
- 7 mixed
- 56 hurt
- Learning-Augmented Scalable Linear Assignment Problem Optimization via Neural Dual Warm-Starts ICML 2026 · add positional encodings to features (d=13 -> d=21) · Block-structured, N=4096 · speedupImproved
- Mesh Based Simulations with Spatial and Temporal awareness ICML 2026 · use learned positional embedding, learnable relative bias or weighted adjacency instead of RoPE · Aneurysm · RMSEHurt
- Sparse ActionGen: Accelerating Diffusion Policy with Real-time Pruning ICML 2026 · add sinusoidal positional encoding to the pruner · Robomimic Tool Hang · success rate (%): 28 -> 54Improved
See this change in detailMatched by the wording of each paper's own result, read by AI from the full text. Papers report gains more often than failures, so this shows what gets published, not the chance a change works.
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add virtual node to MPNN (GCN, GINE, GatedGCN) | MPNN on PEPTIDES-FUNC and PEPTIDES-STRUCT · LRGB | Test AP / Test MAE: GCN AP 0.5930 -> 0.6623; GatedGCN+RWSE AP 0.6069 -> 0.6685 | Improved |
| add virtual node to GatedGCN+RWSE | GatedGCN+RWSE on PEPTIDES-FUNC · LRGB PEPTIDES-FUNC | Test AP: 0.6069 -> 0.6685 | Improved |
| add positional encoding to DeepSets-based MPNN+VN (ablation) | MPNN + VN on OGB graph datasets · ogbg-ppa / ogbg-molhiv / ogbg-molpcba / ogbg-code2 | accuracy / AUROC / AP / F1: ppa 0.8055 (NoPE) vs 0.8027 (PE); molhiv 0.7676 vs 0.7687; molpcba 0.2823 vs 0.2848; code2 0.1727 vs 0.1719 | No clear effect |
| replace GIN+VN virtual node with DeepSets-based MPNN+VN (no PE) | GIN+virtual node on OGB · ogbg-ppa | Accuracy: 0.7037 -> 0.8055 | Improved |
| DeepSets-based MPNN+VN (no PE) vs GIN+VN | PCQM4Mv2 validation · PCQM4Mv2 | MAE: GIN-virtual 0.1083 -> MPNN+VN+NoPE medium 0.0889 | Improved |
| add virtual node to MPNN for sea surface temperature forecasting (ablation) | MPNN on SST · SST 4/2/1 week | MSE: 4wk 0.2917 -> 0.2806; 2wk 0.2281 -> 0.2130; 1wk 0.1613 -> 0.1540 | Improved |
| MPNN+VN vs Linear Transformer+LapPE on SST | MPNN + VN · SST 4/2/1 week | MSE: 4wk 0.2806 vs 0.2818 (better); 2wk 0.2130 vs 0.2191 (better); 1wk 0.1540 vs 0.1610 (better); but TF-Net is best at 2 and | Mixed |
| add virtual node to GCN on PEPTIDES-STRUCT | GCN · LRGB PEPTIDES-STRUCT | Test MAE: 0.3496 -> 0.2488 (lower is better) | Improved |
Closest papers by meaning
- Distinguished In Uniform: Self-Attention Vs. Virtual Nodes (ICLR 2024 · similarity 0.88)
- Classic GNNs are Strong Baselines: Reassessing GNNs for Node Classification (NeurIPS 2024 · similarity 0.88)
- Can Classic GNNs Be Strong Baselines for Graph-level Tasks? Simple Architectures Meet Excellence (ICML 2025 · similarity 0.88)
- Polynormer: Polynomial-Expressive Graph Transformer in Linear Time (ICLR 2024 · similarity 0.87)
- Probabilistic Graph Rewiring via Virtual Nodes (NeurIPS 2024 · similarity 0.87)
- Accurate and Scalable Graph Neural Networks via Message Invariance (ICLR 2025 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
