Research map / Graph learning / GNN architectures
A Generalization of ViT/MLP-Mixer to Graphs
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 graph patch extraction with METIS partitioning and one-hop extension (ablation) | GCN-MLP-Mixer · ZINC | MAE: 0.2495 ± 0.0040 -> 0.1347 ± 0.0020 | Helped |
| Add graph patch extraction with METIS partitioning and one-hop extension (ablation) | GCN-MLP-Mixer · Peptides-func | AP: 0.6341 ± 0.0139 -> 0.6832 ± 0.0061 | Helped |
| Replace METIS partitioning with random node partitioning (ablation) | Graph MLP-Mixer · ZINC | MAE: 0.1347 ± 0.0020 -> 0.1435 ± 0.0122; 0.0733 ± 0.0014 -> 0.0708 ± 0.0020 | Mixed |
| Increase the number of patches from 2 to 16 (ablation) | GCN-MLP-Mixer · EXP | Accuracy: 57.54 ± 3.87 -> 100.00 ± 0.00 | Helped |
| Add node positional encoding (ablation) | GINE-MLP-Mixer · ZINC | MAE: 0.1225 ± 0.0070 -> 0.0733 ± 0.0014 | Helped |
| Add node positional encoding (ablation) | GINE-MLP-Mixer · Peptides-func | AP: 0.6868 ± 0.0034 -> 0.6970 ± 0.0080 | Helped |
| Add patch positional encoding (ablation) | GINE-MLP-Mixer · ZINC | MAE: 0.0746 ± 0.0010 -> 0.0733 ± 0.0014 | Helped |
| Use one-hop-overlapping patches instead of non-overlapping patches (ablation) | Graph MLP-Mixer · ZINC and Peptides-func | performance | Helped |
| Extend patches beyond one hop (ablation) | Graph MLP-Mixer · ZINC and Peptides-func | performance | Mixed |
| Use standard full attention instead of Hadamard attention (ablation) | Graph ViT · ZINC | MAE: 0.0849 ± 0.0047 -> 0.1784 ± 0.0238 | Hurt |
| Use standard full attention instead of Hadamard attention (ablation) | Graph ViT · Peptides-func | AP: 0.6919 ± 0.0085 -> 0.6778 ± 0.0039 | Hurt |
| Use Graph MLP-Mixer rather than the base GINE model | GINE · Peptides-func | AP: 0.6405 ± 0.0077 -> 0.6970 ± 0.0080 | Helped |
Closest papers by meaning
- VQGraph: Rethinking Graph Representation Space for Bridging GNNs and MLPs (ICLR 2024 · similarity 0.89)
- Generalizing CNNs to graphs with learnable neighborhood quantization (NeurIPS 2024 · similarity 0.88)
- NN-Former: Rethinking Graph Structure in Neural Architecture Representation (CVPR 2025 · similarity 0.88)
- Graph Neural Networks Are More Than Filters: Revisiting and Benchmarking from A Spectral Perspective (ICLR 2025 · similarity 0.88)
- On the Stability of Graph Convolutional Neural Networks: A Probabilistic Perspective (NeurIPS 2025 · similarity 0.87)
- A Flexible, Equivariant Framework for Subgraph GNNs via Graph Products and Graph Coarsening (NeurIPS 2024 · 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.