Research map / Graph learning / GNN architectures
Towards Scale-Invariant Graph-related Problem Solving by Iterative Homogeneous GNNs
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 iterative module (adaptive stopping) (ablation) | Homo-Path · Shortest path lobster 1000 nodes | success rate (%): 50.2 -> 100.0 | Helped |
| add homogeneous prior (ablation) | Path · Weighted shortest path, weights [8, 24) | relative loss: 0.54 -> 0.03 | Helped |
| add homogeneous prior on top of iterative module (ablation) | Iter-Path · Weighted shortest path, weights [8, 24) | relative loss: 0.08 vs Homo-Path 0.03 | Hurt |
| add iterative module (ablation) | Homo-Path · Physical simulation 100 balls | MSE: 21.45 -> 2.01 | Helped |
| add homogeneous prior (ablation) | Iter-Path · Physical simulation 100 balls | MSE: 1.68 -> 2.01 | No clear effect |
| add homogeneous prior (ablation) | Iter-Path · Image-based navigation 33x33 | success rate: 78.6% -> 91.7% | Helped |
| add homogeneous prior (ablation) | Iter-Path · Component counting lobster 500 | accuracy: 96.1% -> 97.5% | Helped |
| replace weight-sharing fixed-depth iteration with adaptive iterative module (ablation) | Shared-Homo-Path · Shortest path lobster 1000 nodes | success rate (%): 91.7 -> 100.0 | Helped |
| add PathGNN layers (Bellman-Ford alignment) in place of GCN/GAT (ablation) | GCN/GAT · Shortest path lobster 1000 nodes | success rate (%): 2.4 -> 10.3 | Helped |
| add adaptive iterative module in place of ACT (ablation) | Homo-Path · Shortest path lobster 1000 nodes | success rate | Helped |
| use deeper 100-layer Path model (ablation) | Path · Weighted shortest path lobster | relative loss: ~0.13 | Hurt |
Closest papers by meaning
- Covered Forest: Fine-grained generalization analysis of graph neural networks (ICML 2025 · similarity 0.88)
- A Manifold Perspective on the Statistical Generalization of Graph Neural Networks (ICML 2025 · similarity 0.88)
- Learning to Reweight for Generalizable Graph Neural Network (AAAI 2024 · similarity 0.87)
- On the Power of Small-size Graph Neural Networks for Linear Programming (NeurIPS 2024 · similarity 0.87)
- Dynamic Rescaling for Training GNNs (NeurIPS 2024 · similarity 0.87)
- Graph neural networks and non-commuting operators (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.