Research map / Graph learning / GNN architectures
MeGraph: Capturing Long-Range Interactions by Alternating Local and Hierarchical Aggregation on Multi-Scaled Graph Hierarchy
Honghua Dong, Jiawei Xu, Yu Yang, Rui Zhao, Shiwen Wu, Chun Yuan, Xiu Li, Chris J. Maddison, Lei Han
Read the paperCheck your idea against 137,217 papers
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 graph hierarchy (h=1 -> h=5) with S-EdgePool, n=5 (ablation) | MeGraph · Graph Theory Benchmark (medium) | avg MSE: 8.440 -> 0.675 | Improved |
| repeat information exchange over mega graph (n=1 -> n=5) at h=5 (ablation) | MeGraph with EdgePool · Graph Theory Benchmark (medium) | avg MSE: 5.368 -> 2.215 | Improved |
| increase layers of flat GNN without hierarchy (n=5 -> n=10) (ablation) | MeGraph h=1 · Graph Theory Benchmark (medium) | avg MSE: 8.440 -> 7.006 | Improved |
| switch EdgePool (tau_c=2) to S-EdgePool with eta_v=0.3, tau_c=4 (ablation) | MeGraph · Graph Theory Benchmark (medium) | avg MSE: 2.215 -> 0.675 | Improved |
| replace Graph-UNets with MeGraph | MeGraph vs Graph-UNets · Graph Theory Benchmark (medium) | avg MSE: 1.581 -> 0.675 | Improved |
| replace X-Conv cross update with X-Pool (unpool) (ablation) | MeGraph · Graph Theory Benchmark (medium) | avg MSE: 0.675 -> 1.203 | Hurt |
| remove bidirectional pathway (w/o pw) (ablation) | MeGraph · Graph Theory Benchmark (medium) | avg MSE: 0.675 -> 1.065 | Hurt |
| increase node pooling ratio eta_v from 0.3 to 0.5 (ablation) | MeGraph S-EdgePool · Graph Theory Benchmark (medium) | avg MSE: 0.675 -> 1.197 | Hurt |
| random pooling instead of EdgePool/Louvain (ablation) | MeGraph · TreeCycle and TreeGrid | node classification accuracy | Hurt |
| add MeGraph on large-size graphs vs h=1 baseline | MeGraph · Graph Theory Benchmark (large) | avg MSE: 220.169 -> 23.9676 | Improved |
| repeat local aggregation (n=1 -> MeGraph) on CLUSTER (ablation) | MeGraph · CLUSTER | accuracy: 50.396 -> 68.603 | Improved |
| increase n from 1 to 4 at h=9 (ablation) | MeGraph · LRGB Peptide-func | AP: 67.52 -> 69.45 | Improved |
Closest papers by meaning
- On Measuring Long-Range Interactions in Graph Neural Networks (ICML 2025 · similarity 0.88)
- ENAHPool: The Edge-Node Attention-based Hierarchical Pooling for Graph Neural Networks (ICML 2025 · similarity 0.87)
- Adaptive Message Passing: A General Framework to Mitigate Oversmoothing, Oversquashing, and Underreaching (ICML 2025 · similarity 0.87)
- Improving Graph Neural Networks by Learning Continuous Edge Directions (ICLR 2025 · similarity 0.86)
- Relieving the Over-Aggregating Effect in Graph Transformers (NeurIPS 2025 · similarity 0.86)
- PANDA: Prior-guided Attentional Dual-path Architecture (NeurIPS 2026 · similarity 0.86)
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.
