Research map / Graph learning / GNN architectures
Memory-Based Graph Networks
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 |
|---|---|---|---|
| replace local adjacency-based query information with global RWR topology embeddings | GMN · graph classification benchmarks | accuracy | Mixed |
| use edge features in attention (e-GAT instead of GAT) (ablation) | MemGNN query network · ESOL, Lipophilicity, BACE | validation performance | Helped |
| use RWR instead of adjacency or normalized adjacency as topology embedding (ablation) | GMN · Enzymes | accuracy: 78.66%, 77.16%, and 77.33% | Mixed |
| sort topological embeddings to ensure permutation invariance (ablation) | GMN · DD | accuracy: 82.24% to 84.40% | Helped |
| randomly sample 10% of edges instead of selecting top edges by RWR (ablation) | MemGNN · Collab | accuracy: 73.1% to 73.9% | Helped |
| increase the number of keys from 8 to 64 (ablation) | MemGNN · ESOL | RMSE: 0.56, 0.52 | Helped |
| increase the number of keys from 64 to 160 (ablation) | MemGNN · ESOL | RMSE: 0.52 to 0.54 | Hurt |
| use 5 heads with 32 keys instead of 1 head with 160 keys at fixed parameter count (ablation) | MemGNN · ESOL | RMSE: 0.54 to 0.53 | Helped |
| initialize memory keys with K-Means rather than random initialization (ablation) | MemGNN · graph benchmarks | performance | No clear effect |
| use GMN rather than MemGNN with local message passing | proposed graph models · Collab | accuracy: 80.18 vs 77.0 | Mixed |
Closest papers by meaning
- Fast and Deep Graph Neural Networks (AAAI 2020 · similarity 0.89)
- Distance Encoding: Design Provably More Powerful Neural Networks for Graph Representation Learning (NeurIPS 2020 · similarity 0.89)
- Counting Graph Substructures with Graph Neural Networks (ICLR 2024 · similarity 0.89)
- On the Scalability of GNNs for Molecular Graphs (NeurIPS 2024 · similarity 0.88)
- Pre-Training Graph Neural Networks on Molecules by Using Subgraph-Conditioned Graph Information Bottleneck (AAAI 2025 · similarity 0.87)
- Can Classic GNNs Be Strong Baselines for Graph-level Tasks? Simple Architectures Meet Excellence (ICML 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.