Research map / Graph learning / Graph self-supervised learning
Size-Invariant Graph Representations for Graph Classification Extrapolations
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 |
|---|---|---|---|
| Use induced-subgraph-density representations with a GNN and attribute regularization instead of standard whole-graph GNNs | Attributed SBM classification · Train size 20, test size 40 | accuracy: 0.65 (0.10) -> 1.00 (0.00) | Helped |
| Use induced-subgraph-density representations with a GNN and attribute regularization instead of standard whole-graph GNNs | Attributed SBM classification · Train sizes 14 and 20, test size 40 | accuracy: 0.57 (0.19) -> 0.95 (0.05) | Helped |
| Use induced-subgraph-density representations with a GNN and attribute regularization instead of standard whole-graph GNNs | Attributed SBM classification · Train sizes 20 and 30, test size 40 | accuracy: 0.69 (0.19) -> 0.93 (0.05) | Helped |
| Use induced-subgraph-density representations with a GNN | Unattributed schizophrenia graphs · Training size 264, smaller-size test | accuracy: 0.61 (0.08) -> 0.72 (0.04) | Helped |
| Use IRM with multiple training graph sizes | Erdős–Rényi graph classification · Train sizes 70 and 80, test size 140 | accuracy: 0.65 (0.12) -> 0.66 (0.08) | No clear effect |
| Train on graph sizes closer to the test size | Attributed SBM classification · Train sizes 20 and 30 versus 14 and 20; test size 40 | accuracy: 0.95 (0.05) -> 0.93 (0.05) | Mixed |
| Apply the E-invariant subgraph representation to real-world OOD graph datasets | Γ_GIN and Γ_RPGIN · NCI1, NCI109, PROTEINS, and DD | MCC | Mixed |
| Use the one-hot subgraph representation instead of attribute-regularized GNN subgraph representations | Attributed SBM classification · Train size 20, test size 40 | accuracy: 1.00 (0.00) -> 0.50 (0.07) | Hurt |
| Use the one-hot subgraph representation instead of attribute-regularized GNN subgraph representations | Attributed SBM classification · Train sizes 14 and 20, test size 40 | accuracy: 0.95 (0.05) -> 0.50 (0.07) | Hurt |
| Use the one-hot subgraph representation instead of attribute-regularized GNN subgraph representations | Attributed SBM classification · Train sizes 20 and 30, test size 40 | accuracy: 0.93 (0.05) -> 0.45 (0.05) | Hurt |
Closest papers by meaning
- Dissecting the Failure of Invariant Learning on Graphs (NeurIPS 2024 · similarity 0.90)
- Disentangled Graph Self-supervised Learning for Out-of-Distribution Generalization (ICML 2024 · similarity 0.88)
- Subgraph Invariant Learning Towards Large-Scale Graph Node Classification (AAAI 2025 · similarity 0.88)
- Graph Contrastive Invariant Learning from the Causal Perspective (AAAI 2024 · similarity 0.88)
- Graph Invariant Learning with Subgraph Co-mixup for Out-of-Distribution Generalization (AAAI 2024 · similarity 0.87)
- Improving Out-of-Distribution Generalization in Graphs via Hierarchical Semantic Environments (CVPR 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.