Research map / Graph learning / Graph self-supervised learning
Learning Causally Invariant Representations for Out-of-Distribution Generalization on 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 |
|---|---|---|---|
| use CIGA (invariant subgraph with contrastive + hinge objectives) instead of ERM | GNN graph classifier · SPMotif Struc/Mixed avg | accuracy: 52.24 -> 67.92 | Helped |
| use CIGAv2 instead of ERM on CMNIST-sp | GNN · CMNIST-sp | accuracy: 13.96 -> 44.91 | Helped |
| use CIGAv2 instead of ERM on Drug-Assay | GNN · DrugOOD Assay (ROC-AUC) | accuracy: 71.79 -> 73.17 | Helped |
| use CIGAv2 instead of ERM on Twitter | GNN · Twitter | accuracy: 60.81 -> 64.45 | Helped |
| add CIGA's decomposition/objective vs. plain interpretable GNN DIR | DIR · Drug-Assay | accuracy: 68.25 -> 73.17 | Helped |
| use interpretable GNN DIR instead of ERM | ERM · SPMotif-Mixed bias=0.90 | accuracy: 41.36 -> 38.58 | Hurt |
| use contrastive objective CNC alone instead of CIGA decomposition | CIGA · SPMotif-Struc bias=0.90 | accuracy: 63.41 -> 50.25 | Hurt |
| use CIGAv2 (estimates s_c) instead of CIGAv1 (assumes known s_c) | CIGAv1 · SPMotif-Struc bias=0.90 | accuracy: 51.78 -> 63.41 | Helped |
| use CIGAv1 instead of CIGAv2 with accurate s_c (ablation) | CIGAv2 · PROTEINS (MCC) | accuracy: 0.31 -> 0.40 | Mixed |
| make contrastive coefficient alpha too small (ablation) | CIGA · SPMotif-Mixed / DrugOOD-Scaffold / NCI109 | accuracy | Hurt |
| make hinge loss coefficient beta too small (ablation) | CIGA · SPMotif-Mixed / DrugOOD-Scaffold / NCI109 | accuracy | Hurt |
| make alpha and beta too large (ablation) | CIGA · hardest datasets | accuracy | Hurt |
Closest papers by meaning
- Dissecting the Failure of Invariant Learning on Graphs (NeurIPS 2024 · similarity 0.94)
- Quantifying Distributional Invariance in Causal Subgraph for IRM-Free Graph Generalization (NeurIPS 2025 · similarity 0.93)
- Out-of-Distribution Generalization on Graphs via Progressive Inference (AAAI 2025 · similarity 0.92)
- Graph Contrastive Invariant Learning from the Causal Perspective (AAAI 2024 · similarity 0.90)
- Subgraph Aggregation for Out-of-Distribution Generalization on Graphs (AAAI 2025 · similarity 0.90)
- Empowering Graph Invariance Learning with Deep Spurious Infomax (ICML 2024 · similarity 0.90)
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.