Research map / Graph learning / Graph self-supervised learning
Graph out-of-distribution generalization
Learn invariant graph representations for out-of-distribution generalization and detection.
58 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2021–2026) · an approach in Graph self-supervised learning, keyword cluster “detection · anomaly · ood”
State History Tried Baselines Checklist Preprints Who Export
Where it stands
Not enough benchmark data Fewer than four papers report the same benchmark, so there is no record to follow.
21 papers in 2023–24 and 31 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic shrank, from 6.1% to 5.3%.
Open questions the data points to
- Papers report GNNSafe on Amazon-Photo (FPR95) anywhere from 2.1 to 22.39: comparisons against it depend on whose number you take.
- Papers report GNNSafe on Coauthor-CS (FPR95) anywhere from 3.18 to 16.64: comparisons against it depend on whose number you take.
How this is decided
A benchmark is tracked when at least four papers of this approach report their own method on the same dataset and metric in at least two different years. Its record is the best number reported up to 2024 compared with the best reported in 2025–26: still improving means at least 1% better, flat means at most 0.2% better with at least two papers from 2025–26 trying. Numbers on a different scale from the rest (a fraction among percentages) and far outliers are left out. Results come from different papers with different setups, so they are signals, not head-to-head comparisons. The share of papers is counted at ICML, NeurIPS, ICLR, CVPR and AAAI.
How it got here
Most cited paper of each year
- 2021 · Size-Invariant Graph Representations for Graph Classification Extrapolations (ICML 2021 · 9 citations)
- 2022 · Learning Causally Invariant Representations for Out-of-Distribution Generalization on Graphs (NeurIPS 2022 · 47 citations)
- 2023 · DrugOOD: Out-of-Distribution Dataset Curator and Benchmark for AI-Aided Drug Discovery – a Focus on Affinity Prediction Problems with Noise Annotations (AAAI 2023 · 48 citations)
- 2024 · GOODAT: Towards Test-Time Graph Out-of-Distribution Detection (AAAI 2024 · 23 citations)
- 2025 · Structural Entropy Guided Unsupervised Graph Out-Of-Distribution Detection (AAAI 2025 · 5 citations)
- 2026 · LLM-Enhanced Energy Contrastive Learning for Out-of-Distribution Detection in Text-Attributed Graphs (AAAI 2026 · 1 citation)
Earliest papers here
- Size-Invariant Graph Representations for Graph Classification Extrapolations (ICML 2021 · 9 citations)
- Learning Causally Invariant Representations for Out-of-Distribution Generalization on Graphs (NeurIPS 2022 · 47 citations)
- Handling Distribution Shifts on Graphs: An Invariance Perspective (ICLR 2022 · 30 citations)
What has been tried
37 papers here report 402 measured changes (a component added, removed or swapped, compared with a baseline): helped 66%, no clear effect 5%, hurt 22%, mixed 6%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| Add masked graph GIB loss to subgraph GIB loss | BZR/COX2 · AUC (%) | 39.10{\scriptstyle±1.59} -> 83.06{\scriptstyle±1.45} | Helped | GOODAT: Towards Test-Time Graph Out-of-Distribution Detection (AAAI 2024 · 23 citations) |
| Add both environment Mixup and invariant Mixup | Graph-Twitter, DD, and BBBP · performance | 4.6% | Helped | Graph Invariant Learning with Subgraph Co-mixup for Out-of-Distribution Generalization (AAAI 2024 · 23 citations) |
| Use inferred hierarchical environments instead of two randomly assigned environments | DrugOOD IC50-SCA · ROC-AUC | 68.54±0.64 -> 70.72±0.30 | Helped | Improving Out-of-Distribution Generalization in Graphs via Hierarchical Semantic Environments (CVPR 2024 · 6 citations) |
| Remove local-level contrastive loss | OOD dataset pairs (10 pairs) · AUC | 85.02±0.94 -> 87.71±1.35; 87.55±0.13 -> 88.34±1.64; 78.99±2.81 -> 79.21±4.55 | Mixed | Structural Entropy Guided Unsupervised Graph Out-Of-Distribution Detection (AAAI 2025 · 5 citations) |
| Add non-causal representation permutation | CMNIST-75sp, CFashion-75sp, and CKuzushiji-75sp, bias 0.9 · accuracy (%) | ↓0.74, ↓1.22, ↓0.72 | Helped | Out-of-Distribution Generalization on Graphs via Progressive Inference (AAAI 2025 · 3 citations) |
| Add subgraph sampling while retaining the diversity regularizer | SPMotif-0.9, EC50-Scaffold, and EC50-Size · classification performance | Helped | Subgraph Aggregation for Out-of-Distribution Generalization on Graphs (AAAI 2025 · 3 citations) | |
| Use top 50% of training nodes ranked by h(i), rather than random selection | Common benchmarks (average) · AUROC | 81.19 -> 85.19 | Helped | Revisiting Score Propagation in Graph Out-of-Distribution Detection (NeurIPS 2024 · 3 citations) |
| Add global relationship modeling | Arxiv and Twitch OOD subsets · accuracy / ROC-AUC | Helped | Expand Horizon: Graph Out-of-Distribution Generalization via Multi-Level Environment Inference (AAAI 2025 · 2 citations) | |
| Remove all augmented representations (100% removal) | graph classification test accuracy · accuracy | Hurt | Graph Neural Network Generalization With Gaussian Mixture Model Based Augmentation (ICML 2025 · 1 citation) | |
| Align only local same-class pairs within 1 hop rather than all pairs | Toy concept shift, epoch 200 · accuracy | 0.253 -> 0.354; 0.250 -> 0.312 | Helped | Dissecting the Failure of Invariant Learning on Graphs (NeurIPS 2024 · 1 citation) |
| Add labeled ID connectivity to self-supervised graph learning | Toy graph, covariate OOD · linear probing error | 0 -> 0; 2 -> 2 | Mixed | Bridging OOD Detection and Generalization: A Graph-Theoretic View (NeurIPS 2024 · 1 citation) |
| Reconnect labeled nodes with added edges (TAR-N) instead of preserving the original graph | WebKB, CBAS, Twitch, Cora, and long-tailed datasets · accuracy or ROC-AUC | Mixed | Topology-Aware Dynamic Reweighting for Distribution Shifts on Graph (ICML 2025 · 0 citations) |
Baseline numbers that disagree
The same baseline on the same benchmark and metric, reported more than 5% apart by different papers here. Training budgets, data splits and reimplementations differ, so state which number you compare with.
| Baseline | Benchmark | Metric | Reported | Papers |
|---|---|---|---|---|
| GNNSafe | Amazon-Photo | FPR95 | 2.1 – 22.39 | 2 papers, e.g. ICLR 2025 and NeurIPS 2024 |
| GNNSafe | Coauthor-CS | FPR95 | 3.18 – 16.64 | 2 papers, e.g. ICLR 2025 and NeurIPS 2024 |
What reviewers will expect
Datasets and baselines named in the 19 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- Cora · 4 of 19 papers
- Twitch · 3 of 19 papers
- Citeseer · 2 of 19 papers
- Pubmed · 2 of 19 papers
- MUTAG · 2 of 19 papers
- Twitter · 2 of 19 papers
- GOOD-Motif · 2 of 19 papers
- GOOD-HIV · 2 of 19 papers
Baselines
- ERM · 7 of 19 papers
- IRM · 7 of 19 papers
- VREx · 6 of 19 papers
- DIR · 6 of 19 papers
- CIGA · 5 of 19 papers
- GSAT · 5 of 19 papers
- GroupDRO · 4 of 19 papers
- CAL · 4 of 19 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 52 papers; from OpenAlex and the papers' affiliations).
- Tsinghua University · 5 papers
- National University of Singapore · 4 papers
- Cornell University · 4 papers
- Peking University · 3 papers
- The Chinese University of Hong Kong · 3 papers
- Tencent · 2 papers
- Beijing University of Posts and Telecommunications · 2 papers
- Shanghai Jiao Tong University · 2 papers
- Texas A&M University · 2 papers
- Texas A&M University - College Station · 2 papers
Use it
Download BibTeX of 33 key papersOpen the topic in the interactive map
Get alerts. Describe your idea in scime atlas and save it (free scime account): you get a weekly email when new preprints land close to it.
Related approaches
Same keyword cluster
- Graph domain adaptation (82 papers)
- Graph anomaly detection (101 papers)
Elsewhere in the topic
- Semi-supervised node representation learning (214 papers)
- Heterophily and imbalance on graphs (178 papers)
- Diffusion-based graph generation (138 papers)
- LLMs and graph foundation models (117 papers)
- Molecular graph pretraining (101 papers)
- Brain and biological graph learning (78 papers)
- Multi-view graph clustering (70 papers)
- Graph contrastive learning analysis (61 papers)
Benchmark numbers, ablations and baseline numbers are extracted by AI from each paper's full text and can be wrong; the paper is the authority. Data: results index of 2026-11-02, citation counts from OpenAlex (2026-10-09).