Research map / Graph learning / Graph self-supervised learning
Augmentation-free graph learning
Replace or learn graph augmentations, including spectral approaches.
25 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2021–2026) · an approach in Graph self-supervised learning, keyword cluster “graph contrastive · gcl · augmentation”
State History Benchmarks Tried Baselines Checklist Preprints Who Export
Where it stands
Still moving The best reported result is still improving on 3 of 4 tracked benchmarks.
9 papers in 2023–24 and 11 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic shrank, from 2.6% to 1.9%.
Open questions the data points to
- NCI1 (accuracy): best went from 78.7 (up to 2024) to 80.04 in 2025 — Equivalence is All: A Unified View for Self-supervised Graph Learning (ICML 2025)
- PubMed (accuracy): best went from 86.29 (up to 2024) to 87.34 in 2025 — Learn Beneficial Noise as Graph Augmentation (ICML 2025 · 1 citation)
- WikiCS (accuracy): best went from 79.98 (up to 2024) to 82.15 in 2025 — Equivalence is All: A Unified View for Self-supervised Graph Learning (ICML 2025)
What looks settled
- Cora (accuracy): best up to 2024 85.89, best in 2025–26 86.25 from 3 papers, less than 1% better.
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 · Multi-view Contrastive Graph Clustering (NeurIPS 2021 · 2 citations)
- 2022 · Augmentation-Free Self-Supervised Learning on Graphs (AAAI 2022 · 189 citations)
- 2023 · Spectral Feature Augmentation for Graph Contrastive Learning and Beyond (AAAI 2023 · 84 citations)
- 2024 · GAMC: An Unsupervised Method for Fake News Detection Using Graph Autoencoder with Masking (AAAI 2024 · 55 citations)
- 2025 · Through the Dual-Prism: A Spectral Perspective on Graph Data Augmentation for Graph Classifications (AAAI 2025 · 1 citation)
- 2026 · From Semantics to Spectrum: A New Lens on Graph Augmentation Strategy (AAAI 2026 · 1 citation)
Earliest papers here
- Multi-view Contrastive Graph Clustering (NeurIPS 2021 · 2 citations)
- Augmentation-Free Self-Supervised Learning on Graphs (AAAI 2022 · 189 citations)
- Regularizing Graph Neural Networks via Consistency-Diversity Graph Augmentations (AAAI 2022 · 38 citations)
Benchmark records
The best number each year that beat every earlier one, among papers of this approach reporting their own method. Numbers are extracted by AI from each paper's text; check the paper before citing.
Cora · accuracy (higher is better)
5 papers, 2023–2025 · small gains only
| Year | Best so far | Set by |
|---|---|---|
| 2023 | 85.89 | Spectral Feature Augmentation for Graph Contrastive Learning and Beyond (AAAI 2023 · 84 citations) |
| 2025 | 86.25 | Learn Beneficial Noise as Graph Augmentation (ICML 2025 · 1 citation) |
NCI1 · accuracy (higher is better)
4 papers, 2023–2025 · still improving
| Year | Best so far | Set by |
|---|---|---|
| 2023 | 78.7 | Spectral Feature Augmentation for Graph Contrastive Learning and Beyond (AAAI 2023 · 84 citations) |
| 2025 | 80.04 | Equivalence is All: A Unified View for Self-supervised Graph Learning (ICML 2025) |
PubMed · accuracy (higher is better)
4 papers, 2023–2025 · still improving
| Year | Best so far | Set by |
|---|---|---|
| 2023 | 86.29 | Spectral Feature Augmentation for Graph Contrastive Learning and Beyond (AAAI 2023 · 84 citations) |
| 2025 | 87.34 | Learn Beneficial Noise as Graph Augmentation (ICML 2025 · 1 citation) |
WikiCS · accuracy (higher is better)
4 papers, 2022–2025 · still improving
| Year | Best so far | Set by |
|---|---|---|
| 2022 | 77.62 | Augmentation-Free Self-Supervised Learning on Graphs (AAAI 2022 · 189 citations) |
| 2023 | 79.98 | Spectral Feature Augmentation for Graph Contrastive Learning and Beyond (AAAI 2023 · 84 citations) |
| 2025 | 82.15 | Equivalence is All: A Unified View for Self-supervised Graph Learning (ICML 2025) |
What has been tried
19 papers here report 201 measured changes (a component added, removed or swapped, compared with a baseline): helped 66%, no clear effect 4%, hurt 25%, mixed 4%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| Add node feature masking and edge dropping | PolitiFact · accuracy | 0.793 -> 0.828 | Helped | GAMC: An Unsupervised Method for Fake News Detection Using Graph Autoencoder with Masking (AAAI 2024 · 55 citations) |
| add node dropping and edge dropping instead of supervised GCN alone | Weibo · accuracy | 0.927 -> 0.962 | Helped | Propagation Tree Is Not Deep: Adaptive Graph Contrastive Learning Approach for Rumor Detection (AAAI 2024 · 42 citations) |
| Perturb high-frequency rather than low-frequency eigenvalues | IMDB-BINARY · accuracy | Helped | Through the Dual-Prism: A Spectral Perspective on Graph Data Augmentation for Graph Classifications (AAAI 2025 · 1 citation) | |
| Add PiNGDA augmentation to GRACE | Cora, CiteSeer, PubMed, Amazon-Photo, Coauthor-Phy · accuracy | 83.43 ± 0.32 -> 84.28 ± 0.24; 70.93 ± 0.21 -> 71.47 ± 0.20; 85.90 ± 0.24 -> 86.79 ± 0.27; 93.13 | Helped | Learn Beneficial Noise as Graph Augmentation (ICML 2025 · 1 citation) |
| Add dynamic confidence-factor selection instead of static high-confidence sample selection | Six benchmark datasets · clustering performance | Helped | Hybrid-Collaborative Augmentation and Contrastive Sample Adaptive-Differential Awareness for Robust Attributed Graph Clustering (NeurIPS 2025 · 0 citations) | |
| Use cosine similarity instead of negative Euclidean distance for query selection | Cora-Full, 5-way 1-shot · accuracy | 45.60% -> 64.20% | Helped | Unsupervised Episode Generation for Graph Meta-learning (ICML 2024 · 0 citations) |
| Add Lcau to ERM | ABIDE I · accuracy | 71.13±4.01 -> 72.17±4.10 | Helped | Causal Invariance-aware Augmentation for Brain Graph Contrastive Learning (ICML 2025) |
| Replace SWIRL's GCN encoder with GPR | D1 · HRScore | 96.24 -> 87.72 | Hurt | Towards Understanding Parametric Generalized Category Discovery on Graphs (ICML 2025) |
| remove graph augmentation | Cora, CiteSeer, PubMed, Amazon-Computers, Coauthor-CS · accuracy | 82.12 ± 0.60; 72.79 ± 0.21; 80.86 ± 0.72; 88.42 ± 0.44; 92.22 ± 0.29 (Aug+Pos: 83.26 ± 0.56; 73.20 ± 0.37; 81.33 ± 0.58; | No clear effect | Equivalence is All: A Unified View for Self-supervised Graph Learning (ICML 2025) |
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 |
|---|---|---|---|---|
| AD-GCL | RDT-B | accuracy | 85.52 – 90.04 | 2 papers, e.g. ICML 2025 and ICML 2025 |
What reviewers will expect
Datasets and baselines named in the 6 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 6 papers
- CiteSeer · 4 of 6 papers
- PROTEINS · 3 of 6 papers
- NCI1 · 3 of 6 papers
- DD · 3 of 6 papers
- MUTAG · 3 of 6 papers
- PubMed · 2 of 6 papers
- Wiki-CS · 2 of 6 papers
Baselines
- GCN · 3 of 6 papers
- GraphCL · 2 of 6 papers
- InfoGraph · 2 of 6 papers
- AD-GCL · 2 of 6 papers
- DGI · 2 of 6 papers
- BGRL · 2 of 6 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 20 papers; from OpenAlex and the papers' affiliations).
- Beijing University of Technology · 3 papers
- The Chinese University of Hong Kong · 2 papers
- Northwestern Polytechnical University · 2 papers
Use it
Download BibTeX of 24 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 contrastive learning analysis (61 papers)
- Contrastive learning with negatives and views (51 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)
- Graph anomaly detection (101 papers)
- Molecular graph pretraining (101 papers)
- Graph domain adaptation (82 papers)
- Brain and biological graph learning (78 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).