Research map / Graph learning / Graph self-supervised learning
Adversarial Graph Augmentation to Improve Graph Contrastive Learning
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 |
|---|---|---|---|
| learn input-graph-dependent adversarial edge-drop probabilities instead of using fixed or uniform random augmentation (ablation) | GCL with GIN encoder · unsupervised graph property prediction | performance across datasets: AD-GCL clearly improves upon NAD-GCL across all the datasets | Helped |
| tune the edge-drop regularization strength on validation data rather than fix it (ablation) | AD-GCL · unsupervised graph property prediction | performance | Mixed |
| vary regularization strength over a broad range (ablation) | AD-GCL · classification tasks | validation performance: 0.1 to 10.0; 80% edges to <10% edges | No clear effect |
| vary random edge-drop ratio (ablation) | NAD-GCL · molecule classification | validation performance | Mixed |
| use large edge-drop regularization strengths, leading to near-zero edge dropping (ablation) | AD-GCL · classification tasks | performance: λreg≥5.0 | Helped |
| use a low regularization strength, allowing more than half of edges to be dropped (ablation) | AD-GCL · classification tasks | performance: λreg=0.3; greater than 0.6 | No clear effect |
| use larger edge-drop ratios (ablation) | AD-GCL and NAD-GCL · regression tasks | validation performance | Helped |
| pre-train with AD-GCL rather than no pre-training | GIN for transfer learning · BBBP | ROC-AUC: 65.8 ± 4.5 -> 70.01 ±1.07 | Helped |
| pre-train with AD-GCL rather than GraphCL | GIN for transfer learning · ClinTox | ROC-AUC: 75.99 ± 2.65 -> 79.78 ± 3.52 | Helped |
| pre-train with AD-GCL rather than GraphCL | GIN for transfer learning · MUV | ROC-AUC: 69.8 ± 2.66 -> 72.30 ± 1.61 | Mixed |
| pre-train with AD-GCL rather than GraphCL | GIN for transfer learning · ToxCast | ROC-AUC: 62.40 ± 0.57 -> 63.07 ± 0.72 | Hurt |
| self-supervised pre-train with AD-GCL before fine-tuning | GCN for semi-supervised graph classification · TU datasets with 10% labels | accuracy: AD-GCL-FIX significantly outperforms baselines in 3 out of 6 datasets | Helped |
Closest papers by meaning
- Adversarial Contrastive Graph Augmentation with Counterfactual Regularization (AAAI 2025 · similarity 0.93)
- GNN-Transformer Cooperative Architecture for Trustworthy Graph Contrastive Learning (AAAI 2025 · similarity 0.90)
- Unified Graph Augmentations for Generalized Contrastive Learning on Graphs (NeurIPS 2024 · similarity 0.90)
- Does GCL Need a Large Number of Negative Samples? Enhancing Graph Contrastive Learning with Effective and Efficient Negative Sampling (AAAI 2025 · similarity 0.90)
- ReGCL: Rethinking Message Passing in Graph Contrastive Learning (AAAI 2024 · similarity 0.89)
- SOLA-GCL: Subgraph-Oriented Learnable Augmentation Method for Graph Contrastive Learning (AAAI 2025 · similarity 0.89)
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.