Research map / Graph learning / Graph self-supervised learning
Automated Data Augmentations for Graph Classification
Youzhi Luo, Michael Curtis McThrow, Wing Yee Au, Tao Komikado, Kanji Uchino, Koji Maruhashi, Shuiwang Ji
AI-extracted. The measured effects below were read by AI from the paper's full text (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 |
|---|---|---|---|
| Replace learned per-element transformations with uniform transformations (remove learnable graph transformation) (ablation) | GraphAug · PROTEINS, IMDB-BINARY, NCI1 | accuracy: PROTEINS 0.722 vs 0.696; IMDB-BINARY 0.762 vs 0.724; NCI1 0.816 vs 0.760 | Improved |
| Random augmentation category selection instead of learned selection (ablation) | GraphAug · PROTEINS, IMDB-BINARY, NCI1 | accuracy: -0.020 PROTEINS, -0.016 IMDB-BINARY, -0.020 NCI1 | Hurt |
| Single augmentation step (T=1) instead of sequential multi-step augmentation (ablation) | GraphAug · PROTEINS, IMDB-BINARY, NCI1 | accuracy: -0.010 PROTEINS, -0.009 IMDB-BINARY, -0.007 NCI1 | Hurt |
| Uniform MaskNF augmentation (label-violating transformation) | GIN classifier · COLORS | testing accuracy: 0.507 vs 0.578 without augmentation | Hurt |
| Uniform PerturbEdge augmentation | GIN classifier · TRIANGLES | testing accuracy: 0.303 vs 0.506 without augmentation | Hurt |
| Learned label-invariant augmentation (GraphAug) instead of no augmentation | GIN classifier · COLORS and TRIANGLES | testing accuracy: +0.055 COLORS; +0.007 TRIANGLES | Improved |
| GraphAug augmentation vs no augmentation on graph benchmarks | GIN classifier · PROTEINS, IMDB-BINARY, COLLAB, MUTAG, NCI109, NCI1, ogbg-molhiv | accuracy / ROC-AUC: gains from +0.002 (ogbg-molhiv 0.756→0.774 is +0.018) to +0.026 (MUTAG); largest gain IMDB +0.031 | Improved |
| Uniform augmentations on molecular datasets (MaskNF, DropNode, PerturbEdge) | GIN classifier · MUTAG, NCI109, NCI1 | accuracy: e.g., NCI109 PerturbEdge 0.555 vs 0.794; NCI1 PerturbEdge 0.545 vs 0.804 | Hurt |
| Uniform DropNode augmentation on ogbg-molhiv | GIN classifier · ogbg-molhiv | ROC-AUC: 0.717 vs 0.756 | Hurt |
| Uniform MaskNF augmentation on COLLAB | GIN classifier · COLLAB | accuracy: +0.009 vs 0.806 | Improved |
Closest papers by meaning
- Label-invariant Augmentation for Semi-Supervised Graph Classification (NeurIPS 2022 · similarity 0.90)
- Through the Dual-Prism: A Spectral Perspective on Graph Data Augmentation for Graph Classifications (AAAI 2025 · similarity 0.89)
- Data Augmentation for Graph Neural Networks (AAAI 2021 · similarity 0.89)
- Learning the Latent Structure: A Feature-Centric Approach to Graph Data Augmentation (AAAI 2026 · similarity 0.88)
- Unleashing the Power of Graph Data Augmentation on Covariate Distribution Shift (NeurIPS 2023 · similarity 0.88)
- Community-Invariant Graph Contrastive Learning (ICML 2024 · similarity 0.86)
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.
