Research map / Graph learning / Graph self-supervised learning
AutoGCL: Automated Graph Contrastive Learning via Learnable View Generators
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 learnable, input-conditioned node drop/keep/mask view generators instead of fixed augmentations | Graph contrastive learning · Unsupervised graph classification | accuracy | Mixed |
| Use AutoGCL pretraining instead of AD-GCL | GIN transfer learning · BACE | ROC-AUC: 78.51±0.80 -> 83.26±1.13 | Helped |
| Use AutoGCL pretraining instead of the strongest listed baseline | GIN transfer learning · Tox21 | ROC-AUC: 76.7±0.4 -> 75.69±0.29 | Hurt |
| Use the joint strategy instead of naive contrastive pretraining followed by fine-tuning (ablation) | AutoGCL semi-supervised learning · COLLAB | accuracy: 76.60±2.15 -> 77.16±1.48 | Helped |
| Add the similarity loss to joint training (ablation) | AutoGCL semi-supervised learning · PROTEINS | accuracy: 74.66±2.58 -> 75.65±2.40 | Helped |
| Add the contrastive loss to joint cls+sim training (ablation) | AutoGCL semi-supervised learning · COLLAB | accuracy: 75.60±2.08 -> 77.16±1.48 | Helped |
| Add the similarity loss to joint cl+cls training (ablation) | AutoGCL semi-supervised learning · NCI1 | accuracy: 73.07±2.31 -> 73.75±2.25 | Helped |
| Use edge perturbation as the only GraphCL augmentation instead of random4 (ablation) | GraphCL · NCI1 | accuracy: 72.94 ± 3.23 -> 71.85 ± 3.45 | Hurt |
| Use subgraph augmentation as the only GraphCL augmentation instead of random4 (ablation) | GraphCL · PROTEINS | accuracy: 71.08 ± 5.03 -> 62.99 ± 4.92 | Hurt |
| Vary the node-dropping augmentation ratio from 0.1 to 0.4 (ablation) | GraphCL · NCI1 | accuracy: 75.01 ± 2.91 -> 73.94 ± 2.32 | Hurt |
| Use the joint strategy rather than naive pretraining/fine-tuning (ablation) | AutoGCL semi-supervised learning · COLLAB | test accuracy | Helped |
Closest papers by meaning
- Simple and Asymmetric Graph Contrastive Learning without Augmentations (NeurIPS 2023 · similarity 0.92)
- An Empirical Study of Graph Contrastive Learning (NeurIPS 2021 · similarity 0.92)
- Neighbor Contrastive Learning on Learnable Graph Augmentation (AAAI 2023 · similarity 0.91)
- S3GCL: Spectral, Swift, Spatial Graph Contrastive Learning (ICML 2024 · similarity 0.91)
- InfoGCL: Information-Aware Graph Contrastive Learning (NeurIPS 2021 · similarity 0.91)
- Efficient Contrastive Learning for Fast and Accurate Inference on Graphs (ICML 2024 · 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.