Research map / Graph learning / Graph self-supervised learning
Let Invariant Rationale Discovery Inspire Graph Contrastive Learning
Sihang Li, Xiang Wang, An Zhang, Yingxin Wu, Xiangnan He, Tat-Seng Chua
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 |
|---|---|---|---|
| Use rationale-aware views instead of random node-dropping views (ablation) | RGCL on MoleculeNet transfer tasks · MoleculeNet transfer, AVG. | ROC-AUC (%): 70.83 -> 73.16 | Improved |
| Include the independence loss term (ablation) | RGCL on MoleculeNet transfer tasks · MoleculeNet transfer, AVG. | ROC-AUC (%): 72.60 -> 73.16 | Improved |
| Use rationale-aware views instead of GraphCL's random augmentations | GraphCL on MoleculeNet transfer tasks · MoleculeNet transfer, AVG. | GAIN: 3.77 to 6.16 | Improved |
| Use rationale-aware views instead of random augmentation | GraphCL on MNIST-Superpixel · MNIST-Superpixel, 10% MNIST | accuracy (%): 93.11 -> 94.92 | Improved |
| Use rationale-aware views instead of random augmentation | GraphCL on MNIST-Superpixel · MNIST-Superpixel, 1% MNIST | accuracy (%): 83.41 -> 83.84 | Improved |
| Use RGCL rather than GraphCL | MoleculeNet transfer tasks · HIV | ROC-AUC (%): 78.47 -> 77.90 | Hurt |
| Remove independence loss term (ablation) | RGCL on MoleculeNet transfer tasks · MoleculeNet transfer, AVG. | ROC-AUC (%): 73.16 -> 72.60 | Hurt |
| Replace rationale-aware augmentation with random node discarding (ablation) | RGCL on MoleculeNet transfer tasks · MoleculeNet transfer, AVG. | ROC-AUC (%): 73.16 -> 70.83 | Hurt |
| Increase the contrastive temperature above 0.2 (ablation) | RGCL · MoleculeNet and TU datasets | average performance | Hurt |
| Set contrastive temperature too small (e.g., 0.05) (ablation) | RGCL · MoleculeNet and TU datasets | average performance | Hurt |
| Overemphasize the independence regularizer with a large lambda (ablation) | RGCL · MoleculeNet and TU datasets | average performance | Hurt |
| Use a proper independence regularizer (lambda=0.1) (ablation) | RGCL · MoleculeNet and TU datasets | average performance | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- InfoGraph — InfoGraph: Unsupervised and Semi-supervised Graph-Level Representation Learning via Mutual Information Maximization (ICLR 2020) · better on 1 of 3 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Rethinking Dimensional Rationale in Graph Contrastive Learning from Causal Perspective (AAAI 2024 · similarity 0.92)
- Graph Contrastive Invariant Learning from the Causal Perspective (AAAI 2024 · similarity 0.89)
- Does GCL Need a Large Number of Negative Samples? Enhancing Graph Contrastive Learning with Effective and Efficient Negative Sampling (AAAI 2025 · similarity 0.89)
- S3GCL: Spectral, Swift, Spatial Graph Contrastive Learning (ICML 2024 · similarity 0.89)
- Efficient Contrastive Learning for Fast and Accurate Inference on Graphs (ICML 2024 · similarity 0.88)
- Perfect Alignment May be Poisonous to Graph Contrastive Learning (ICML 2024 · similarity 0.88)
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.
