Research map / Graph learning / GNN architectures
Graph Self-supervised Learning with Accurate Discrepancy Learning
Dongki Kim, Jinheon Baek, Sung Ju Hwang
Read the paperCheck your idea against 137,217 papers
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 |
|---|---|---|---|
| add graph discrimination loss L_GD (ablation) | D-SLA · ClinTox | ROC-AUC: 58.00 -> 70.83 | Improved |
| add L_edit alone (without L_GD) (ablation) | D-SLA · ClinTox | ROC-AUC: 58.00 -> 57.46 | No clear effect |
| add L_edit on top of L_GD (ablation) | D-SLA · ClinTox | ROC-AUC: 70.83 -> 77.48 | Improved |
| add L_edit on top of L_GD (ablation) | D-SLA · BACE | ROC-AUC: 81.58 -> 83.53 | Improved |
| add triplet margin loss L_margin (ablation) | D-SLA · ClinTox | ROC-AUC: 77.48 -> 80.17 | Improved |
| add triplet margin loss L_margin (ablation) | D-SLA · BACE | ROC-AUC: 83.53 -> 83.81 | Improved |
| add L_edit on top of L_GD (ablation) | D-SLA · COLLAB link prediction | accuracy: 74.23 -> 76.19 | Improved |
| increase margin alpha from 5.0 to 10.0 (ablation) | D-SLA · BACE | ROC-AUC: 83.81 -> 78.34 | Hurt |
| increase lambda2 from 0.5 to 0.9 (ablation) | D-SLA · BACE | ROC-AUC: 83.81 -> 80.72 | Hurt |
| reduce perturbation magnitude from 10% to 0.1% (ablation) | D-SLA · COLLAB link prediction | accuracy: 70.67 -> 76.19 | Improved |
| add node attribute masking (ablation) | D-SLA · COLLAB link prediction | accuracy: 70.42 -> 76.19 | Improved |
| D-SLA vs best contrastive/predictive baselines | GIN pretraining · Molecule/PPI average | ROC-AUC avg: 72.65 -> 73.85 | Improved |
Closest papers by meaning
- Disentangled Contrastive Learning on Graphs (NeurIPS 2021 · similarity 0.89)
- From Canonical Correlation Analysis to Self-supervised Graph Neural Networks (NeurIPS 2021 · similarity 0.89)
- Uncovering the Redundancy in Graph Self-supervised Learning Models (NeurIPS 2024 · similarity 0.88)
- Decoupled Self-supervised Learning for Graphs (NeurIPS 2022 · similarity 0.88)
- Equivalence is All: A Unified View for Self-supervised Graph Learning (ICML 2025 · similarity 0.87)
- Disentangled Graph Self-supervised Learning for Out-of-Distribution Generalization (ICML 2024 · similarity 0.87)
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.
