Research map / Graph learning / GNN architectures
Empowering Graph Representation Learning with Test-Time Graph Transformation
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 |
|---|---|---|---|
| Apply test-time graph transformation instead of ERM training alone | GCN · Cora OOD | accuracy: +3.1% | Helped |
| Apply test-time graph transformation instead of ERM training alone | GCN · Elliptic OOD | F1 score: +5.0% | Helped |
| Apply test-time graph transformation instead of ERM training alone | GCN · OGB-Arxiv OOD | accuracy: +2.0% | Helped |
| Transform the test graph rather than adapt model parameters | GCN · Abnormal features on Cora, Citeseer, Pubmed, and arXiv | accuracy | Mixed |
| Add feature adaptation (ablation) | GTrans · Abnormal-feature robustness | accuracy | Helped |
| Add structure adaptation (ablation) | GTrans · Structural attack robustness | accuracy | Helped |
| Combine feature and structure adaptation (ablation) | GTrans · Graph corruption settings | accuracy | Helped |
| Use only the positive-view similarity term and omit negative samples (ablation) | GTrans contrastive surrogate loss · Test-time graph transformation | representation collapse | Hurt |
| Use GTrans rather than GCNJaccard preprocessing | GCN · arXiv structural attack at all listed perturbation rates | accuracy: 66.29±0.25 vs 65.01±0.26; 65.16±0.52 vs 63.25±0.30; 64.40±0.38 vs 61.83±0.29; 63.44±0.50 vs 60.57±0.34; 62.95±0.67 vs 59 | Helped |
| Optimize test graph with the contrastive surrogate loss | GTrans · OOD distribution shifts | accuracy | Helped |
Closest papers by meaning
- Can Classic GNNs Be Strong Baselines for Graph-level Tasks? Simple Architectures Meet Excellence (ICML 2025 · similarity 0.89)
- Classic GNNs are Strong Baselines: Reassessing GNNs for Node Classification (NeurIPS 2024 · similarity 0.88)
- Towards Robust Fidelity for Evaluating Explainability of Graph Neural Networks (ICLR 2024 · similarity 0.87)
- TRACI: A Data-centric Approach for Multi-Domain Generalization on Graphs (AAAI 2025 · similarity 0.87)
- G-Adapter: Towards Structure-Aware Parameter-Efficient Transfer Learning for Graph Transformer Networks (AAAI 2024 · similarity 0.87)
- Online GNN Evaluation Under Test-time Graph Distribution Shifts (ICLR 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.