Research map / Graph learning / GNN architectures
No Change, No Gain: Empowering Graph Neural Networks with Expected Model Change Maximization for Active Learning
Zixing Song, Yifei Zhang, Irwin King
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 |
|---|---|---|---|
| DOCTOR acquisition vs random selection | SGC · Cora | accuracy: 78.9±0.9 -> 86.9±0.7 | Improved |
| DOCTOR vs strongest baseline IGP | SGC · Cora | accuracy: 86.3±0.7 -> 86.9±0.7 | Improved |
| DOCTOR vs strongest baseline IGP | SGC · Citeseer | accuracy: 75.8±0.4 -> 76.5±0.8 | Improved |
| DOCTOR vs strongest baseline IGP | SGC · PubMed | accuracy: 83.5±0.5 -> 84.3±0.9 | Improved |
| DOCTOR vs strongest baseline IGP | SGC · Reddit | accuracy: 93.3±0.4 -> 93.5±0.5 | Improved |
| DOCTOR vs strongest baseline GraphPart | SGC · ogbn-arxiv | accuracy: 72.3±2.1 -> 73.0±1.2 | Improved |
| Increase labeling budget from 2c to 20c | DOCTOR · Cora/Citeseer/PubMed | accuracy | Improved |
| Swap SGC backbone for GCN | DOCTOR · Citeseer | accuracy: 76.5±0.8 -> 76.2±0.7 | No clear effect |
| Swap SGC backbone for APPNP | DOCTOR · Citeseer | accuracy: 76.5±0.8 -> 76.6±0.9 | No clear effect |
| Swap SGC backbone for GraphSAGE | DOCTOR · Citeseer | accuracy: 76.5±0.8 -> 76.4±0.9 | No clear effect |
| DOCTOR vs GEEM running time per AL iteration | DOCTOR · Cora | relative running time | Improved |
| Approximate EMC (Eq. 11) instead of exact EMC via re-training | DOCTOR · Cora case study | EMC value | Improved |
Closest papers by meaning
- Better with Less: A Data-Active Perspective on Pre-Training Graph Neural Networks (NeurIPS 2023 · similarity 0.88)
- Structural Fairness-aware Active Learning for Graph Neural Networks (ICLR 2024 · similarity 0.88)
- Graph Policy Network for Transferable Active Learning on Graphs (NeurIPS 2020 · similarity 0.87)
- How Graph Neural Networks Learn: Lessons from Training Dynamics (ICML 2024 · similarity 0.87)
- Almost Surely Asymptotically Constant Graph Neural Networks (NeurIPS 2024 · similarity 0.87)
- Robust Offline Active Learning on Graphs (NeurIPS 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.
