Research map / Graph learning / Graph self-supervised learning
Graph Policy Network for Transferable Active Learning on Graphs
Shengding Hu, Zheng Xiong, Meng Qu, Xingdi Yuan, Marc-Alexandre Côté, Zhiyuan Liu, Jian Tang
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 |
|---|---|---|---|
| Use GPA instead of AGE for zero-shot active learning | GNN node classification on Reddit graphs · Reddit3, Reddit4, and Reddit5; same-domain transfer | Micro-F1: 91.09 -> 92.85; 87.55 -> 91.57; 88.02 -> 91.60 | Improved |
| Use GPA instead of AGE for zero-shot active learning | GNN node classification · Pubmed, Reddit1–5, Physics, and CS; train on Cora + Citeseer | Micro-F1: 74.78 -> 77.80; 83.76 -> 88.10; 92.56 -> 95.19; 90.61 -> 92.07; 86.94 -> 91.39; 87.73 -> 90.66; 84.68 -> 88.08; 86.33 -> | Improved |
| Use GPA rather than single-criterion uncertainty, centrality, or coreset selection | Active learning for GNN node classification · Reddit3–5 | Micro-F1: GPA 92.85, 91.57, 91.60; Random 88.21, 84.81, 86.20 | Mixed |
| Add graph-structure modeling with a GCN policy rather than an MLP policy (ablation) | GPA · Pubmed and Reddit1–5, Physics | Micro-F1: Pubmed: 70.49 -> 77.80; Reddit1: 72.65 -> 88.10; Reddit2: 85.61 -> 95.19; Reddit3: 81.66 -> 92.07; Reddit4: 80.03 -> 91. | Improved |
| Add graph-structure modeling with a GCN policy rather than an MLP policy (ablation) | GPA · CS | Micro-F1: 88.53 -> 87.90 | Hurt |
| Add entropy to the node state (ablation) | GPA · Reddit3–5 | Micro-F1: 92.48 -> 92.85; 90.12 -> 91.57; 90.35 -> 91.60 | Improved |
| Add degree to the node state (ablation) | GPA · Reddit3–5 | Micro-F1: 92.36 -> 92.85; 91.48 -> 91.57; 90.56 -> 91.60 | Mixed |
| Add KL divergence to the node state (ablation) | GPA · Reddit3–5 | Micro-F1: 92.76 -> 92.85; 91.12 -> 91.57; 91.29 -> 91.60 | Improved |
| Add the labeled-node indicator to the node state (ablation) | GPA · Reddit3–5 | Micro-F1: 91.11 -> 92.85; 88.90 -> 91.57; 90.72 -> 91.60 | Improved |
| Train the policy on multiple source graphs instead of one (ablation) | GPA · Reddit graphs, evaluation on remaining graphs | Micro-F1 | Improved |
| Use a smaller query budget with GPA to reach the same Micro-F1 | GNN node classification on Reddit4 · Reddit4 | Micro-F1: 100 budget -> 90.0; 30 budget -> 90.0 | Improved |
| Use a smaller query budget with GPA to match AGE performance | GNN node classification on Reddit4 · Reddit4 | Micro-F1: AGE: 100 budget -> 91.7; GPA: 50 budget -> 91.7 | Improved |
Closest papers by meaning
- Batch Active Learning with Graph Neural Networks via Multi-Agent Deep Reinforcement Learning (AAAI 2022 · similarity 0.89)
- Information Gain Propagation: a New Way to Graph Active Learning with Soft Labels (ICLR 2022 · similarity 0.89)
- Efficient Subgraph GNNs by Learning Effective Selection Policies (ICLR 2024 · similarity 0.89)
- Robust Offline Active Learning on Graphs (NeurIPS 2024 · similarity 0.86)
- A Transfer Approach Using Graph Neural Networks in Deep Reinforcement Learning (AAAI 2024 · similarity 0.86)
- Generating In-Distribution Proxy Graphs for Explaining Graph Neural Networks (ICML 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.
