Research map / Graph learning / Graph self-supervised learning
Learning to Propagate for Graph Meta-Learning
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 |
|---|---|---|---|
| propagate class prototypes over the class-relation graph | Prototypical Net · tieredImageNet-Close and tieredImageNet-Far, random and snowball 5-way tasks | accuracy | Helped |
| include test-class hierarchy connections during inference (GPN+) | GPN · tieredImageNet-Close | accuracy: 1∼2% improvement on average | Helped |
| include test-class hierarchy connections during inference (GPN+) | GPN · tieredImageNet-Far | accuracy | No clear effect |
| use hybrid random and snowball task sampling (ablation) | GPN · tieredImageNet-Close 5-way-1-shot random | accuracy: 46.20$\pm$1.70\% -> 50.54$\pm$1.67\% | Helped |
| use random sampling only rather than hybrid task sampling (ablation) | GPN · tieredImageNet-Close 5-way-1-shot random | accuracy: 50.54$\pm$1.67\% -> 49.33$\pm$1.68\% | Hurt |
| use neighbor-to-class propagation rather than parent-to-class propagation (ablation) | GPN · tieredImageNet-Close 5-way-1-shot random | accuracy: 42.60$\pm$1.61\% -> 50.54$\pm$1.67\% | Helped |
| use neighbor-to-class propagation rather than child-to-class propagation (ablation) | GPN · tieredImageNet-Close 5-way-1-shot random | accuracy: 37.90$\pm$1.50\% -> 50.54$\pm$1.67\% | Helped |
| use neighbor-to-class propagation rather than forward-backward belief propagation (ablation) | GPN · tieredImageNet-Close 5-way-1-shot random | accuracy: 47.90$\pm$1.72\% -> 50.54$\pm$1.67\% | Helped |
| add auxiliary supervised tasks during training (ablation) | GPN · tieredImageNet-Close 5-way-1-shot random | accuracy: 41.87$\pm$1.72\% -> 50.54$\pm$1.67\% | Helped |
| add maximum spanning tree propagation pathways (ablation) | GPN · tieredImageNet-Close 5-way-1-shot random | accuracy: 45.83$\pm$1.64\% -> 50.54$\pm$1.67\% | Helped |
| use multi-head multiplicative attention instead of single-head additive attention (ablation) | GPN · tieredImageNet-Close 5-way-1-shot random | accuracy: 46.74$\pm$1.71\% -> 50.54$\pm$1.67\% | Helped |
Closest papers by meaning
- Graph Meta Learning via Local Subgraphs (NeurIPS 2020 · similarity 0.91)
- Unlocking the Potential of Black-box Pre-trained GNNs for Graph Few-shot Learning (AAAI 2025 · similarity 0.87)
- When Sparse Graph Representation Learning Falls into Domain Shift: Data Augmentation for Cross-Domain Graph Meta-Learning (Student Abstract) (AAAI 2024 · similarity 0.87)
- Beyond the Golden Teacher: Enhancing Graph Learning through LLM-GNN Co-teaching (NeurIPS 2026 · similarity 0.86)
- Unsupervised Episode Generation for Graph Meta-learning (ICML 2024 · similarity 0.86)
- Graph Metanetworks for Processing Diverse Neural Architectures (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.