Research map / Graph learning / Knowledge and temporal graphs
Rethinking Knowledge Graph Propagation for Zero-Shot Learning
Michael Kampffmeyer, Yinbo Chen, Xiaodan Liang, Hao Wang, Yujia Zhang, Eric P. Xing
Check 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 dense ancestor/descendant propagation with distance-based weights instead of a shallow GCN | Zero-shot classifier prediction with ResNet-50 · ImageNet 2-hops | Hit@1 accuracy (%): 26.2 -> 26.6 | Improved |
| Use dense ancestor/descendant propagation with distance-based weights instead of a shallow GCN | Zero-shot classifier prediction with ResNet-50 · ImageNet 3-hops | Hit@1 accuracy (%): 6.0 -> 6.3 | Improved |
| Use dense ancestor/descendant propagation with distance-based weights instead of a shallow GCN | Zero-shot classifier prediction with ResNet-50 · ImageNet All | Hit@1 accuracy (%): 2.8 -> 3.0 | Improved |
| Use DGP instead of SGCN when seen classes are included among candidate labels | Zero-shot classifier prediction with ResNet-50 · ImageNet 2-hops+1K | Hit@1 accuracy (%): 11.9 -> 10.3 | Hurt |
| Use DGP instead of GCNZ | Zero-shot classification · AWA2 unseen classes | Top-1 accuracy (%): 70.7 -> 77.3 | Improved |
| Add distance weighting (ablation) | DGP · ImageNet 2-hops | Hit@1 accuracy (%): 25.4 -> 26.6 | Improved |
| Add two-stage finetuning (ablation) | DGP · ImageNet 2-hops | Hit@1 accuracy (%): 24.6 -> 26.6 | Improved |
| Add two-stage finetuning (ablation) | SGCN · ImageNet 2-hops | Hit@1 accuracy (%): 24.8 -> 26.2 | Improved |
| Increase GCN depth from one layer to two (ablation) | SGCN without finetuning · ImageNet 2-hops | Hit@1 accuracy (%): 24.8 -> 24.2 | Hurt |
| Increase GCN depth from one layer to three (ablation) | SGCN without finetuning · ImageNet 2-hops | Hit@1 accuracy (%): 24.8 -> 23.9 | Hurt |
| Separate ancestor and descendant propagation rather than use the full adjacency matrix twice (ablation) | DGP · ImageNet 2-hops | Hit@1 accuracy (%): 26.0 -> 26.6 | Improved |
| Replace standard graph propagation with graph attention (ablation) | Single-hidden-layer graph model · ImageNet 2-hops | Hit@1 accuracy (%): 24.8 -> 24.1 | Hurt |
Closest papers by meaning
- Distilling Knowledge From Graph Convolutional Networks (CVPR 2020 · similarity 0.89)
- Learning Graphs for Knowledge Transfer With Limited Labels (CVPR 2021 · similarity 0.89)
- Generative Adversarial Zero-Shot Relational Learning for Knowledge Graphs (AAAI 2020 · similarity 0.88)
- Convolutional 2D Knowledge Graph Embeddings (AAAI 2018 · similarity 0.88)
- LLMs as Zero-shot Graph Learners: Alignment of GNN Representations with LLM Token Embeddings (NeurIPS 2024 · similarity 0.86)
- Towards Graph Foundation Models: Training on Knowledge Graphs Enables Transferability to General Graphs (NeurIPS 2025 · 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.
