Research map / Graph learning / Graph self-supervised learning
A Hyperbolic-to-Hyperbolic Graph Convolutional Network
Check 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 |
|---|---|---|---|
| Perform graph convolution directly on the hyperbolic manifold instead of using tangent-space graph operations | H2H-GCN versus HGCN · Disease link prediction | ROC AUC: 90.8 ± 0.3 -> 97.8 ± 0.0 | Helped |
| Perform graph convolution directly on the hyperbolic manifold instead of using tangent-space graph operations | H2H-GCN versus HGCN · Disease node classification | F1 score: 74.5 ± 0.9 -> 90.3 ± 0.8 | Helped |
| Perform graph convolution directly on the hyperbolic manifold instead of using tangent-space graph operations | H2H-GCN versus HGCN · Airport, PubMed, and Cora link prediction and node classification | ROC AUC or classification score: 96.4 ± 0.1 -> 96.7 ± 0.0; 96.3 ± 0.0 -> 97.1 ± 0.0; 90.6 ± 0.2 -> 91.0 ± 0.3; 80.3 ± 0.3 -> 82.3 ± 0.4 | Helped |
| Perform graph convolution directly on the hyperbolic manifold instead of using tangent-space graph operations | H2H-GCN versus GAT · Cora link prediction and node classification | ROC AUC or accuracy: 93.7 ± 0.1 -> 95.4 ± 0.0; 83.0 ± 0.7 -> 83.6 ± 0.8 | Helped |
| Use direct hyperbolic graph convolution rather than tangent-space convolution | H2H-GCN versus HGNN · Synthetic graph classification at dimensionality 256 | macro F1: 95.3 ± 0.28 -> 97.2 ± 0.03 | Helped |
| Use direct hyperbolic graph convolution rather than tangent-space convolution | H2H-GCN versus HGNN · D&D, Proteins, and Enzymes graph classification | accuracy: 75.8 ± 3.3 -> 78.2 ± 3.3; 73.7 ± 2.3 -> 74.4 ± 3.0; 51.3 ± 6.1 -> 61.3 ± 4.9 | Helped |
| Use direct hyperbolic graph convolution rather than tangent-space convolution | H2H-GCN versus HGCN · Cora node classification across tested embedding dimensionalities | accuracy: 70.8 ± 1.6 -> 73.1 ± 2.5; 75.4 ± 1.7 -> 77.8 ± 0.6; 78.1 ± 0.8 -> 79.9 ± 0.9; 79.7 ± 1.4 -> 81.2 ± 0.9; 81.7 ± 0.7 -> 83 | Helped |
| Increase embedding dimensionality | H2H-GCN · Cora node classification | accuracy: 73.1 ± 2.5; 77.8 ± 0.6; 79.9 ± 0.9; 81.2 ± 0.9; 83.6 ± 0.8 | Helped |
| Use hyperbolic rather than Euclidean embeddings | HGNN versus Euclidean embeddings · Synthetic graph classification at low embedding dimensionalities | macro F1: 77.2 ± 0.12 -> 94.1 ± 0.03; 90.0 ± 0.21 -> 95.6 ± 0.14; 90.6 ± 0.17 -> 96.4 ± 0.23; 94.8 ± 0.25 -> 96.6 ± 0.22 | Helped |
| Use hyperbolic rather than Euclidean embeddings | HGCN versus Euclidean GCNs · Link prediction and node classification across the reported graph datasets | ROC AUC or classification score: The paper reports average gains of 22.7% (LP) and 19.9% (NC) over state-of-the-art Euclidean GCNs. | Helped |
Closest papers by meaning
- Hyperbolic Graph Convolutional Neural Networks (NeurIPS 2019 · similarity 0.94)
- Residual Hyperbolic Graph Convolution Networks (AAAI 2024 · similarity 0.93)
- Hyperbolic Graph Neural Networks at Scale: A Meta Learning Approach (NeurIPS 2023 · similarity 0.91)
- Random Laplacian Features for Learning with Hyperbolic Space (ICLR 2023 · similarity 0.89)
- Fully Hyperbolic Convolutional Neural Networks for Computer Vision (ICLR 2024 · similarity 0.89)
- Hgformer: Hyperbolic Graph Transformer for Collaborative Filtering (ICML 2025 · similarity 0.88)
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.