Research map / Architectures and efficiency / CNN design
Hyperbolic Neural Networks++
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 |
|---|---|---|---|
| Use hyperbolic RNN+FFNN with Euclidean MLR instead of a fully Euclidean RNN | RNN sentence-pair classifier · PREFIX-10% | accuracy: 89.62 % -> 96.36 % | Helped |
| Use a fully hyperbolic RNN instead of a fully Euclidean RNN | RNN sentence-pair classifier · PREFIX-50% | accuracy: 72.10 % -> 62.94 % | Hurt |
| Use hyperbolic GRU+FFNN with Euclidean MLR instead of a fully Euclidean GRU | GRU sentence-pair classifier · PREFIX-30% | accuracy: 86.47 % -> 88.47 % | Helped |
| Use hyperbolic rather than Euclidean sentence encoders | SNLI sentence-pair classifier · SNLI | accuracy: 79.34 % -> 79.18 %; 81.52 % -> 81.19 % | No clear effect |
| Use hyperbolic MLR rather than Euclidean MLR on Poincaré embeddings | WordNet subtree classifier · WordNet subtrees, d=5 | F1: 98.07 ± 0.55% vs 95.59 ± 1.18%; 87.89 ± 0.80% vs 67.82 ± 0.81%; 55.46 ± 5.49% vs 35.23 ± 3.16%; 88.73 ± 3.22% vs 59.35 ± | Helped |
| Use hyperbolic MLR rather than Euclidean MLR after mapping embeddings to the tangent space | WordNet subtree classifier · WordNet subtrees, d=5 | F1: 98.07 ± 0.55% vs 89.21 ± 1.34%; 87.89 ± 0.80% vs 67.92 ± 0.74%; 55.46 ± 5.49% vs 26.47 ± 0.78%; 88.73 ± 3.22% vs 52.51 ± | Helped |
| Use identity rather than tanh in the hyperbolic RNN/GRU cell and identity rather than ReLU after the hyperbolic FFNN (ablation) | Hyperbolic sentence classifier · SNLI and PREFIX datasets | accuracy | Helped |
| Use tanh and ReLU rather than identity nonlinearities in the Euclidean models (ablation) | Fully Euclidean sentence classifier · SNLI and PREFIX datasets | accuracy | Helped |
Closest papers by meaning
- The Numerical Stability of Hyperbolic Representation Learning (ICML 2023 · similarity 0.90)
- Fully Hyperbolic Convolutional Neural Networks for Computer Vision (ICLR 2024 · similarity 0.89)
- Hyperbolic Embeddings of Supervised Models (NeurIPS 2024 · similarity 0.87)
- Residual Hyperbolic Graph Convolution Networks (AAAI 2024 · similarity 0.85)
- Fast Hyperboloid Decision Tree Algorithms (ICLR 2024 · similarity 0.85)
- Hyperbolic-PDE GNN: Spectral Graph Neural Networks in the Perspective of A System of Hyperbolic Partial Differential Equations (ICML 2025 · similarity 0.85)
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.