Research map / Label-efficient and robust learning / Recognition and long-tail
Probing BERT in Hyperbolic Spaces
Boli Chen, Yao Fu, Guangwei Xu, Pengjun Xie, Chuanqi Tan, Mosha Chen, Liping Jing
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 a Poincaré rather than Euclidean probe | BERTbase7 dependency-distance probing · Penn Treebank | UUAS: 79.8 -> 83.7 | Improved |
| Use a Poincaré rather than Euclidean probe | BERTbase7 dependency-distance probing · Penn Treebank | DSpr.: 0.85 -> 0.88 | Improved |
| Use a Poincaré rather than Euclidean probe | BERTbase7 dependency-depth probing · Penn Treebank | Root %: 88.0 -> 91.3 | Improved |
| Use a Poincaré rather than Euclidean probe | BERTlarge16 dependency-distance probing · Penn Treebank | UUAS: 81.7 -> 85.9 | Improved |
| Use a Poincaré rather than Euclidean probe | BERTlarge16 dependency-distance probing · Penn Treebank | DSpr.: 0.87 -> 0.90 | Improved |
| Use a Poincaré rather than Euclidean probe | BERTbase9 sentiment probing · Movie Review | accuracy: 81.7 -> 84.9 | Improved |
| Fix rather than train the class meta embeddings (ablation) | Poincaré sentiment probe · Movie Review, BERTbase10 | accuracy: 84.2 -> 84.5 | Improved |
| Fix rather than train the class meta embeddings (ablation) | Poincaré sentiment probe · Movie Review, BERTbase9 | accuracy: 84.9 -> 84.2 | Hurt |
| Fix rather than train the class meta embeddings (ablation) | Euclidean sentiment probe · Movie Review, BERTbase9 | accuracy: 81.7 -> 78.4 | Hurt |
| Fix rather than train the class meta embeddings (ablation) | Euclidean sentiment probe · Movie Review, BERTbase10 | accuracy: 83.5 -> 79.1 | Hurt |
| Use hyperbolic rather than Euclidean geometry | sentiment probing of Linear embeddings · Movie Review | accuracy: 48.4 -> 48.4 | No clear effect |
| Use curvature -0.1 instead of -1 (ablation) | 2d Poincaré syntax probe · Penn Treebank, BERTbase7 | UUAS: 40.5 -> 38.7 | Hurt |
Closest papers by meaning
- Visualizing and Measuring the Geometry of BERT (NeurIPS 2019 · similarity 0.87)
- Isotropy in the Contextual Embedding Space: Clusters and Manifolds (ICLR 2021 · similarity 0.87)
- Learning Continuous Hierarchies in the Lorentz Model of Hyperbolic Geometry (ICML 2018 · similarity 0.87)
- A Polar coordinate system represents syntax in large language models (NeurIPS 2024 · similarity 0.85)
- Token Embeddings Violate the Manifold Hypothesis (NeurIPS 2025 · similarity 0.84)
- Invariant Features in Language Models: Geometric Characterization and Model Attribution (NeurIPS 2026 · similarity 0.84)
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.
