Research map / Label-efficient and robust learning / Recognition and long-tail
[Re] A Cluster-based Approach for Improving Isotropy in Contextual Embedding Space
Benjamin Džubur
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 |
|---|---|---|---|
| apply global dominant-direction removal | GPT-2 · STS-B | Spearman correlation: 24.8 -> 55.7 | Improved |
| use cluster-specific instead of globally computed dominant directions | GPT-2 · STS-B | Spearman correlation: 55.7 -> 62.3 | Improved |
| apply global dominant-direction removal | BERT-base · STS-B | Spearman correlation: 47.4 -> 65.5 | Improved |
| use cluster-specific instead of globally computed dominant directions | BERT-base · STS-B | Spearman correlation: 65.5 -> 66.0 | Improved |
| apply global dominant-direction removal | RoBERTa-base · STS-B | Spearman correlation: 56.0 -> 70.1 | Improved |
| use cluster-specific instead of globally computed dominant directions | RoBERTa-base · STS-B | Spearman correlation: 70.1 -> 71.4 | Improved |
| apply cluster-based isotropy enhancement | BERT · RTE | accuracy: 54.4 -> 56.5 | Improved |
| apply cluster-based isotropy enhancement | BERT · CoLA | Matthew's correlation: 38.0 -> 40.7 | Improved |
| apply cluster-based isotropy enhancement | BERT · SST-2 | accuracy: 80.1 -> 82.5 | Improved |
| apply cluster-based isotropy enhancement | BERT · MRPC | accuracy: 70.2 -> 72.4 | Improved |
| apply cluster-based isotropy enhancement | BERT · WiC | accuracy: 60.0 -> 61.0 | Improved |
| apply cluster-based isotropy enhancement | BERT · BoolQ | accuracy: 64.7 -> 66.4 | Improved |
Closest papers by meaning
- Isotropy in the Contextual Embedding Space: Clusters and Manifolds (ICLR 2021 · similarity 0.90)
- Stable Anisotropic Regularization (ICLR 2024 · similarity 0.85)
- Contextures: Representations from Contexts (ICML 2025 · similarity 0.81)
- Metric Space Magnitude for Evaluating the Diversity of Latent Representations (NeurIPS 2024 · similarity 0.81)
- Isometric Representation Learning for Disentangled Latent Space of Diffusion Models (ICML 2024 · similarity 0.81)
- Contextual Document Embeddings (ICLR 2025 · similarity 0.81)
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.
