Research map / Label-efficient and robust learning / Recognition and long-tail
von Mises-Fisher Loss: An Exploration of Embedding Geometries for Supervised Learning
Tyler R. Scott, Andrew C. Gallagher, Michael C. Mozer
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 cosine distance at test time (ablation) | Euclidean loss · Cars196 retrieval | mAP@R: 21.3 -> 23.2 | Improved |
| use cosine distance at test time (ablation) | Euclidean loss · SOP retrieval | mAP@R: 39.7 -> 42.1 | Improved |
| use cosine distance at test time (ablation) | Hyperbolic loss · SOP retrieval | mAP@R: 41.0 -> 44.0 | Improved |
| use cosine distance at test time (ablation) | Hyperbolic loss · Cars196 retrieval | mAP@R: 22.9 -> 25.0 | Improved |
| use vMF loss instead of Spherical loss | Spherical loss · CIFAR100 ECE | ECE (%): 21.8 -> 7.9 | Improved |
| use vMF loss instead of Euclidean loss | Euclidean loss · CIFAR10 ECE | ECE (%): 8.8 -> 5.9 | Improved |
| apply temperature scaling post hoc (ablation) | Euclidean loss · CIFAR100 ECE | ECE (%): 20.6 -> 2.2 | Improved |
| apply temperature scaling post hoc (ablation) | Spherical loss · CIFAR100 ECE | ECE (%): 21.8 -> 10.8 | Improved |
| apply temperature scaling post hoc (ablation) | vMF loss · CIFAR100 ECE | ECE (%): 7.9 -> 8.0 | No clear effect |
| use spherical geometry (Spherical) instead of Euclidean | Euclidean loss · CIFAR100 accuracy | accuracy (%): 69.21 -> 70.57 | Improved |
| use spherical geometry (Spherical) instead of Euclidean | Euclidean loss · CIFAR10 accuracy | accuracy (%): 94.13 -> 93.99 | No clear effect |
| use vMF loss instead of Spherical loss | Spherical loss · SOP retrieval | mAP@R: 44.3 -> 38.3 | Hurt |
Closest papers by meaning
- Analyzing and Improving Representations with the Soft Nearest Neighbor Loss (ICML 2019 · similarity 0.87)
- Geometric Losses for Distributional Learning (ICML 2019 · similarity 0.87)
- Hyperbolic Embeddings of Supervised Models (NeurIPS 2024 · similarity 0.87)
- Learning Representations on the Unit Sphere: Investigating Angular Gaussian and Von Mises-Fisher Distributions for Online Continual Learning (AAAI 2024 · similarity 0.86)
- Trading off Consistency and Dimensionality of Convex Surrogates for Multiclass Classification (NeurIPS 2024 · similarity 0.86)
- Generative Cross-Entropy: A Strictly Proper Loss for Data-Efficient Classification (NeurIPS 2026 · 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.
