Research map / Label-efficient and robust learning / Recognition and long-tail
Hyperspherical Prototype 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 |
|---|---|---|---|
| Replace one-hot or word2vec class prototypes with separated hyperspherical prototypes | ResNet-32 classifier · CIFAR-100, 100 dimensions | accuracy (%): 62.1 { $\pm$ 0.1} -> 65.0 { $\pm$ 0.3}; 57.6 { $\pm$ 0.6} -> 65.0 { $\pm$ 0.3} | Helped |
| Use separated hyperspherical prototypes instead of word2vec prototypes | ResNet-32 classifier · CIFAR-100, 10 dimensions | accuracy (%): 29.0 { $\pm$ 0.0} -> 51.1 { $\pm$ 0.7} | Helped |
| Use separated hyperspherical prototypes instead of word2vec prototypes | ResNet-32 classifier · ImageNet-200, 200 dimensions | accuracy (%): 30.0 { $\pm$ 0.4} -> 44.7 { $\pm$ 0.3} | Helped |
| Reduce output dimensionality to 25 from the class-count-sized output | Hyperspherical prototype classifier · CIFAR-100, ResNet-32 | accuracy (%): 65.0 { $\pm$ 0.3} -> 63.0 { $\pm$ 0.1} | No clear effect |
| Add privileged information to prototype construction (ablation) | Hyperspherical prototype classifier · CIFAR-100, ResNet-32, 5 dimensions | accuracy (%): 28.7 { $\pm$ 0.4} -> 37.0 { $\pm$ 0.8} | Helped |
| Add privileged information to prototype construction (ablation) | Hyperspherical prototype classifier · CIFAR-100, ResNet-32, 3 dimensions | accuracy (%): 5.5 { $\pm$ 0.3} -> 11.5 { $\pm$ 0.4} | Helped |
| Add privileged information to prototype construction (ablation) | Hyperspherical prototype classifier · CIFAR-100, ResNet-32, 10 dimensions | accuracy (%): 51.1 { $\pm$ 0.7} -> 57.0 { $\pm$ 0.6} | Helped |
| Add privileged information to prototype construction (ablation) | Hyperspherical prototype classifier · CIFAR-100, ResNet-32, 25 dimensions | accuracy (%): 63.0 { $\pm$ 0.1} -> 64.0 { $\pm$ 0.2} | Helped |
| Add privileged information to prototype construction (ablation) | Hyperspherical prototype classifier · CIFAR-100, ResNet-32, 100 dimensions | accuracy (%): 65.0 { $\pm$ 0.3} -> 63.8 { $\pm$ 0.1} | Hurt |
| Add privileged information to prototype construction (ablation) | Hyperspherical prototype classifier · CIFAR-100, ResNet-32, 200 dimensions | accuracy (%): 63.7 { $\pm$ 0.4} -> 64.7 { $\pm$ 0.1} | Helped |
| Use hyperspherical classification rather than softmax cross-entropy | ResNet-32 classifier · CIFAR-100, 500 examples/class | accuracy (%): 64.4 { $\pm$ 0.4} -> 65.0 { $\pm$ 0.3} | No clear effect |
| Use hyperspherical classification rather than softmax cross-entropy | ResNet-32 classifier · CIFAR-100, 2 to 200 examples/class | accuracy (%): 44.2 { $\pm$ 0.0} -> 46.4 { $\pm$ 0.0} | Helped |
Closest papers by meaning
- Hyperspherical Classification with Dynamic Label-to-Prototype Assignment (CVPR 2024 · similarity 0.88)
- Hyperbolic Busemann Learning with Ideal Prototypes (NeurIPS 2021 · similarity 0.87)
- HYPO: Hyperspherical Out-Of-Distribution Generalization (ICLR 2024 · similarity 0.86)
- Beyond Federated Prototype Learning: Learnable Semantic Anchors with Hyperspherical Contrast for Domain-Skewed Data (AAAI 2025 · similarity 0.85)
- A Robust Prototype-Based Network with Interpretable RBF Classifier Foundations (AAAI 2025 · similarity 0.85)
- MCPNet: An Interpretable Classifier via Multi-Level Concept Prototypes (CVPR 2024 · similarity 0.83)
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.