Research map / Architectures and efficiency / RNNs and normalization
Learning towards Minimum Hyperspherical Energy
Weiyang Liu, Rongmei Lin, Zhen Liu, Lixin Liu, Zhiding Yu, Bo Dai, Le Song
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 |
|---|---|---|---|
| add MHE regularization (s=2) | CNN-9 · CIFAR-100 | test error (%): 28.13 -> 27.15 | Improved |
| use half-space MHE instead of full MHE (ablation) | CNN-9 · CIFAR-100 | test error (%): 27.15 -> 25.61 | Improved |
| Euclidean vs angular distance in MHE (ablation) | CNN-9 · CIFAR-10 | test error (%): 6.22 -> 6.21 | No clear effect |
| add MHE to network of narrow width 16/32/64 (ablation) | CNN-9 · CIFAR-100 | test error (%): 47.72 -> 36.84 | Improved |
| add half-space MHE to widest network 256/512/1024 (ablation) | CNN-9 · CIFAR-100 | test error (%): 25.45 -> 21.83 | Improved |
| add MHE to 15-layer CNN (ablation) | CNN-15 · CIFAR-100 | test error (%): N/C -> 26.9 | Improved |
| add MHE to 6-layer CNN (ablation) | CNN-6 · CIFAR-100 | test error (%): 32.08 -> 28.16 | Improved |
| apply MHE to both hidden and output layers vs hidden only (ablation) | CNN-9 MHE · CIFAR-100 | test error (%): 26.85 -> 26.16 | Improved |
| vary MHE hyperparameter from 10^-2 to 10^2 (ablation) | CNN-9 MHE · CIFAR-100 | test error (%) | No clear effect |
| add MHE to ResNet-32 | ResNet-32 · CIFAR-10 | test error (%): 5.19 -> 4.72 | Improved |
| add half-space MHE to ResNet-34 | ResNet-34 · ImageNet-2012 val | top-1 error (%): 30.04 -> 29.50 | Improved |
| replace orthonormal regularizer with MHE | ResNet-18 · ImageNet-2012 val | top-1 error (%): 33.61 -> 33.50 | Improved |
Closest papers by meaning
- Revisiting Orbital Minimization Method for Neural Operator Decomposition (NeurIPS 2025 · similarity 0.85)
- Refined generalization analysis of the Deep Ritz Method and Physics-Informed Neural Networks (ICML 2025 · similarity 0.85)
- Improving Energy Natural Gradient Descent through Woodbury, Momentum, and Randomization (NeurIPS 2025 · similarity 0.84)
- Diagonal Symmetrization of Neural Network Solvers for the Many-Electron Schrödinger Equation (ICML 2025 · similarity 0.84)
- One More Time: Revisiting Neural Quantum States from a Reinforcement Learning Perspective (NeurIPS 2026 · similarity 0.84)
- Neural Pfaffians: Solving Many Many-Electron Schrödinger Equations (NeurIPS 2024 · 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.
