Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Meta-Curvature
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 |
|---|---|---|---|
| Learn gradient-transform curvature matrices jointly with the initialization | MAML · Sinusoidal regression, 5-shot | MSE: 0.686 $\pm$ 0.070 -> \textbf{0.405} $\pm$ \textbf{0.048} | Helped |
| Learn all three curvature matrices instead of fixing the output-channel matrix to identity (ablation) | MC1 · Sinusoidal regression, 10-shot | MSE: 0.239 $\pm$ 0.025 -> \textbf{0.201} $\pm$ \textbf{0.020} | Helped |
| Learn gradient-transform curvature matrices jointly with the initialization | MAML · miniImagenet test, 5-way 1-shot, 1 inner step, 4-layer CNN | accuracy: 46.28 $\pm$ 0.89 -> \textbf{54.23} $\pm$ \textbf{0.88} | Helped |
| Learn all three curvature matrices instead of fixing the output-channel matrix to identity (ablation) | MC1 · miniImagenet test, 5-way 1-shot, 1 inner step, 4-layer CNN | accuracy: \textbf{53.37} $\pm$ \textbf{0.88} -> \textbf{54.23} $\pm$ \textbf{0.88} | Helped |
| Learn gradient-transform curvature matrices jointly with the initialization | MAML · Omniglot, 20-way 1-shot | accuracy: 95.8 $\pm$ 0.3 -> 97.86 $\pm$ 0.26 | Helped |
| Ensemble three MC2 models | MC2 · Omniglot, 20-way 1-shot | accuracy: 97.86 $\pm$ 0.26 -> \textbf{99.12} $\pm$ \textbf{0.16} | Helped |
| Learn gradient-transform curvature matrices jointly with the initialization | MAML · Omniglot, 5-way 5-shot | accuracy: 99.9 $\pm$ 0.1 -> 99.79 $\pm$ 0.10 | No clear effect |
| Use three-model ensembling | MC2 · Omniglot, 5-way 5-shot | accuracy: 99.79 $\pm$ 0.10 -> 99.89 $\pm$ 0.06 | Helped |
| Use the learned curvature transformation during task adaptation | MAML · tieredImagenet, 5-way 5-shot, center features | accuracy: 69.04 $\pm$ 0.22 -> 82.21 $\pm$ 0.08 | Helped |
| Use meta-trained curvature | MAML · Meta-training across few-shot classification experiments | convergence rate | Helped |
Closest papers by meaning
- Global curvature for second-order optimization of neural networks (ICML 2025 · similarity 0.89)
- Riemannian Geometric-based Meta Learning (AAAI 2025 · similarity 0.88)
- Curvature Tuning: Provable Training-free Model Steering From a Single Parameter (NeurIPS 2025 · similarity 0.87)
- Kronecker-Factored Approximate Curvature for Physics-Informed Neural Networks (NeurIPS 2024 · similarity 0.86)
- CAMEx: Curvature-aware Merging of Experts (ICLR 2025 · similarity 0.86)
- FOSI: Hybrid First and Second Order Optimization (ICLR 2024 · 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.