Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Improving Generalization of Meta-Learning With Inverted Regularization at Inner-Level
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 |
|---|---|---|---|
| Add ordinary outer-level and inverted inner-level regularization | MAML++ · Mini-ImageNet 5-way 5-shot | accuracy: 70.02±0.55% -> 71.70±0.23% | Helped |
| Add ordinary outer-level and inverted inner-level regularization | MAML++ · Omniglot 20-way 1-shot | accuracy: 97.21±0.51% -> 97.77±0.06% | Helped |
| Add ordinary outer-level and inverted inner-level L2 regularization | MAML with three inner steps · PhysioNet-GCS | RMSE: 1.532±0.074 -> 1.483±0.052 | Helped |
| Add ordinary outer-level and inverted inner-level L2 regularization | MAML with three inner steps · SwissFEL | RMSE: 0.535±0.042 -> 0.477±0.026 | Helped |
| Add ordinary outer-level and inverted inner-level L2 regularization | MAML with three inner steps · PhysioNet-HCT | RMSE: 2.396±0.047 -> 2.343±0.019 | Helped |
| Add ordinary outer-level and inverted inner-level L2 regularization | MAML with three inner steps · Cauchy | RMSE: 0.212±0.003 -> 0.201±0.002 | Helped |
| Add ordinary outer-level L2 regularization (ablation) | MAML · Mini-ImageNet 5-way 5-shot | accuracy: 65.39±0.50% -> 66.47±1.21% | Helped |
| Add ordinary inner-level L2 regularization (ablation) | MAML · Mini-ImageNet 5-way 5-shot | accuracy: 65.39±0.50% -> 64.80±0.25% | Hurt |
| Add inverted inner-level L2 regularization (ablation) | MAML · Mini-ImageNet 5-way 5-shot | accuracy: 65.39±0.50% -> 66.05±0.68% | Helped |
| Add ordinary outer-level and inverted inner-level L2 regularization (ablation) | MAML · Mini-ImageNet 5-way 5-shot | accuracy: 65.39±0.50% -> 68.17±0.92% | Helped |
| Replace L2-only regularization with combined L2-Norm and output-entropy regularization (ablation) | MAML with Minimax-Meta Regularization · Mini-ImageNet 5-way 5-shot | accuracy: 68.17±0.92% -> 69.36±0.34% | Helped |
Closest papers by meaning
- Meta-Learning Requires Meta-Augmentation (NeurIPS 2020 · similarity 0.92)
- Improving Generalization in Meta-learning via Task Augmentation (ICML 2021 · similarity 0.92)
- On the Stability and Generalization of Meta-Learning: the Impact of Inner-Levels (NeurIPS 2025 · similarity 0.91)
- Theoretical Characterization of the Generalization Performance of Overfitted Meta-Learning (ICLR 2023 · similarity 0.91)
- MetaReg: Towards Domain Generalization using Meta-Regularization (NeurIPS 2018 · similarity 0.90)
- Task Groupings Regularization: Data-Free Meta-Learning with Heterogeneous Pre-trained Models (ICML 2024 · similarity 0.88)
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.