Research map / Learning theory and algorithms / Kernels and regression theory
Theoretical Characterization of the Generalization Performance of Overfitted Meta-Learning
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 |
|---|---|---|---|
| Increase number of features/parameters p (overparameterization) in overfitted min l2-norm MAML | MAML linear model with Gaussian features · theoretical model error bound | model error of ideal interpolator | Helped |
| Increase task diversity (nu) so that overparameterization yields double descent even with zero noise | MAML linear model with Gaussian features · simulation with nu=60, sigma=0 | model error | Mixed |
| Set alpha_r = n_r/(n_r+p+1) for the test-task one-step adaptation instead of optimal step | test-task one-step gradient adaptation · theoretical test error | expected test error | No clear effect |
| Overparameterized min l2-norm solution (large p) versus underparameterized solution (p=s=1 or p=5=s) | MAML with large noise and task diversity · theoretical comparison and simulation (Fig. a, p=5 vs p=1000) | model error | Helped |
| Increase noise sigma and task diversity nu | overfitted min l2-norm MAML · theoretical model error curve | descent floor value and location | Hurt |
Closest papers by meaning
- Provable Generalization of Overparameterized Meta-learning Trained with SGD (NeurIPS 2022 · similarity 0.92)
- Understanding Benign Overfitting in Gradient-Based Meta Learning (NeurIPS 2022 · similarity 0.92)
- Towards Sample-efficient Overparameterized Meta-learning (NeurIPS 2021 · similarity 0.92)
- Improving Generalization of Meta-Learning With Inverted Regularization at Inner-Level (CVPR 2023 · similarity 0.91)
- Benign Overfitting in Adversarial Training of Neural Networks (ICML 2024 · similarity 0.87)
- Improving Generalization via Meta-Learning on Hard Samples (CVPR 2024 · similarity 0.87)
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.