Research map / Learning theory and algorithms / Kernels and regression theory
Generalization Bounds for Meta-Learning via PAC-Bayes and Uniform Stability
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 |
|---|---|---|---|
| Combine base-level uniform stability with meta-level PAC-Bayes | PAC-BUS · Classification on Ball | generalization bound: 0.3422±0.0006 -> 0.2213±0.0012 | Helped |
| Use PAC-BUS instead of MR-MAML | Classification on Ball · Classification on Ball | test loss: 0.1584±0.0012 -> 0.1657±0.0014 | Hurt |
| Combine base-level uniform stability with meta-level PAC-Bayes | PAC-BUS · Mini-Wiki 4-way 1-, 3-, and 5-shot | generalization bound: 0.6638±0.0011 -> 0.4999±0.0003; 0.6366±0.0006 -> 0.5058±0.0002; 0.6343±0.0014 -> 0.5101±0.0002 | Helped |
| Use PAC-BUS instead of MAML | Mini-Wiki · Mini-Wiki 4-way 1-, 3-, and 5-shot | meta-test loss: 0.3916±0.0009 -> 0.3922±0.0009; 0.3868±0.0005 -> 0.3878±0.0003; 0.3883±0.0005 -> 0.3895±0.0005 | No clear effect |
| Add PAC-BUS(H)'s KL-to-prior and uniform-stability regularizers | Omniglot NME · Omniglot NME 20-way 1-shot | meta-test accuracy (%): 23.4±2.2 -> 87.9±0.5; 22.4±0.5 -> 87.9±0.5; 84.2±2.2 -> 87.9±0.5 | Helped |
| Add PAC-BUS(H)'s KL-to-prior and uniform-stability regularizers | Omniglot NME · Omniglot NME 20-way 5-shot | meta-test accuracy (%): 75.1±4.8 -> 95.0±0.9; 39.1±0.5 -> 95.0±0.9; 94.3±0.3 -> 95.0±0.9 | Helped |
Closest papers by meaning
- Fast-Rate PAC-Bayesian Generalization Bounds for Meta-Learning (ICML 2022 · similarity 0.94)
- On the Stability and Generalization of Meta-Learning (NeurIPS 2024 · similarity 0.91)
- Certification-Enhanced Generalization Bounds (NeurIPS 2026 · similarity 0.91)
- Fast Rate Bounds for Multi-Task and Meta-Learning with Different Sample Sizes (NeurIPS 2025 · similarity 0.90)
- Generalization Bounds via Meta-Learned Model Representations: PAC-Bayes and Sample Compression Hypernetworks (ICML 2025 · similarity 0.90)
- More Flexible PAC-Bayesian Meta-Learning by Learning Learning Algorithms (ICML 2024 · similarity 0.89)
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.