Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
PACOH: Bayes-Optimal Meta-Learning with PAC-Guarantees
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 |
|---|---|---|---|
| Meta-learn the prior with PACOH-NN instead of using a standard BNN prior | BNN · Omniglot | accuracy: 0.795 -> 0.885 | Helped |
| Meta-learn the prior with PACOH-NN instead of using a standard BNN prior | BNN · Omniglot | calibration error: 0.135 -> 0.091 | Helped |
| Use PACOH-GP rather than MLL-GP without meta-level regularization | GP prior meta-learning · synthetic regression environments | meta-test RMSE | Helped |
| Add a hyper-prior as meta-level regularization | GP prior meta-learning · synthetic regression environments, especially with fewer than 100 tasks | meta-test RMSE | Helped |
| Meta-learn a BNN prior with PACOH-NN rather than use a standard BNN/GP prior | vaccine peptide Bayesian optimization · MHC-I allele A-6901 | regret | Helped |
| Use PACOH-NN instead of a vanilla BNN | BNN uncertainty estimates · regression environments | calibration error | Helped |
| Use PACOH-GP instead of a Vanilla GP | GP uncertainty estimates · regression environments | calibration error | Mixed |
| Use PACOH-NN rather than maintain nested task and hyper-posterior optimization | BNN meta-learning · meta-training as task count grows | memory and compute scaling | Helped |
| Use MAML rather than a standard BNN | Omniglot classification · Omniglot | accuracy: 0.795 -> 0.693 | Hurt |
| Remove meta-level regularization by using MLL-GP (ablation) | PACOH-GP · synthetic regression environments | meta-test RMSE | Hurt |
Closest papers by meaning
- More Flexible PAC-Bayesian Meta-Learning by Learning Learning Algorithms (ICML 2024 · similarity 0.93)
- Learning via Surrogate PAC-Bayes (NeurIPS 2024 · similarity 0.90)
- Recursive PAC-Bayes: A Frequentist Approach to Sequential Prior Updates with No Information Loss (NeurIPS 2024 · similarity 0.89)
- Generalization Bounds via Meta-Learned Model Representations: PAC-Bayes and Sample Compression Hypernetworks (ICML 2025 · similarity 0.89)
- Fast Rate Bounds for Multi-Task and Meta-Learning with Different Sample Sizes (NeurIPS 2025 · similarity 0.88)
- MALIBO: Meta-learning for Likelihood-free Bayesian Optimization (ICML 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.