Research map / Learning theory and algorithms / Sample complexity
A Distribution-dependent Analysis of 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 |
|---|---|---|---|
| Use the MLE meta-mean in weighted biased regularization, with weight set by the inverse task covariance | Fixed-design Gaussian linear regression with known task covariance · Fixed-design Gaussian linear regression | expected transfer risk: MLE plug-in risk equals the unbiased-estimator lower bound | Helped |
| Estimate the meta-mean from past tasks and use it to bias prediction on the target task | MLE plug-in with weighted BRLS · Fixed-design Gaussian linear regression | expected transfer risk: any predictor has lower bound xᵀMx/(16√e) + xᵀTx + σ²; MLE plug-in attains xᵀMx + xᵀTx + σ² | Helped |
| Use EM to estimate the unknown environment parameters | EM Learner · Synthetic Fourier and spherical experiments | test error | Helped |
| Use EM rather than the method-of-moments representation estimator | Representation learning · Representation learning experiment II | max-correlation | Helped |
| Use EM-estimated low-rank task structure rather than the method-of-moments estimator on the Fourier low-rank setting | Representation learning · Fourier synthetic experiment with low-rank covariance | test error | Mixed |
| Use EM-estimated coefficients biased toward the meta-mean rather than least squares in the known lower-dimensional subspace | EM Learner versus Oracle Representation · Fourier synthetic experiment with low-rank covariance | test error | Helped |
| Use meta-learning rather than ordinary regression | Biased Regression and EM Learner · School Dataset | test error | Helped |
Closest papers by meaning
- Provable Meta-Learning of Linear Representations (ICML 2021 · similarity 0.90)
- Learning To Learn Around A Common Mean (NeurIPS 2018 · similarity 0.88)
- Task-Robust Model-Agnostic Meta-Learning (NeurIPS 2020 · similarity 0.88)
- Generalization Bounds For Meta-Learning: An Information-Theoretic Analysis (NeurIPS 2021 · similarity 0.88)
- Theoretical Investigations and Practical Enhancements on Tail Task Risk Minimization in Meta Learning (NeurIPS 2024 · similarity 0.87)
- More Flexible PAC-Bayesian Meta-Learning by Learning Learning Algorithms (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.