Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
Hyperparameter Learning via Distributional Transfer
Ho Chung Law, Peilin Zhao, Leung Sing Chan, Junzhou Huang, Dino Sejdinovic
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 learned dataset-distribution embeddings to transfer information from source tasks | distBO · Unsupervised toy | maximum observed objective | Improved |
| Use learned dataset-distribution embeddings instead of handcrafted meta-features | distBO · Regression: handcrafted meta-features counterexample | maximum observed R² | Improved |
| Use dataset meta-information to transfer across source tasks | distGP · Protein dataset | maximum observed accuracy | Improved |
| Use learned dataset representations for transfer when a source task has matching underlying bandwidths | distBLR · Classification task experiment A | maximum observed AUC | Improved |
| Transfer dataset meta-information when no similar source task exists | distBLR · Classification task experiment B | maximum observed AUC | Hurt |
| Use learned distribution embeddings rather than handcrafted meta-features | distBLR · Classification task experiment A | convergence speed | Improved |
| Warm-start with top hyperparameters from tasks selected using handcrafted meta-features | initBO · Regression: handcrafted meta-features counterexample | convergence speed | Hurt |
| Use all source-task evaluations rather than only selected previous best hyperparameters | distBO · Regression: handcrafted meta-features counterexample | convergence speed | Improved |
| Embed the joint distribution instead of separately embedding marginal and conditional distributions (ablation) | distBO · Regression: joint versus concatenation embeddings | performance | No clear effect |
| Use dataset meta-information for a target task dissimilar to all source tasks | distBLR · Classification task experiment B | convergence speed | Improved |
Closest papers by meaning
- A Quantile-based Approach for Hyperparameter Transfer Learning (ICML 2020 · similarity 0.86)
- On the Theory of Transfer Learning: The Importance of Task Diversity (NeurIPS 2020 · similarity 0.85)
- Provable Sample-Efficient Transfer Learning Conditional Diffusion Models via Representation Learning (NeurIPS 2025 · similarity 0.85)
- Bayesian Deconditional Kernel Mean Embeddings (ICML 2019 · similarity 0.85)
- Scalable Hyperparameter Transfer Learning (NeurIPS 2018 · similarity 0.85)
- Supervised Distributional Reduction via Optimal Transport and Dependence Maximization (NeurIPS 2026 · similarity 0.84)
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.
