Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Auto-Sklearn 2.0: Hands-free AutoML via Meta-Learning
Matthias Feurer, Katharina Eggensperger, Stefan Falkner, Marius Lindauer, Frank Hutter
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 |
|---|---|---|---|
| add portfolio warmstarting instead of plain BO | Auto-sklearn (FB; holdout) · AutoML benchmark 39 datasets (10 min) | normalized balanced error rate (ADTM): 5.98 -> 3.70 | Improved |
| replace KND warmstarting with a greedy portfolio | Auto-sklearn (FB; holdout) · AutoML benchmark 39 datasets (10 min) | normalized balanced error rate (ADTM): 5.29 -> 3.70 | Improved |
| use successive halving instead of full budget with portfolio and holdout | Auto-sklearn · AutoML benchmark 39 datasets (10 min) | normalized balanced error rate (ADTM): 3.70 -> 4.11 | No clear effect |
| add per-dataset policy selector on top of PoSH-Auto-sklearn | PoSH-Auto-sklearn · AutoML benchmark 39 datasets (10 min) | normalized balanced error rate (ADTM): 4.11 -> 3.58 | Improved |
| add per-dataset policy selector on top of PoSH-Auto-sklearn | PoSH-Auto-sklearn · AutoML benchmark 39 datasets (60 min) | normalized balanced error rate (ADTM): 3.19 -> 2.47 | Improved |
| Auto-sklearn 2.0 versus Auto-sklearn 1.0 | Auto-sklearn 1.0 · AutoML benchmark 39 datasets (10 min) | normalized balanced error rate (ADTM): 16.21 -> 3.58 | Improved |
| add per-dataset policy selector instead of single best policy (ablation) | Auto-sklearn 2.0 · AutoML benchmark 39 datasets (10 min) | normalized balanced error rate (ADTM): 3.69 -> 3.58; 2.44 -> 2.47 | Mixed |
| add fallback to holdout with SH for out-of-distribution datasets (ablation) | policy selector · AutoML benchmark 39 datasets (60 min) | normalized balanced error rate (ADTM): 5.43 -> 2.47 | Improved |
| add portfolio warmstarting to policy selector system (ablation) | policy selector · AutoML benchmark 39 datasets (10 min) | normalized balanced error rate (ADTM): 5.63 -> 3.58 | Improved |
| add portfolio warmstarting to policy selector system (ablation) | policy selector · AutoML benchmark 39 datasets (60 min) | normalized balanced error rate (ADTM): 3.42 -> 2.47 | Improved |
| allow cross-validation policies in addition to holdout (ablation) | policy selector · AutoML benchmark 39 datasets (10 min) | normalized balanced error rate (ADTM): 4.03 -> 3.58 | Improved |
| restrict policy set to full budget only (ablation) | policy selector · AutoML benchmark 39 datasets (10 min) | normalized balanced error rate (ADTM): 3.58 -> 3.50 | No clear effect |
Closest papers by meaning
- Towards Automated Semi-Supervised Learning (AAAI 2019 · similarity 0.89)
- AutoML-Zero: Evolving Machine Learning Algorithms From Scratch (ICML 2020 · similarity 0.89)
- Evolving Fully Automated Machine Learning via Life-Long Knowledge Anchors (TPAMI 2021 · similarity 0.88)
- Pipeline Combinators for Gradual AutoML (NeurIPS 2021 · similarity 0.88)
- Position: A Call to Action for a Human-Centered AutoML Paradigm (ICML 2024 · similarity 0.88)
- MetaOptimize: A Framework for Optimizing Step Sizes and Other Meta-parameters (ICML 2025 · similarity 0.86)
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.
