Research map / Architectures and efficiency / Neural architecture search
Auto-Pytorch: Multi-Fidelity MetaLearning for Efficient and Robust AutoDL
Lucas Zimmer, 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 |
|---|---|---|---|
| use BOHB multi-fidelity instead of plain BO (ablation) | Auto-PyTorch · 8 meta-test OpenML datasets | anytime performance: BOHB performed as well or better than plain BO | Improved |
| add greedy portfolio warmstarting (ablation) | Auto-PyTorch BOHB · 8 meta-test OpenML datasets | anytime performance | Improved |
| add simple portfolio of all 100 incumbents (ablation) | Auto-PyTorch BOHB · 8 meta-test OpenML datasets | anytime performance | Mixed |
| add ensembles of DNNs across fidelities (ablation) | Auto-PyTorch · 8 meta-test OpenML datasets | error over time | Improved |
| add diverse ensembles including RF, ExtraTrees, LightGBM, CatBoost, KNN (ablation) | Auto-PyTorch · 8 meta-test OpenML datasets | error over time | Improved |
| add ensembling (ablation) | Auto-PyTorch Tabular · covertype | accuracy (%): 93.35 -> 96.86 | Improved |
| add ensembling (ablation) | Auto-PyTorch Tabular · dilbert | accuracy (%): 98.54 -> 98.70 | Improved |
| add ensembling (ablation) | Auto-PyTorch Tabular · car | accuracy (%): 98.59 -> 96.41 | Hurt |
| run 3 parallel workers instead of sequential | Auto-PyTorch BOHB · larger OpenML datasets | speedup: speedups of 3x or more | Improved |
| use Auto-PyTorch Tabular instead of AutoNet2.0 | Auto-PyTorch · covertype | misclassification rate (%): 31.78% -> 3.14% | Improved |
| use adaptive instead of non-adaptive learning rate schedule (ablation) | LCBench configs · budget pair (12, 50) | Spearman rank correlation: 0.91 -> 0.88 | Hurt |
| use adaptive instead of non-adaptive learning rate schedule (ablation) | LCBench configs · budget pair (12, 25) | Spearman rank correlation: 0.94 -> 0.95 | No clear effect |
Closest papers by meaning
- PyTorch: An Imperative Style, High-Performance Deep Learning Library (NeurIPS 2019 · similarity 0.87)
- AutoKeras: An AutoML Library for Deep Learning (JMLR 2023 · similarity 0.87)
- AutoAL: Automated Active Learning with Differentiable Query Strategy Search (ICML 2025 · similarity 0.85)
- MLZero: A Multi-Agent System for End-to-end Machine Learning Automation (NeurIPS 2025 · similarity 0.84)
- AutoML-Agent: A Multi-Agent LLM Framework for Full-Pipeline AutoML (ICML 2025 · similarity 0.84)
- Pfeife: Automatic Pipeline Parallelism for PyTorch (ICML 2025 · 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.
