Research map / Architectures and efficiency / Neural architecture search
Rethinking Performance Estimation in Neural Architecture Search
Check 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 MIP-selected BPE-1 for CIFAR-10 architecture search | RL, EA, DARTS, and random search · CIFAR-10 test | search cost: 0.05 to 0.33 GPU days for proposed methods; 3150 GPU days for AmoebaNet-B | Helped |
| Use BPE-1 with DARTS | DARTS · CIFAR-10 test | test error: 2.7 ± 0.01 -> 2.89 ± 0.0 | Hurt |
| Use BPE-2 with DARTS | DARTS · CIFAR-10 test | test error: 2.7 ± 0.01 -> 2.72 ± 0.0 | Hurt |
| Use MIP-selected BPE-1 | NAS methods transferred from CIFAR-10 to ImageNet · ImageNet validation, mobile setting | Top-1 accuracy: 73.1 -> 74.0, 74.18, 74.2, 74.56 | Helped |
| Optimize for Spearman rank correlation rather than only the estimated absolute performance | BPE selection · CIFAR-10 sampled architectures | correlation with found-architecture performance: 0.65 correlation | Helped |
| Use the DARTS training settings instead of BPE-1 | CIFAR-10 BPE · CIFAR-10 BPE search | Spearman rank correlation: 0.50 -> 0.57 | Helped |
| Use BPE-1 instead of the DARTS training settings | CIFAR-10 BPE · CIFAR-10 BPE search | training time: 1.38 -> 0.08 | Helped |
| Use BPE-2 instead of the DARTS training settings | CIFAR-10 BPE · CIFAR-10 BPE search | Spearman rank correlation: 0.57 -> 0.63 | Helped |
| Use BPE-2 instead of the DARTS training settings | CIFAR-10 BPE · CIFAR-10 BPE search | training time: 1.38 -> 0.55 | Helped |
| Use local rather than global performance ranking for one-shot networks | One-shot methods · One-shot evaluation, epoch 200 | Spearman rank correlation: -0.06 -> 0.50 | Helped |
Closest papers by meaning
- Speedy Performance Estimation for Neural Architecture Search (NeurIPS 2021 · similarity 0.92)
- Evaluating Efficient Performance Estimators of Neural Architectures (NeurIPS 2021 · similarity 0.90)
- Neural Architecture Search With Representation Mutual Information (CVPR 2022 · similarity 0.90)
- Robustifying and Boosting Training-Free Neural Architecture Search (ICLR 2024 · similarity 0.88)
- MOTE-NAS: Multi-Objective Training-based Estimate for Efficient Neural Architecture Search (NeurIPS 2024 · similarity 0.88)
- HEP-NAS: Towards Efficient Few-shot Neural Architecture Search via Hierarchical Edge Partitioning (AAAI 2025 · similarity 0.88)
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.