Research map / Architectures and efficiency / Neural architecture search
One-Shot Neural Architecture Search via Self-Evaluated Template Network
Xuanyi Dong, Yi Yang
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 |
|---|---|---|---|
| train the template network stochastically (SETN) instead of standard classification training (SETN-NON) (ablation) | SETN · CIFAR-10 (with CutOut) | test error (%): 3.12 -> 2.69 | Improved |
| train the template network stochastically (SETN) instead of standard classification training (SETN-NON) (ablation) | SETN · CIFAR-100 (with CutOut) | test error (%): 18.27 -> 17.25 | Improved |
| use independent per-cell sampling (SETN) instead of sharing indexes across cells (SETN-LR) (ablation) | SETN · CIFAR-10 (with CutOut) | test error (%): 2.81 -> 2.69 | Improved |
| use independent per-cell sampling (SETN) instead of sharing indexes across cells (SETN-LR) (ablation) | SETN · CIFAR-100 (with CutOut) | test error (%): 17.88 -> 17.25 | Improved |
| evaluate T=1K sampled candidates instead of arg max choice (T=1) (ablation) | SETN · CIFAR-10 (with CutOut) | test error (%): 3.41 -> 2.69 | Improved |
| evaluate T=1K sampled candidates instead of arg max choice (T=1) (ablation) | SETN · CIFAR-100 (with CutOut) | test error (%): 18.12 -> 17.25 | Improved |
| add CutOut augmentation (ablation) | SETN (T=1K) · CIFAR-10 | test error (%): 3.56 -> 2.69 | Improved |
| add CutOut augmentation (ablation) | SETN (T=1K) · CIFAR-100 | test error (%): 19.38 -> 17.25 | Improved |
| enlarge search space with two dilated convolutions (100x larger) (ablation) | SETN · CIFAR-10 | test error | No clear effect |
| use SETN-learned candidates instead of random candidates (ablation) | SETN · one-shot candidate quality | one-shot validation accuracy | Improved |
| use SETN instead of SETN-LR for candidate generation (ablation) | SETN · one-shot candidate quality | one-shot validation accuracy | Mixed |
| use template-parameter validation accuracy to rank networks (ablation) | SETN · pairwise ranking of 2000 networks | ranking agreement: >80% of pairs good | Improved |
Closest papers by meaning
- Understanding and Simplifying One-Shot Architecture Search (ICML 2018 · similarity 0.91)
- Few-Shot Neural Architecture Search (ICML 2021 · similarity 0.90)
- Neural Architecture Search on ImageNet in Four GPU Hours: A Theoretically Inspired Perspective (ICLR 2021 · similarity 0.90)
- Evaluating The Search Phase of Neural Architecture Search (ICLR 2020 · similarity 0.90)
- Efficient Few-Shot Neural Architecture Search by Counting the Number of Nonlinear Functions (AAAI 2025 · similarity 0.89)
- Prior Knowledge Guided Neural Architecture Generation (ICML 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.
