Research map / Architectures and efficiency / Neural architecture search
Efficient Forward Architecture Search
Read the paperCheck 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 |
|---|---|---|---|
| Grow a macro-search network by adding selected shortcut connections | modified ResNet seed · CIFAR-10 | test error: 4.6 -> 2.85± 0.12 | Helped |
| Grow a cell-search network by adding selected shortcut connections | modified ResNet seed · CIFAR-10 | test error: 4.6 -> 2.87 ± 0.13 | Helped |
| Transfer the searched cell architecture to ImageNet | Petridish cell · ILSVRC2012 | top-1 test error: 26.3 ± 0.20 | Helped |
| Search recurrent cells through iterative growth | Petridish · PTB | test perplexity: 55.85 | 56.39± 0.38; Random search WS: 55.5 | Mixed |
| Search in proxy space rather than directly in final-scale space (ablation) | Petridish cell · ILSVRC2012 | top-1 test error: 26.9 -> 26.3 | Helped |
| Add concat-projection each time weak learners are finalized instead of retaining weighted sums (ablation) | Petridish macro · ILSVRC2012 | top-1 test error: 32.5 -> 29.4 | Helped |
| Add concat-projection each time weak learners are finalized instead of retaining weighted sums (ablation) | Petridish cell · ILSVRC2012 | top-1 test error: 32.7 -> 26.9 | Helped |
| Add concat-projection during weak-learner finalization rather than only switching in the final model (ablation) | Petridish macro · ILSVRC2012 | top-1 test error: 29.1 -> 29.4 | No clear effect |
| Add concat-projection during weak-learner finalization rather than only switching in the final model (ablation) | Petridish cell · ILSVRC2012 | top-1 test error: 27.2 -> 26.9 | Helped |
| Isolate weak learners from the existing model during weak learning instead of jointly training random weak learners with it (ablation) | Petridish cell · ILSVRC2012 | top-1 test error: 32.8 -> 26.9 | Helped |
| Select 3 operations per weak learner rather than 2 (ablation) | Petridish · CIFAR-10 | average lowest error rate: 3.08 -> 2.88 | Helped |
| Select 3 operations per weak learner rather than 4 (ablation) | Petridish · CIFAR-10 | average lowest error rate: 2.93 -> 2.88 | Helped |
Closest papers by meaning
- Per-Architecture Training-Free Metric Optimization for Neural Architecture Search (NeurIPS 2025 · similarity 0.90)
- AdvantageNAS: Efficient Neural Architecture Search with Credit Assignment (AAAI 2021 · similarity 0.89)
- Neural Architecture Generator Optimization (NeurIPS 2020 · similarity 0.89)
- EG-NAS: Neural Architecture Search with Fast Evolutionary Exploration (AAAI 2024 · similarity 0.88)
- Boosting Order-Preserving and Transferability for Neural Architecture Search: a Joint Architecture Refined Search and Fine-tuning Approach (CVPR 2024 · similarity 0.88)
- Efficient Few-Shot Neural Architecture Search by Counting the Number of Nonlinear Functions (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.