Research map / Architectures and efficiency / Neural architecture search
SGAS: Sequential Greedy Architecture Search
Guohao Li, Guocheng Qian, Itzel C. Delgadillo, Matthias Muller, Ali Thabet, Bernard Ghanem
Check 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 sequential greedy operation selection and pruning rather than selecting all operations at the end | DARTS-style architecture search · CIFAR-10 | Test Err. (%): 2.76±0.09 -> 2.39 | Improved |
| Use sequential greedy operation selection and pruning rather than selecting all operations at the end | DARTS-style architecture search · CIFAR-10 | Search-evaluation Kendall τ: −0.29 -> 0.56 | Improved |
| Add selection stability to the criterion (ablation) | SGAS · CIFAR-10 | Search-evaluation Kendall τ: 0.56 -> 0.42 | Hurt |
| Add selection stability to the criterion (ablation) | SGAS · ImageNet | Test error top-1 (%): 24.2 -> 24.1 | Improved |
| Add selection stability to the criterion (ablation) | SGAS · ModelNet40 | OA (%): 92.87 -> 93.23 | Improved |
| Add selection stability to the criterion (ablation) | SGAS · PPI | micro-F1 (%): 99.38±0.17 -> 99.40±0.09; best: 99.46 -> 99.46 | Mixed |
| Reduce the greedy decision interval from T=5 to T=3 (ablation) | SGAS (Cri.2) · CIFAR-10 | Test Err. (%): 2.67±0.21 -> 2.86±0.12 | Hurt |
| Increase the greedy decision interval from T=5 to T=7 (ablation) | SGAS (Cri.2) · CIFAR-10 | Test Err. (%): 2.67±0.21 -> 2.65±0.08 | No clear effect |
| Reduce the history window from K=4 to K=2 (ablation) | SGAS (Cri.2) · CIFAR-10 | Test Err. (%): 2.67±0.21 -> 2.73±0.16 | Hurt |
| Increase the history window from K=4 to K=6 (ablation) | SGAS (Cri.2) · CIFAR-10 | Test Err. (%): 2.67±0.21 -> 2.67±0.17 | No clear effect |
| Reduce channel size from 512 to 64 (ablation) | SGAS (Cri.1 best), 5-cell PPI model · PPI | micro-F1 (%): 99.462 -> 98.894 | Hurt |
| Reduce channel size from 512 to 128 (ablation) | SGAS (Cri.1 best), 5-cell PPI model · PPI | micro-F1 (%): 99.462 -> 99.369 | Hurt |
Closest papers by meaning
- Bridging the Gap between Sample-based and One-shot Neural Architecture Search with BONAS (NeurIPS 2020 · similarity 0.89)
- Fitting the Search Space of Weight-sharing NAS with Graph Convolutional Networks (AAAI 2021 · similarity 0.89)
- Evaluating The Search Phase of Neural Architecture Search (ICLR 2020 · similarity 0.89)
- Neural Architecture Search on ImageNet in Four GPU Hours: A Theoretically Inspired Perspective (ICLR 2021 · similarity 0.89)
- 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.
