Research map / Architectures and efficiency / Neural architecture search
Angle-based Search Space Shrinking for Neural Architecture Search
Yiming Hu, Yuding Liang, Zichao Guo, Ruosi Wan, Xiangyu Zhang, Yichen Wei, Qingyi Gu, Jian Sun
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 |
|---|---|---|---|
| apply ABS shrinking before search | FairNAS · ImageNet, MobileNet-like space | top-1 accuracy: 74.24% -> 74.42% | Improved |
| apply ABS shrinking before search | SPOS · ImageNet, MobileNet-like space | top-1 accuracy: 75.33% -> 75.97% | Improved |
| apply ABS shrinking before search | ProxylessNAS · ImageNet, MobileNet-like space | top-1 accuracy: 75.56% -> 76.14% | Improved |
| apply ABS shrinking before search | DARTS · ImageNet, DARTS space | top-1 accuracy: 73.39% -> 75.59 | Improved |
| apply ABS shrinking before search | PDARTS · ImageNet, DARTS space | top-1 accuracy: 75.02% -> 75.89 | Improved |
| scale channels down to 45 to match FLOPs after ABS | DARTS+ABS and PDARTS+ABS · ImageNet, DARTS space | top-1 accuracy: DARTS 75.59 -> 75.19; PDARTS 75.89 -> 75.64 (still above baselines 73.39%, 75.02%) | No clear effect |
| use angle metric instead of accuracy (Re-BN) to rank child models (ablation) | supernet ranking · NAS-Bench-201 CIFAR-10 | Kendall's Tau: 0.5436 -> 0.5748 | Improved |
| use angle metric instead of accuracy (Re-BN) to rank child models (ablation) | supernet ranking · NAS-Bench-201 CIFAR-100 | Kendall's Tau: 0.5329 -> 0.6040 | Improved |
| use angle metric instead of accuracy (Re-BN) to rank child models (ablation) | supernet ranking · NAS-Bench-201 ImageNet-16-120 | Kendall's Tau: 0.5391 -> 0.5445 | Improved |
| use angle metric instead of accuracy metric for computing child model scores (ablation) | metric computation · NAS-Bench-201 CIFAR-10 | processing time (s): 561.75 -> 0.92 | Improved |
| change weight initialization (Kaiming, Xavier, orthogonal) (ablation) | angle metric · NAS-Bench-201 CIFAR-10 | Kendall's Tau: 0.622 / 0.609 / 0.609 | No clear effect |
| use ABS-found shrunk space instead of human-designed best space S8 (ablation) | five NAS algorithms · NAS-Bench-201 CIFAR-100 | accuracy: DARTS 65.70% -> 67.57%; SETN 70.12% -> 69.12%; GDAS 71.01% -> 70.35%; SPOS 70.49% -> 71.95% | Mixed |
Closest papers by meaning
- Evolving Search Space for Neural Architecture Search (ICCV 2021 · similarity 0.90)
- AtomNAS: Fine-Grained End-to-End Neural Architecture Search (ICLR 2020 · similarity 0.90)
- Understanding and Accelerating Neural Architecture Search With Training-Free and Theory-Grounded Metrics (TPAMI 2023 · similarity 0.90)
- Per-Architecture Training-Free Metric Optimization for Neural Architecture Search (NeurIPS 2025 · similarity 0.90)
- Efficient Few-Shot Neural Architecture Search by Counting the Number of Nonlinear Functions (AAAI 2025 · similarity 0.89)
- einspace: Searching for Neural Architectures from Fundamental Operations (NeurIPS 2024 · similarity 0.89)
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.
