Research map / Architectures and efficiency / Neural architecture search
AOWS: Adaptive and Optimal Network Width Search With Latency Constraints
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 |
|---|---|---|---|
| Replace greedy channel trimming with global Viterbi width selection (ablation) | MobileNet-v1 width search · ImageNet, TensorRT, 0.04ms target | Top-1 error (%): 29.3 -> 28.2 | Helped |
| Add adaptive channel-configuration sampling to global Viterbi selection (ablation) | OWS · ImageNet, TensorRT, 0.04ms target | Top-1 error (%): 28.2 -> 27.8 | Helped |
| Train AOWS for 10 rather than 20 epochs (ablation) | AOWS · ImageNet, TensorRT, 0.04ms target | Top-1 error (%): 27.8 -> 28.1 | Hurt |
| Train AOWS for 30 rather than 20 epochs (ablation) | AOWS · ImageNet, TensorRT, 0.04ms target | Top-1 error (%): 27.8 -> 27.9 | No clear effect |
| Use AOWS instead of greedy search | MobileNet-v1 · ImageNet, FLOPs target 150 MFLOPs | Top-1 error (%): 35.8 -> 35.9 | No clear effect |
| Use AOWS instead of greedy search | MobileNet-v1 · ImageNet, FLOPs target 325 MFLOPs | Top-1 error (%): 31.0 -> 29.7 | Helped |
| Use AOWS instead of greedy search | MobileNet-v1 · ImageNet, FLOPs target 572 MFLOPs | Top-1 error (%): 27.6 -> 26.7 | Helped |
| Optimize for realistic CPU latency with AOWS rather than the greedy comparison configurations | MobileNet-v1 · ImageNet, CPU latency | Top-1 error (%): 35.8 -> 33.8 | Helped |
| Optimize for realistic GPU latency with AOWS rather than the greedy comparison configurations | MobileNet-v1 · ImageNet, GPU latency | Top-1 error (%): 35.8 -> 28.5 | Helped |
| Use AOWS rather than AutoSlim | MobileNet-v1 · ImageNet, FLOPs target 150 MFLOPs | Top-1 error (%): 32.1 -> 35.9 | Hurt |
| Use AOWS rather than AutoSlim | MobileNet-v1 · ImageNet, FLOPs target 325 MFLOPs | Top-1 error (%): 28.5 -> 29.7 | Hurt |
| Use AOWS rather than AutoSlim | MobileNet-v1 · ImageNet, FLOPs target 572 MFLOPs | Top-1 error (%): 27.0 -> 26.7 | Helped |
Closest papers by meaning
- Neural Architecture Search on ImageNet in Four GPU Hours: A Theoretically Inspired Perspective (ICLR 2021 · similarity 0.90)
- NetAdaptV2: Efficient Neural Architecture Search With Fast Super-Network Training and Architecture Optimization (CVPR 2021 · similarity 0.90)
- Block-Wisely Supervised Neural Architecture Search With Knowledge Distillation (CVPR 2020 · similarity 0.89)
- Efficient Few-Shot Neural Architecture Search by Counting the Number of Nonlinear Functions (AAAI 2025 · similarity 0.89)
- Subnet-Aware Dynamic Supernet Training for Neural Architecture Search (CVPR 2025 · 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.