Research map / Architectures and efficiency / Neural architecture search
GreedyNAS: Towards Fast One-Shot NAS With Greedy Supernet
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 |
|---|---|---|---|
| train the supernet with greedy path filtering rather than uniform path sampling | GreedyNAS · ImageNet, FLOPs ≤322M | Top-1 accuracy (%): 74.50 -> 74.85 | Helped |
| use greedy path filtering and candidate-pool training | GreedyNAS · ImageNet, latency ≤80ms | corrected #optimization: 147.6M -> 89.7M | Helped |
| add path filtering (ablation) | GreedyNAS, Net1 to Net2 · ImageNet, search under 330M FLOPs | Top-1 accuracy (%): 74.31 -> 74.59 | Helped |
| add candidate-pool sampling, evolutionary initialization, and pool-based stopping (ablation) | GreedyNAS, Net2 to Net6 · ImageNet, search under 330M FLOPs | Top-1 accuracy (%): 74.59 -> 74.89 | Helped |
| initialize evolutionary search from the candidate pool instead of random paths (ablation) | GreedyNAS · ImageNet, search under 330M FLOPs | Top-1 accuracy (%): +0.18% | Helped |
| continue full training with candidate-pool exploitation instead of stopping using the pool (ablation) | GreedyNAS, Net6 to Net5 · ImageNet, search under 330M FLOPs | Top-1 accuracy (%): 74.89 -> 74.84 | Hurt |
| rank paths using loss on 1K validation images rather than accuracy on the full 50K (ablation) | Greedy supernet · ImageNet, ranking 1000 paths | Spearman rho: 0.997 | Helped |
| use uniform sampling during warm-up before greedy filtering (ablation) | GreedyNAS · ImageNet, path ranking after 10K iterations | Kendall tau: 0.81 | Helped |
| replace uniform supernet training with greedy supernet training (ablation) | Random search · ImageNet, random architecture search | Top-1 accuracy (%): 74.07 -> 74.22 | Helped |
| replace uniform supernet training with greedy supernet training (ablation) | Evolutionary search · ImageNet, evolutionary architecture search | Top-1 accuracy (%): 74.50 -> 74.85 | Helped |
Closest papers by meaning
- GreedyNASv2: Greedier Search With a Greedy Path Filter (CVPR 2022 · similarity 0.94)
- Generalizing Few-Shot NAS with Gradient Matching (ICLR 2022 · similarity 0.93)
- Efficient Few-Shot Neural Architecture Search by Counting the Number of Nonlinear Functions (AAAI 2025 · similarity 0.93)
- Subnet-Aware Dynamic Supernet Training for Neural Architecture Search (CVPR 2025 · similarity 0.92)
- HEP-NAS: Towards Efficient Few-shot Neural Architecture Search via Hierarchical Edge Partitioning (AAAI 2025 · similarity 0.91)
- Boosting Order-Preserving and Transferability for Neural Architecture Search: a Joint Architecture Refined Search and Fine-tuning Approach (CVPR 2024 · similarity 0.90)
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.