Research map / Architectures and efficiency / Neural architecture search
DSNAS: Direct Neural Architecture Search Without Parameter Retraining
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 |
|---|---|---|---|
| jointly optimize architecture and weights in one stage, without retraining | DSNAS on ImageNet · ImageNet classification | total GPU hours: reducing the total time by more than 34% | Helped |
| use one-stage DSNAS instead of SPOSsearch120retrain240 | ImageNet classification · ImageNet validation | Top-1 accuracy: comparable accuracy | No clear effect |
| use DSNAS rather than SPOSsearch120tune120 | ImageNet classification · ImageNet validation | Top-1 accuracy: 1.4% | Helped |
| use DSNAS rather than SPOSsearch240retrain240 | ImageNet classification · ImageNet validation | Top-1 accuracy: 4% | Helped |
| add progressive early stopping instead of selecting an architecture with SPOS evolution at epoch 120 (ablation) | DSNASsearch240 · ImageNet validation | Top-1 accuracy: 1.3% | Helped |
| replace the continuous all-candidate SNAS computation with sampled-subnetwork DSNAS computation | DSNAS · CIFAR-10 | forward time per batch: 0.26s -> 0.05s | Helped |
| replace ProxylessNAS with DSNAS's compact-subnetwork implementation | DSNAS · ImageNet classification | forward time per batch: 3.3s -> 1.9s | Helped |
| use two-stage SPOS or ProxylessNAS, whose search and evaluation performance correlate weakly | two-stage NAS · ImageNet architecture search | Kendall Tau correlation: 0.33; -0.07; -0.33 | Hurt |
Closest papers by meaning
- TransNAS-Bench-101: Improving Transferability and Generalizability of Cross-Task Neural Architecture Search (CVPR 2021 · similarity 0.90)
- Neural Architecture Search on ImageNet in Four GPU Hours: A Theoretically Inspired Perspective (ICLR 2021 · similarity 0.90)
- ProxylessNAS: Direct Neural Architecture Search on Target Task and Hardware (ICLR 2019 · similarity 0.90)
- Posterior-Guided Neural Architecture Search (AAAI 2020 · similarity 0.90)
- Training-free Neural Architecture Search through Variance of Knowledge of Deep Network Weights (CVPR 2025 · 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.