Research map / Architectures and efficiency / Neural architecture search
ISTA-NAS: Efficient and Consistent Neural Architecture Search by Sparse Coding
Read the paperCheck 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 |
|---|---|---|---|
| formulate search as sparse coding with ISTA recovery (two-stage ISTA-NAS) vs DARTS 2nd order | DARTS search space · CIFAR-10 | test error (%): 2.76±0.09 -> 2.54±0.05 | Helped |
| sparse super-net training vs DARTS 1st order | DARTS · CIFAR-10 search | search cost: 0.70 day -> 0.05 day | Helped |
| sparse super-net training vs DARTS 1st order | DARTS · CIFAR-10 search, batch 64 | GPU memory: 9.1 G -> 1.9 G | Helped |
| increase batch size 256 to 512 in two-stage ISTA-NAS (ablation) | ISTA-NAS · CIFAR-10 search | search cost: 0.05 day -> 0.03 day | Helped |
| one-stage search under target-net settings instead of two-stage (ablation) | ISTA-NAS · CIFAR-10 | test error (%): 2.54±0.05 -> 2.36±0.06 | Helped |
| one-stage vs two-stage ISTA-NAS (ablation) | ISTA-NAS · CIFAR-10 | search+eval cost: 0.05 -> 2.3 GPU-day | Hurt |
| ISTA-NAS sparse search vs DARTS/PC-DARTS for rank correlation | DARTS · CIFAR-10 | Kendall tau: -0.36 -> 0.43 | Helped |
| one-stage vs two-stage ISTA-NAS rank correlation (ablation) | ISTA-NAS · CIFAR-10 | Kendall tau: 0.43 -> 0.57 | Helped |
| search directly on ImageNet instead of transfer from CIFAR-10 (ablation) | one-stage ISTA-NAS · ImageNet | top-1 error (%): 25.1 -> 24.0 | Helped |
| one-stage ISTA-NAS vs PC-DARTS on ImageNet | PC-DARTS · ImageNet | top-1 error (%): 24.2 -> 24.0 | Helped |
Closest papers by meaning
- Neural Architecture Search on ImageNet in Four GPU Hours: A Theoretically Inspired Perspective (ICLR 2021 · similarity 0.93)
- AtomNAS: Fine-Grained End-to-End Neural Architecture Search (ICLR 2020 · similarity 0.92)
- Neural Architecture Search as Sparse Supernet (AAAI 2021 · similarity 0.92)
- AdvantageNAS: Efficient Neural Architecture Search with Credit Assignment (AAAI 2021 · similarity 0.92)
- BayesNAS: A Bayesian Approach for Neural Architecture Search (ICML 2019 · similarity 0.91)
- Efficient Few-Shot Neural Architecture Search by Counting the Number of Nonlinear Functions (AAAI 2025 · similarity 0.91)
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.