Research map / Architectures and efficiency / Neural architecture search
DARTS-: Robustly Stepping out of Performance Collapse Without Indicators
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 |
|---|---|---|---|
| Add a decaying auxiliary skip connection to DARTS | DARTS · CIFAR-10 S0 | test error (%): 2.91$\pm$0.25 -> 2.63$\pm$0.07 | Helped |
| Add a decaying auxiliary skip connection to DARTS | DARTS · CIFAR-100 S0 | test error (%): 20.58$\pm$0.44 -> 17.51$\pm$0.25 | Helped |
| Add a decaying auxiliary skip connection | DARTS · CIFAR-10 S1-S4 | test error (%): S1: 3.84 -> 2.68; S2: 4.85 -> 2.63; S3: 3.34 -> 2.42; S4: 7.20 -> 2.86 | Helped |
| Add a decaying auxiliary skip connection | DARTS · CIFAR-100 S1-S4 | test error (%): S1: 29.46 -> 22.41; S2: 26.05 -> 21.61; S3: 28.90 -> 21.13; S4: 22.85 -> 21.55 | Mixed |
| Add a decaying auxiliary skip connection | DARTS · ImageNet validation S5 | top-1 accuracy (%): 66.4\% -> 76.2\% | Helped |
| Add DARTS- auxiliary skip to P-DARTS without its priors | P-DARTS · CIFAR-10 S0 | accuracy (%): 96.48$\pm$0.55 -> 97.28$\pm$0.04 | Helped |
| Add DARTS- auxiliary skip to PC-DARTS at K=2 | PC-DARTS · CIFAR-10 S0 | accuracy (%): 97.09$\pm$0.14 -> 97.35$\pm0.02$ | Helped |
| Use cosine or step decay instead of linear decay (ablation) | DARTS- · CIFAR-10 S3 | accuracy (%): 97.33\%$\pm$0.09; 97.30\% | No clear effect |
| Increase the initial auxiliary weight β₀ (ablation) | DARTS- · CIFAR-10 S3 | test error (%): 4.58$\pm$1.3 (β₀=0) -> 2.65$\pm$0.04 (β₀=1) | Helped |
| Replace the auxiliary skip with an identity-initialized learnable 1×1 projection (ablation) | DARTS- · CIFAR-10 S3 | accuracy (%): 97.25\%$\pm$0.09 | Helped |
| Extend search to 200 epochs (ablation) | DARTS- · CIFAR-10 S0 | test error (%): 2.59$\pm$0.08 -> 2.71$\pm$0.11 | Hurt |
| Extend search to 150 or 200 epochs (ablation) | DARTS- · CIFAR-10 S2 and S3 | test error (%): S2 best: 2.63% -> 2.50%; S3 best: 2.42% -> 2.53% | Mixed |
Closest papers by meaning
- $\Lambda$-DARTS: Mitigating Performance Collapse by Harmonizing Operation Selection among Cells (ICLR 2023 · similarity 0.92)
- Stabilizing Differentiable Architecture Search via Perturbation-based Regularization (ICML 2020 · similarity 0.91)
- IS-DARTS: Stabilizing DARTS through Precise Measurement on Candidate Importance (AAAI 2024 · similarity 0.90)
- Understanding and Robustifying Differentiable Architecture Search (ICLR 2020 · similarity 0.89)
- Interpreting Operation Selection in Differentiable Architecture Search: A Perspective from Influence-Directed Explanations (NeurIPS 2022 · similarity 0.89)
- EG-NAS: Neural Architecture Search with Fast Evolutionary Exploration (AAAI 2024 · similarity 0.86)
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.