Research map / Architectures and efficiency / Neural architecture search
Rethinking Bi-Level Optimization in Neural Architecture Search: A Gibbs Sampling Perspective
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 |
|---|---|---|---|
| Use Gibbs sampling for architecture search instead of DARTS-style deterministic updates | One-shot NAS on CIFAR-10 · CIFAR-10 | test error (%): 3.00±0.14 -> 2.53±0.02 | Helped |
| Use GibbsNAS rather than DARTS-V1 | NAS-Bench-201 search · ImageNet-16-120 | test accuracy (Top-1, %): 16.32±0.00 -> 42.08±1.95 | Helped |
| Use GibbsNAS rather than GDAS | NAS-Bench-201 search · ImageNet-16-120 | test accuracy (Top-1, %): 41.71±0.98 -> 42.08±1.95 | Helped |
| Use GibbsNAS rather than GDAS | NAS-Bench-201 search · CIFAR-10 | test accuracy (Top-1, %): 93.61±0.09 -> 92.72±0.60 | Hurt |
| Use GibbsNAS rather than GDAS | NAS-Bench-201 search · CIFAR-100 | test accuracy (Top-1, %): 70.70±0.30 -> 69.20±1.40 | Hurt |
| Use uncertainty-aware sampling rather than updating with posterior means (ablation) | GibbsNAS · CIFAR-10 | Top-1 test accuracy | Helped |
| Use the second-order GibbsNAS variant rather than the first-order variant (ablation) | GibbsNAS · CIFAR-10 search | ImageNet Top-1 accuracy: 0.743 -> 0.741 | Hurt |
| Use the P-DARTS learning-rate decay policy instead of the DARTS policy (ablation) | GibbsNAS · ImageNet classification | Top-1 accuracy: 0.741 -> 0.754 | Helped |
| Use GibbsNAS rather than DARTS | Transfer from CIFAR-10 to ImageNet · ImageNet classification | Top-1 accuracy: 0.731 -> 0.754 | Helped |
| Use GibbsNAS rather than P-DARTS | Transfer from CIFAR-10 to ImageNet · ImageNet classification | Top-1 accuracy: 0.756 -> 0.754 | Hurt |
Closest papers by meaning
- BaLeNAS: Differentiable Architecture Search via the Bayesian Learning Rule (CVPR 2022 · similarity 0.91)
- EG-NAS: Neural Architecture Search with Fast Evolutionary Exploration (AAAI 2024 · similarity 0.91)
- Bridging the Gap between Sample-based and One-shot Neural Architecture Search with BONAS (NeurIPS 2020 · similarity 0.91)
- iDARTS: Differentiable Architecture Search with Stochastic Implicit Gradients (ICML 2021 · similarity 0.91)
- Rethinking Architecture Selection in Differentiable NAS (ICLR 2021 · similarity 0.90)
- Revisiting Neural Networks for Few-Shot Learning: A Zero-Cost NAS Perspective (ICML 2025 · similarity 0.89)
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.