Research map / Architectures and efficiency / Neural architecture search
Posterior-Guided Neural Architecture Search
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 |
|---|---|---|---|
| guide candidate sampling using the learned posterior rather than predefined random dropout (ablation) | PGNAS on SupNet-MI · CIFAR-10 | error (%): 9.75 -> 2.06 | Helped |
| guide candidate sampling using the learned posterior rather than predefined random dropout (ablation) | PGNAS on SupNet-EI · CIFAR-10 | error (%): 13.45 -> 2.56 | Helped |
| sample architecture-weight pairs and use sampled weights directly, without fine-tuning (ablation) | PGNAS on SupNet-MI · CIFAR-10 | error (%): 9.75 -> 2.06 | Helped |
| replace the architecture-dependent L2-like regularizer with conventional constant-weight L2 (ablation) | PGNAS-MI · CIFAR-100 | error (%): 14.28 -> 17.41 | Hurt |
| increase the number of sampled candidates from 0.05K to 0.5k (ablation) | PGNAS-MI · CIFAR-10 | error (%): 2.17 -> 2.06 | Helped |
| increase the number of sampled candidates from 0.5k to 1.5k (ablation) | PGNAS-MI · CIFAR-10 | error (%): 2.06 -> 2.06 | No clear effect |
| increase the number of sampled candidates from 1.5k to 10k (ablation) | PGNAS-MI · CIFAR-10 | error (%): 2.06 -> 1.98 | Helped |
| increase weight prior l² from 50 to 150 (ablation) | PGNAS · CIFAR-10, SupNet-EI | error (%): 2.13 -> 2.06 | Helped |
| increase weight prior l² from 150 to 250 (ablation) | PGNAS · CIFAR-10, SupNet-EI | error (%): 2.06 -> 2.27 | Hurt |
| increase weight prior l² from 250 to 500 (ablation) | PGNAS · CIFAR-10, SupNet-EI | error (%): 2.27 -> 2.39 | Hurt |
| lower Gumbel-softmax temperature from 2/3 to 1/5 (ablation) | PGNAS · CIFAR-10, SupNet-EI | error (%): 2.74 -> 2.56 | Helped |
| lower Gumbel-softmax temperature from 2/3 to 1/5 (ablation) | PGNAS · CIFAR-10, SupNet-M | error (%): 2.49 -> 2.20 | Helped |
Closest papers by meaning
- Improving One-Shot NAS by Suppressing the Posterior Fading (CVPR 2020 · similarity 0.92)
- BayesNAS: A Bayesian Approach for Neural Architecture Search (ICML 2019 · similarity 0.92)
- Neural Architecture Search on ImageNet in Four GPU Hours: A Theoretically Inspired Perspective (ICLR 2021 · similarity 0.91)
- Transfer NAS with Meta-learned Bayesian Surrogates (ICLR 2023 · similarity 0.91)
- Bridging the Gap between Sample-based and One-shot Neural Architecture Search with BONAS (NeurIPS 2020 · similarity 0.91)
- Efficient Few-Shot Neural Architecture Search by Counting the Number of Nonlinear Functions (AAAI 2025 · 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.