Research map / Architectures and efficiency / Neural architecture search
MiLeNAS: Efficient Neural Architecture Search via Mixed-Level Reformulation
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 |
|---|---|---|---|
| Use mixed-level optimization with λ=1 instead of single-level or bilevel optimization | gradient-based NAS on CIFAR-10 · CIFAR-10 search | validation accuracy | Helped |
| Increase λ from 0.1 toward 999 (ablation) | MiLeNAS · CIFAR-10 search | training-validation accuracy gap | Mixed |
| Use λ=0, the single-level special case (ablation) | MiLeNAS · CIFAR-10 search | validation accuracy | Hurt |
| Use λ=999, approximating the bilevel special case (ablation) | MiLeNAS · CIFAR-10 search | validation accuracy | Hurt |
| Replace first-order mixed-level updates with second-order mixed-level updates (ablation) | MiLeNAS · CIFAR-10 search | validation accuracy | Hurt |
| Replace GDAS bilevel optimization with mixed-level optimization | GDAS / Gumbel-Softmax search · CIFAR-10 search | validation accuracy: 65.79\% -> 69.56\% | Helped |
| Replace GDAS bilevel optimization with mixed-level optimization | GDAS / Gumbel-Softmax search · CIFAR-10 test | test error (%): 2.82\% -> 2.57\% | Helped |
| Add training-loss information to architecture updates alongside validation loss | MiLeNAS-1st versus DARTS-1st · CIFAR-10 search | validation accuracy | Helped |
| Use first-order updates instead of second-order finite-difference approximation (ablation) | MiLeNAS · CIFAR-10 search | search time | Helped |
| Add model size-based search to evaluate architectures at multiple sizes (ablation) | MiLeNAS · CIFAR-10 architecture search | architectures obtained per search run | Helped |
| Add early stopping when the model size falls below the expected size (ablation) | MiLeNAS · CIFAR-10 search | search time: around 5 hours | Helped |
Closest papers by meaning
- Per-Architecture Training-Free Metric Optimization for Neural Architecture Search (NeurIPS 2025 · similarity 0.92)
- Unifying and Boosting Gradient-Based Training-Free Neural Architecture Search (NeurIPS 2022 · similarity 0.92)
- Neural Architecture Search on ImageNet in Four GPU Hours: A Theoretically Inspired Perspective (ICLR 2021 · similarity 0.91)
- Insights from the Use of Previously Unseen Neural Architecture Search Datasets (CVPR 2024 · similarity 0.90)
- Efficient Few-Shot Neural Architecture Search by Counting the Number of Nonlinear Functions (AAAI 2025 · similarity 0.90)
- EG-NAS: Neural Architecture Search with Fast Evolutionary Exploration (AAAI 2024 · 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.