Research map / Architectures and efficiency / Neural architecture search
Shapley-NAS: Discovering Operation Contribution for Neural Architecture Search
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 |
|---|---|---|---|
| Replace gradient-based operation ranking with Shapley contribution evaluation and use it during supernet optimization | DARTS search on CIFAR-10 S1 · CIFAR-10 S1 | test error (%): 3.84 -> 2.82 | Helped |
| Replace DARTS operation ranking with Shapley values only at final discretization (ablation) | DARTS search · CIFAR-10 S1 | test error (%): 3.84 -> 3.11 | Helped |
| Keep α frozen and use Shapley values for final operation selection (ablation) | DARTS search · CIFAR-10 S1 | test error (%): 3.84 -> 2.95 | Helped |
| Use Shapley values during supernet optimization rather than only at final discretization (ablation) | DARTS+Shapley on CIFAR-10 S1 · CIFAR-10 S1 | test error (%): 3.11 -> 2.82 | Helped |
| Use Shapley values during supernet optimization rather than only at final discretization (ablation) | DARTS+Shapley on CIFAR-100 S1 · CIFAR-100 S1 | test error (%): 28.21 -> 23.60 | Helped |
| Use Shapley values during supernet optimization rather than only at final discretization (ablation) | DARTS+Shapley on SVHN S1 · SVHN S1 | test error (%): 2.59 -> 2.36 | Helped |
| Set momentum coefficient μ to 0.8 rather than 0.2, 0.5, or 0.9 (ablation) | Shapley-NAS · CIFAR-10 | test error (%): 2.82±0.11, 2.66±0.10, 2.61±0.06 -> 2.47±0.04 | Helped |
| Set step size ε to 0.1 rather than 0.01, 0.05, or 0.5 at μ=0.8 (ablation) | Shapley-NAS · CIFAR-10 | test error (%): 2.67±0.06, 2.55±0.07, 2.71±0.12 -> 2.47±0.04 | Helped |
| Use fewer than 10 Monte-Carlo samples (ablation) | Shapley-NAS · CIFAR-10 | test error (%) | Hurt |
| Use more than 10 Monte-Carlo samples (ablation) | Shapley-NAS · CIFAR-10 | test error (%) | No clear effect |
| Apply early truncation during Monte-Carlo estimation (ablation) | Shapley-NAS · Operation contribution estimation | search cost: nearly half | Helped |
| Use Shapley operation strength instead of DARTS α magnitude to rank operations | NAS-Bench-201 sampled architectures · CIFAR-10, CIFAR-100, ImageNet-16-120 | Kendall Tau correlation with test accuracy: τ = 0.526, 0.474, 0.357 | Helped |
Closest papers by meaning
- Rethinking Architecture Selection in Differentiable NAS (ICLR 2021 · similarity 0.91)
- Interpreting Operation Selection in Differentiable Architecture Search: A Perspective from Influence-Directed Explanations (NeurIPS 2022 · similarity 0.90)
- Zero-Cost Operation Scoring in Differentiable Architecture Search (AAAI 2023 · similarity 0.90)
- Efficient Few-Shot Neural Architecture Search by Counting the Number of Nonlinear Functions (AAAI 2025 · similarity 0.89)
- Boosting Order-Preserving and Transferability for Neural Architecture Search: a Joint Architecture Refined Search and Fine-tuning Approach (CVPR 2024 · similarity 0.88)
- EG-NAS: Neural Architecture Search with Fast Evolutionary Exploration (AAAI 2024 · similarity 0.88)
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.