Research map / Architectures and efficiency / Neural architecture search
InTrain: Intrinsic Trainability for Zero-Cost Neural Architecture Search
Qinqin Zhou, Fuhai Chen, Jipeng Wu, Zhiwei Chen, Zhikai Hu, Weiwei Cai
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 gradient health to participation-ratio score (multiplicative) (ablation) | InTrain · NAS-Bench-201 (CIFAR-10) | Kendall tau: 0.61 -> 0.67 | Improved |
| add participation-ratio capacity to gradient-health score (multiplicative) (ablation) | InTrain · NAS-Bench-201 (CIFAR-10) | Kendall tau: 0.63 -> 0.67 | Improved |
| couple components multiplicatively instead of additively (ablation) | InTrain · NAS-Bench-201 (CIFAR-10) | Kendall tau: 0.59 -> 0.67 | Improved |
| combine components additively instead of using one alone (ablation) | InTrain · NAS-Bench-201 (CIFAR-10) | Kendall tau: 0.63 -> 0.59 | Hurt |
| couple components multiplicatively instead of additively (ablation) | InTrain · NAS-Bench-201 (CIFAR-10) | Spearman rho: 0.80 -> 0.86 | Improved |
| use Gaussian instead of uniform synthetic inputs (ablation) | InTrain · NAS-Bench-201 (CIFAR-10) | Kendall tau | Improved |
| use constant all-ones inputs (ablation) | InTrain · NAS-Bench-201 (CIFAR-10) | Kendall tau | Hurt |
| vary input resolution 32 to 128 (ablation) | InTrain · NAS-Bench-201 (CIFAR-10) | Kendall tau | No clear effect |
| use InTrain-NAS instead of AZ-NAS | InTrain · ImageNet MobileNetV2 450M | top-1 accuracy (%): 78.6 -> 78.9 | Improved |
| use InTrain-NAS instead of AZ-NAS | InTrain · ImageNet MobileNetV2 1000M | top-1 accuracy (%): 81.1 -> 81.3 | Improved |
| use InTrain-NAS instead of AZ-NAS | InTrain · ImageNet MobileNetV2 600M | top-1 accuracy (%): 79.9 -> 79.9 | No clear effect |
| use InTrain instead of ensemble AZ-NAS | InTrain · NAS-Bench-201 CIFAR-10 | Kendall tau: 0.712 -> 0.669 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 2 on most shared results.
- ZiCo — ZiCo: Zero-shot NAS via inverse Coefficient of Variation on Gradients (ICLR 2023) · better on 7 of 7 shared results
- AZ-NAS — AZ-NAS: Assembling Zero-Cost Proxies for Network Architecture Search (CVPR 2024) · better on 3 of 6 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Understanding and Accelerating Neural Architecture Search With Training-Free and Theory-Grounded Metrics (TPAMI 2023 · similarity 0.87)
- Unifying and Boosting Gradient-Based Training-Free Neural Architecture Search (NeurIPS 2022 · similarity 0.87)
- Topology-Weighted Effective Rank: A Zero-Cost Proxy for Training Dynamics Stability in Neural Architecture Search (ECCV 2026 · similarity 0.86)
- Neural Spectral Capacity: An Architectural Quantity from Network Specification Alone (NeurIPS 2026 · similarity 0.86)
- The Computational Complexity of Circuit Discovery for Inner Interpretability (ICLR 2025 · similarity 0.85)
- einspace: Searching for Neural Architectures from Fundamental Operations (NeurIPS 2024 · similarity 0.85)
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.
