Research map / Architectures and efficiency / Neural architecture search
Learning Versatile Neural Architectures by Propagating Network Codes
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 |
|---|---|---|---|
| Jointly optimize classification and segmentation rather than classification alone | NCP architecture search · ImageNet-50-1000 | top1: 85.56 -> 86.05 | Helped |
| Jointly optimize classification and segmentation rather than segmentation alone | NCP architecture search · Cityscapes | mIoU: 77.15 -> 77.16 | No clear effect |
| Jointly optimize segmentation and video recognition | NCP architecture search · Cityscapes and HMDB51 | task performance | Hurt |
| Jointly optimize segmentation and 3D detection | NCP architecture search · Cityscapes and KITTI | task performance: Cityscapes mIoU: 77.15 -> 75.60; KITTI car-3D: 75.44 -> 75.54 | Mixed |
| Initialize segmentation search with a classification architecture instead of searching from scratch | NCP architecture transfer · Cityscapes | mIoU: 77.15 -> 78.27 | Helped |
| Jointly optimize architectures for all three ImageNet subsets | NCP architecture search · ImageNet-50-100 | top1 error: 56.40 -> 58.12 | Helped |
| Jointly optimize using predictors for two Cityscapes input resolutions (ablation) | NCP architecture search · Cityscapes validation, 128 × 512 | mIoU: 65.89 -> 64.98 | Hurt |
| Use continuous propagation instead of winner-takes-all propagation (ablation) | NCP-Net-M · Cityscapes validation | mIoU: 77.91 -> 78.36 | Helped |
| Increase the compute-budget setting from M to L (ablation) | NCP-Net continuous propagation · Cityscapes validation | mIoU: 78.36 -> 80.05 | Helped |
| Optimize video architecture using both scratch and pretrained predictors (ablation) | NCP architecture search · HMDB51, Video-Pretrained | top1: 39.14 -> 40.21 | Helped |
| Optimize with both full- and reduced-resolution segmentation predictors (ablation) | NCP architecture search · ADE20K | mIoU: 33.32 -> 35.65 | Helped |
| Optimize with both full- and reduced-resolution segmentation predictors (ablation) | NCP architecture search · Cityscapes validation, 512 × 1024 | mIoU: 77.15 -> 77.35 | Mixed |
Closest papers by meaning
- TransNAS-Bench-101: Improving Transferability and Generalizability of Cross-Task Neural Architecture Search (CVPR 2021 · similarity 0.91)
- Vision-Oriented Lightweight Neural Architecture Search with Budget-Adaptive Evaluation (CVPR 2026 · similarity 0.90)
- Learning Transferable Architectures for Scalable Image Recognition (CVPR 2018 · similarity 0.89)
- HEP-NAS: Towards Efficient Few-shot Neural Architecture Search via Hierarchical Edge Partitioning (AAAI 2025 · similarity 0.89)
- einspace: Searching for Neural Architectures from Fundamental Operations (NeurIPS 2024 · similarity 0.88)
- Efficient Few-Shot Neural Architecture Search by Counting the Number of Nonlinear Functions (AAAI 2025 · 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.