Research map / Architectures and efficiency / Neural architecture search
Efficient Transfer Learning via Joint Adaptation of Network Architecture and Weight
Ming Sun, Haoxuan Dou, Junjie Yan
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 |
|---|---|---|---|
| search the block allocation of ResNet50 on the target task via transfer (to [1,3,7,5]) | Faster-RCNN FPN ResNet50 · MS-COCO minival | mAP: 36.4 -> 37.8 | Improved |
| search the block allocation of ResNet101 on the target task | Faster-RCNN FPN ResNet101 · MS-COCO minival | mAP: 38.5 -> 39.8 | Improved |
| search MobileNetV2 block allocation on the target task | MobileNetV2 · MS-COCO minival | mAP: 32.2 -> 34.0 | Improved |
| search MnasNet-b0 block allocation on the target task | MnasNet-b0 · MS-COCO minival | mAP: 34.2 -> 34.7 | Improved |
| use full-space greedy search with a single longest-path super-network instead of the reduced search space of prior work (CR-ResNet50) | ResNet50 · MS-COCO minival | mAP: 37.4 -> 37.8 | Improved |
| use transferred ResNet50 backbone within Grid-RCNN | Grid-RCNN-ResNet50 · MS-COCO minival | mAP: 39.5 -> 40.3 | Improved |
| use transferred ResNet101 backbone within Cascade-MaskRCNN | Cascade-MaskRCNN-ResNet101 · MS-COCO minival | mAP: 43.3 -> 44.1 | Improved |
| transfer architecture to SE-ResNeXt50 with multi-branching and SE modules | SE-ResNeXt50 · MS-COCO minival | mAP: 38.9 -> 39.9 | Improved |
| increase computation budget of the transferred ResNet from ResNet37 to ResNet53 (ablation) | Transfer-ResNet · MS-COCO minival | mAP: 36.4 -> 38.0 | Improved |
| transfer ResNet50 architecture to [1,6,6,3] on CUB-200-2011 | ResNet50 · CUB-200-2011 | mAP: 84.05 -> 84.98 | Improved |
| transfer architecture with non-local module on CUB-200-2011 | ResNet50+NL · CUB-200-2011 | mAP: 84.79 -> 85.42 | Improved |
| transfer ResNet50 architecture to [1,6,6,3] on CUB-200-2010 | ResNet50 · CUB-200-2010 | mAP: 68.01 -> 71.35 | Improved |
Closest papers by meaning
- Neural Architecture Transfer (TPAMI 2021 · similarity 0.91)
- Boosting Order-Preserving and Transferability for Neural Architecture Search: a Joint Architecture Refined Search and Fine-tuning Approach (CVPR 2024 · similarity 0.89)
- AlphaNet: Improved Training of Supernets with Alpha-Divergence (ICML 2021 · similarity 0.89)
- Distribution Consistent Neural Architecture Search (CVPR 2022 · similarity 0.89)
- Efficient Few-Shot Neural Architecture Search by Counting the Number of Nonlinear Functions (AAAI 2025 · similarity 0.88)
- Subnet-Aware Dynamic Supernet Training for Neural Architecture Search (CVPR 2025 · similarity 0.87)
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.
