Research map / Architectures and efficiency / Neural architecture search
Searching for Network Width With Bilaterally Coupled Network
Xiu Qin Su, Shan You, Jiyang Xie, Fei Wang, Chen Qian, Changshui Zhang, Chang Xu
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 BCNet bilaterally coupled supernet (greedy search) (ablation) | AutoSlim-style supernet · ImageNet, MobileNetV2 150M | Top-1 accuracy: 69.52% -> 69.87% | Improved |
| add BCNet bilaterally coupled supernet (greedy search) (ablation) | AutoSlim-style supernet · ImageNet, ResNet50 2G | Top-1 accuracy: 75.64% -> 76.30% | Improved |
| add complementary training to BCNet (greedy search) (ablation) | BCNet · ImageNet, MobileNetV2 150M | Top-1 accuracy: 69.87% -> 69.91% | Improved |
| add complementary training to BCNet (random-init evolutionary) (ablation) | BCNet · ImageNet, ResNet50 2G | Top-1 accuracy: 76.41% -> 76.56% | Improved |
| add prior initial population sampling to evolutionary search (ablation) | BCNet + complementary training · ImageNet, MobileNetV2 150M | Top-1 accuracy: 70.04% -> 70.20% | Improved |
| add base widths (BCNetV2 shrunk search space) (ablation) | BCNet + complementary training + prior sampling · ImageNet, MobileNetV2 150M | Top-1 accuracy: 70.20% -> 70.71% | Improved |
| add base widths (BCNetV2 shrunk search space) (ablation) | BCNet + complementary training + prior sampling · ImageNet, ResNet50 2G | Top-1 accuracy: 76.90% -> 77.28% | Improved |
| replace AutoSlim supernet with BCNet in supernet ranking | AutoSlim · Channel-Bench-Macro, MobileNet | Kendall tau (%): 60.4 -> 65.2 | Improved |
| replace AutoSlim supernet with BCNet in supernet ranking | AutoSlim · Channel-Bench-Macro, ResNet | Kendall tau (%): 61.2 -> 68.4 | Improved |
| use iterative updating instead of joint update in BCNetV2 (half memory) (ablation) | BCNetV2 · ImageNet, MobileNetV2 150M | Top-1 accuracy: 70.7% -> 70.7% | No clear effect |
| replace uniform 0.5x ResNet50 backbone with BCNet-searched 2G backbone | Uniform ResNet50 · COCO RetinaNet | detection score: 34.3% -> 35.4% | Improved |
| replace uniform ResNet50 backbone with BCNet-searched backbone | Uniform ResNet50 · COCO Faster R-CNN | detection score: 35.4% -> 36.3% | Improved |
Closest papers by meaning
- BCNet: Searching for Network Width With Bilaterally Coupled Network (CVPR 2021 · similarity 0.98)
- AOWS: Adaptive and Optimal Network Width Search With Latency Constraints (CVPR 2020 · similarity 0.89)
- Locally Free Weight Sharing for Network Width Search (ICLR 2021 · similarity 0.88)
- Unveiling The Matthew Effect Across Channels: Assessing Layer Width Sufficiency via Weight Norm Variance (NeurIPS 2024 · similarity 0.85)
- Subnet-Aware Dynamic Supernet Training for Neural Architecture Search (CVPR 2025 · similarity 0.85)
- Efficient Few-Shot Neural Architecture Search by Counting the Number of Nonlinear Functions (AAAI 2025 · 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.
