Research map / Architectures and efficiency / Neural architecture search
BCNet: Searching for Network Width With Bilaterally Coupled Network
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 |
|---|---|---|---|
| Use bilaterally coupled evaluation instead of a unilateral supernet (ablation) | Width search on ImageNet · MobileNetV2, ResNet50, EfficientNet-B0, and ProxylessNAS | Top-1 accuracy: 69.64% -> 69.92%; 76.12% -> 76.41%; 74.35% -> 74.63%; 71.13% -> 71.48% | Helped |
| Add complementary training to BCNet (ablation) | Width search on ImageNet · MobileNetV2 and ResNet50, evolutionary search | Top-1 accuracy: 69.92% -> 70.04%; 76.41% -> 76.56% | Helped |
| Use prior initialization instead of random initialization (ablation) | BCNet evolutionary search · ImageNet, MobileNetV2, 0.5× FLOPs | Top-1 accuracy: 69.64% -> 70.20% | Helped |
| Use larger width-search group size (ablation) | BCNet width search · ImageNet MobileNetV2 and CIFAR-10 VGGNet | Top-1 accuracy | Mixed |
| Increase UA training epochs and evolutionary-search iterations (ablation) | UA width search · ImageNet, ResNet50 and MobileNetV2, 0.5× FLOPs | Top-1 accuracy: 75.63% -> 75.65%; 75.74% -> 75.72%; 68.95% -> 68.98%; 69.03% -> 69.06% | No clear effect |
| Increase BCNet training from 30 to 300 epochs (ablation) | MobileNetV2 width search · ImageNet, 50% FLOPs | Top-1 accuracy: 68.51% (30 epochs); 70.20% (300 epochs) | Helped |
| Search network widths with BCNet rather than uniform width scaling | MobileNetV2 · CIFAR-10, 200M FLOPs | accuracy: 94.57% -> 95.44% | Helped |
| Refine the original EfficientNet-B0 widths using BCNet | EfficientNet-B0 · ImageNet, 192M FLOPs | Top-1 accuracy: 74.26% -> 74.92% | Helped |
| Use BCNet searched widths rather than AutoSlim widths | MobileNetV2 · ImageNet, 305M FLOPs, same training recipe | Top-1 accuracy: 73.1% -> 73.9% | Helped |
Closest papers by meaning
- 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 0.87)
- GreedyNAS: Towards Fast One-Shot NAS With Greedy Supernet (CVPR 2020 · similarity 0.86)
- Subnet-Aware Dynamic Supernet Training for Neural Architecture Search (CVPR 2025 · similarity 0.86)
- Efficient Few-Shot Neural Architecture Search by Counting the Number of Nonlinear Functions (AAAI 2025 · similarity 0.85)
- Unveiling The Matthew Effect Across Channels: Assessing Layer Width Sufficiency via Weight Norm Variance (NeurIPS 2024 · similarity 0.84)
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.