Research map / Architectures and efficiency / Neural architecture search
Locally Free Weight Sharing for Network Width Search
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 |
|---|---|---|---|
| Use locally free weight sharing with CafeNet-E | ResNet50 · ImageNet, 570M FLOPs | Top-1 accuracy: 72.2% -> 73.3% | Helped |
| Use CafeNet-E to search widths | EfficientNet-B0 · ImageNet, 1× FLOPs | Top-1 accuracy: 76.42% -> 76.83% | Helped |
| Use CafeNet-E to search widths | MobileNetV2 · CIFAR-10, 44M FLOPs | accuracy: 93.00% -> 95.31% | Helped |
| Add min-min optimization rather than randomly optimizing a subnet for each sampled width (ablation) | CafeNet · CIFAR-10, 0.5×-FLOPs MobileNetV2 | accuracy: 0.92% | Helped |
| Use FLOPs-sensitive bins instead of uniform channel groups (ablation) | CafeNet · CIFAR-10, MobileNetV2, 188M FLOPs | accuracy: 95.37% -> 95.56% | Helped |
| Use FLOPs-sensitive bins instead of uniform channel groups (ablation) | CafeNet · CIFAR-10, VGGNet, 189M FLOPs | accuracy: 94.12% -> 94.36% | Helped |
| Increase evolutionary search generations for fixed weight sharing (ablation) | Fixed-pattern MobileNetV2 · CIFAR-10, 0.5× FLOPs | accuracy: 94.53% -> 94.72% | Helped |
| Increase supernet training for fixed weight sharing (ablation) | Fixed-pattern MobileNetV2 · CIFAR-10, 0.5× FLOPs | accuracy: 94.53% -> 94.55% | No clear effect |
| Increase evolutionary search generations for fixed weight sharing (ablation) | Fixed-pattern VGGNet · CIFAR-10, 0.5× FLOPs | accuracy: 93.62% -> 93.69% | Helped |
| Increase supernet training for fixed weight sharing (ablation) | Fixed-pattern VGGNet · CIFAR-10, 0.5× FLOPs | accuracy: 93.62% -> 93.66% | No clear effect |
| Increase the free-channel offset from r=0 to r=3 (ablation) | CafeNet · CIFAR-10, MobileNetV2 | accuracy | Helped |
| Increase the free-channel offset from r=1 to r=3 (ablation) | CafeNet · CIFAR-10, MobileNetV2 | training time per epoch: 61.1s -> 203.6s | Hurt |
Closest papers by meaning
- BCNet: Searching for Network Width With Bilaterally Coupled Network (CVPR 2021 · similarity 0.87)
- Landmark Regularization: Ranking Guided Super-Net Training in Neural Architecture Search (CVPR 2021 · similarity 0.86)
- Distribution Consistent Neural Architecture Search (CVPR 2022 · similarity 0.85)
- Unveiling The Matthew Effect Across Channels: Assessing Layer Width Sufficiency via Weight Norm Variance (NeurIPS 2024 · similarity 0.84)
- Efficient Few-Shot Neural Architecture Search by Counting the Number of Nonlinear Functions (AAAI 2025 · similarity 0.83)
- The Effect of Weight Precision on the Neuron Count in Deep ReLU Networks (ICML 2024 · similarity 0.83)
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.