Research map / Architectures and efficiency / CNN design
The Heterogeneity Hypothesis: Finding Layer-Wise Differentiated Network Architectures
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 |
|---|---|---|---|
| Adjust layer-wise channel widths using single-shot shrinkage | ResNet50 · ImageNet | Top-1 Error (%): 23.28 -> 23.00 | Helped |
| Adjust layer-wise channel widths using single-shot shrinkage | RegNet X-4.0GF · ImageNet | Top-1 Error (%): 23.05 -> 22.74 | Helped |
| Adjust layer-wise channel widths using single-shot shrinkage | MobileNetV3 small · ImageNet | Top-1 Error (%): 34.91 -> 34.84 | Helped |
| Adjust layer-wise channel widths using single-shot shrinkage | MobileNetV1 · Tiny-ImageNet | Top-1 Error (%): 51.87 -> 46.44 | Helped |
| Adjust layer-wise channel widths using single-shot shrinkage | MobileNetV2 · Tiny-ImageNet | Top-1 Error (%): 44.38 -> 40.74 | Helped |
| Adjust layer-wise channel widths using single-shot shrinkage | EDSR · Urban100 | PSNR [dB]: 26.02 -> 26.09 | Helped |
| Replace the baseline ResNet50 backbone with its LW-DNA version | DiMP · LaSOT | Success (AUC) (%): 55.87 -> 57.43 | Helped |
| Use layer-wise differentiated channels | DnCNN · BSD68 | PSNR [dB]: 24.9 -> 24.9 | No clear effect |
| Set rho to 0.1 and tau to 0.45 instead of rho 0.1 and tau 0.4 (ablation) | MobileNetV1 LW-DNA · Tiny-ImageNet | Top-1 Error: 47.02 -> 46.66 | Helped |
| Set rho to 0.2 and tau to 0.4 instead of rho 0.1 and tau 0.4 (ablation) | MobileNetV1 LW-DNA · Tiny-ImageNet | Top-1 Error: 47.02 -> 46.94 | Helped |
| Set rho to 0.2 and tau to 0.45 instead of rho 0.2 and tau 0.4 (ablation) | MobileNetV1 LW-DNA · Tiny-ImageNet | Top-1 Error: 46.94 -> 46.44 | Helped |
Closest papers by meaning
- Revisiting Convolution Architecture in the Realm of DNA Foundation Models (ICLR 2025 · similarity 0.85)
- LSNet: See Large, Focus Small (CVPR 2025 · similarity 0.84)
- Simplicity Bias and Optimization Threshold in Two-Layer ReLU Networks (ICML 2025 · similarity 0.84)
- D^4: Dataset Distillation via Disentangled Diffusion Model (CVPR 2024 · similarity 0.84)
- Explicitly Modeling Subcortical Vision with a Neuro-Inspired Front-End Improves CNN Robustness (NeurIPS 2025 · similarity 0.84)
- Understanding Heterophily for Graph Neural Networks (ICML 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.