Research map / Architectures and efficiency / RNNs and normalization
Dynamic ReLU
Yinpeng Chen, Xiyang Dai, Mengchen Liu, Dongdong Chen, Lu Yuan, Zicheng Liu
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 |
|---|---|---|---|
| replace static ReLU with DY-ReLU-B | MobileNetV2 x1.0 · ImageNet val | top-1 accuracy: 72.0 -> 76.2 | Improved |
| replace SE and h-swish with DY-ReLU | MobileNetV3-Small · ImageNet val | top-1 accuracy: 67.4 -> 69.7 | Improved |
| replace ReLU with DY-ReLU | ResNet-10 · ImageNet val | top-1 accuracy: 63.0 -> 66.3 | Improved |
| replace ReLU with DY-ReLU-C in backbone and head | MobileNetV2 x1.0 · COCO keypoint val2017 | AP: 64.6 -> 68.1 | Improved |
| use channel-wise DY-ReLU-B instead of channel-shared DY-ReLU-A (ablation) | DY-ReLU · ImageNet val (MobileNetV2 x0.35) | top-1 accuracy: 63.28 -> 66.36 | Improved |
| add spatial-wise activation (DY-ReLU-C instead of B) for classification (ablation) | DY-ReLU · ImageNet val (MobileNetV2 x0.35) | top-1 accuracy: 66.36 -> 66.31 | No clear effect |
| use DY-ReLU-C instead of DY-ReLU-A in backbone (ablation) | MobileNetV2 x0.5 keypoint · COCO val2017 | AP: 58.97 -> 62.23 | Improved |
| use DY-ReLU-C instead of DY-ReLU-B in the head (ReLU backbone) (ablation) | MobileNetV2 x0.5 keypoint · COCO val2017 | AP: 58.72 -> 61.03 | Improved |
| use DY-ReLU-C in both backbone and head (ablation) | MobileNetV2 x0.5 keypoint · COCO val2017 | AP: 59.26 -> 63.27 | Improved |
| use DY-ReLU-A in the head (ablation) | MobileNetV2 x0.5 keypoint · COCO val2017 | AP: 59.26 -> 57.12 | Hurt |
| make first segment dynamic (slope a_c(x)) (ablation) | DY-ReLU · ImageNet val (MobileNetV2 x0.35) | top-1 accuracy: 60.3 -> 63.8 | Improved |
| add a second dynamic segment (K=2 max) (ablation) | DY-ReLU · ImageNet val (MobileNetV2 x0.35) | top-1 accuracy: 63.8 -> 65.7 | Improved |
Closest papers by meaning
- ReLUs Are Sufficient for Learning Implicit Neural Representations (ICML 2024 · similarity 0.87)
- Neural Characteristic Activation Analysis and Geometric Parameterization for ReLU Networks (NeurIPS 2024 · similarity 0.87)
- Compelling ReLU Networks to Exhibit Exponentially Many Linear Regions at Initialization and During Training (ICML 2025 · similarity 0.86)
- Most ReLU Networks Admit Identifiable Parameters (NeurIPS 2026 · similarity 0.85)
- Make Haste Slowly: A Theory of Emergent Structured Mixed Selectivity in Feature Learning ReLU Networks (ICLR 2025 · similarity 0.85)
- Activation-Descent Regularization for Input Optimization of ReLU 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.
