Research map / Architectures and efficiency / RNNs and normalization
Learning Specialized Activation Functions With the Piecewise Linear Unit
Yucong Zhou, Zezhou Zhu, Zhao Zhong
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 statistic-based realignment instead of fix-init-10 (channel-wise) (ablation) | PWLU on ResNet-18 · ImageNet | top-1 accuracy: 72.19 -> 72.57 | Improved |
| add statistic-based realignment instead of fix-init-3 (channel-wise) (ablation) | PWLU on MobileNet-V3 · ImageNet | top-1 accuracy: 68.98 -> 69.65 | Improved |
| add statistic-based realignment instead of fix-init-10 (layer-wise) (ablation) | PWLU on ResNet-18 · ImageNet | top-1 accuracy: 71.84 -> 72.22 | Improved |
| add statistic-based realignment instead of fix-init-10 (channel-wise) (ablation) | PWLU on MobileNet-V3 · ImageNet | top-1 accuracy: 68.20 -> 69.65 | Improved |
| use fixed hand-picked boundaries instead of ReLU baseline (layer-wise fix-init-3) (ablation) | PWLU on ResNet-18 · ImageNet | top-1 accuracy: 71.34 -> 71.57 | Improved |
| increase intervals from 4 to 8 (ablation) | PWLU on MobileNet-V3 · ImageNet | top-1 accuracy: 68.70 -> 69.55 | Improved |
| increase intervals from 16 to 20 (ablation) | PWLU on MobileNet-V3 · ImageNet | top-1 accuracy: 69.65 -> 69.37 | Hurt |
| increase intervals from 16 to 20 (ablation) | PWLU on ResNet-18 · ImageNet | top-1 accuracy: 72.57 -> 72.43 | Hurt |
| replace Swish with PWLU | ResNet-18 · ImageNet | top-1 accuracy: 71.61 -> 72.57 | Improved |
| replace Swish with PWLU | MobileNet-V3 · ImageNet | top-1 accuracy: 67.93 -> 69.65 | Improved |
| replace Swish with PWLU | Mask R-CNN ResNet-50-FPN · COCO val2017 | AP bbox: 40.05 -> 40.48 | Improved |
| replace Swish with PWLU | RetinaNet ResNet-50-FPN · COCO val2017 | AP bbox: 37.82 -> 38.31 | Improved |
Closest papers by meaning
- PWLU: Learning Specialized Activation Functions With the Piecewise Linear Unit (TPAMI 2023 · similarity 0.98)
- Smooth Maximum Unit: Smooth Activation Function for Deep Networks Using Smoothing Maximum Technique (CVPR 2022 · similarity 0.88)
- Gompertz Linear Units: Leveraging Asymmetry for Enhanced Learning Dynamics (NeurIPS 2025 · similarity 0.85)
- ReLUs Are Sufficient for Learning Implicit Neural Representations (ICML 2024 · similarity 0.83)
- Better NTK Conditioning: A Free Lunch from (ReLU) Nonlinear Activation in Wide Neural Networks (NeurIPS 2025 · similarity 0.82)
- Compelling ReLU Networks to Exhibit Exponentially Many Linear Regions at Initialization and During Training (ICML 2025 · similarity 0.81)
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.
