Research map / Architectures and efficiency / RNNs and normalization
Toward Principled Flexible Scaling for Self-Gated Neural Activation
Sudong Cai, Shuyuan Zheng, Bingzhi Chen, Shuai Yuan, Chuan Xiao, Jianbin Qin, Bing WANG
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 channel effective mean responses to scaling coefficients (ablation) | FleS · ImageNet (Swin-Min) | Top-1 accuracy (%): 69.1 -> 71.4 | Improved |
| scaling coefficients without channel statistics vs GELU (ablation) | FleS-DG · ImageNet (Swin-Min) | Top-1 accuracy (%): 68.7 -> 69.1 | Improved |
| compute channel indicators from positive features only (ablation) | FleS · ImageNet (Swin-Min) | Top-1 accuracy (%): 69.8 -> 71.4 | Improved |
| replace GELU with FleS | Swin-Min · ImageNet | Top-1 accuracy (%): 68.7 -> 71.4 | Improved |
| replace GELU with FleS | PoolFormer-S12 · ImageNet | Top-1 accuracy (%): 77.2 -> 79.4 | Improved |
| replace GELU with FleS | Swin-Micro · ImageNet | Top-1 accuracy (%): 78.7 -> 80.3 | Improved |
| replace GELU with FleS | Swin-T · ImageNet | Top-1 accuracy (%): 81.3 -> 82.3 | Improved |
| replace GELU with FleS | ViT-B/16 · ImageNet | Top-1 accuracy (%): 79.7 -> 80.7 | Improved |
| replace ReLU with FleS | ResNet-50 · ImageNet | Top-1 accuracy (%): 77.2 -> 80.1 | Improved |
| add FleS scaling to AdaShift | AdaS on Swin-Min · ImageNet | Top-1 accuracy (%): 69.7 -> 73.0 | Improved |
| replace GELU with FleS-Proto | Swin-Micro · ImageNet (non-shuffle eval) | Top-1 accuracy (%): 78.7 -> 85.2 | Improved |
| evaluate FleS-Proto on shuffled batches instead of non-shuffled | FleS-Proto Swin-Micro · ImageNet | Top-1 accuracy (%): 85.2 -> 77.3 | Hurt |
Closest papers by meaning
- AdaShift: Learning Discriminative Self-Gated Neural Feature Activation With an Adaptive Shift Factor (CVPR 2024 · similarity 0.88)
- Self-Attention-Based Contextual Modulation Improves Neural System Identification (ICLR 2025 · similarity 0.86)
- Rethinking Neural Nonlinearity as Gating (NeurIPS 2026 · similarity 0.86)
- From Activation to Initialization: Scaling Insights for Optimizing Neural Fields (CVPR 2024 · similarity 0.85)
- Global Convergence and Rich Feature Learning in $L$-Layer Infinite-Width Neural Networks under $\mu$ Parametrization (ICML 2025 · similarity 0.85)
- Feature Learning Dynamics in Infinite-Depth Neural Networks (NeurIPS 2026 · 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.
