Research map / Architectures and efficiency / RNNs and normalization
Padé Activation Units: End-to-end Learning of Flexible Activation Functions in Deep Networks
Alejandro Molina, Patrick Schramowski, Kristian Kersting
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 fixed activations with learnable PAUs | VGG-8 on MNIST · MNIST | test accuracy: 99.17 ±0.10 -> 99.30 ±0.05 | No clear effect |
| replace fixed activations with learnable PAUs | LeNet on Fashion-MNIST · Fashion-MNIST | test accuracy: 89.86 ±0.32 -> 90.33 ±0.15 | Improved |
| add coefficient noise during training (ablation) | PAU on MNIST · MNIST, LeNet | test accuracy: 99.21 \pm0.04 -> 99.26 \pm0.11 | Improved |
| add coefficient noise during training (ablation) | PAU on Fashion-MNIST · Fashion-MNIST, LeNet | test accuracy: 90.33 \pm0.15 -> 90.20 \pm0.11 | Hurt |
| add coefficient noise during training (ablation) | PAU on CIFAR-10 · CIFAR-10, MobileNetV2 | test accuracy: 94.57 \pm0.21 -> 94.82 \pm0.21 | Improved |
| add coefficient noise during training (ablation) | PAU on CIFAR-10 · CIFAR-10, ResNet101 | test accuracy: 95.16 \pm0.13 -> 95.34 \pm0.13 | Improved |
| add coefficient noise during training (ablation) | PAU on CIFAR-10 · CIFAR-10, DenseNet121 | test accuracy: 95.03 \pm0.07 -> 95.27 \pm0.10 | Improved |
| replace fixed activations with learnable PAUs | MobileNetV2 · ImageNet | top-1 accuracy: 71.24 -> 71.35 | Improved |
| replace fixed activations with learnable PAUs | MobileNetV2 · ImageNet | top-5 accuracy: 89.95 -> 89.85 | Hurt |
| prune convolutional layers of PAU networks (ablation) | ResNet101 with PAU · CIFAR-10 | accuracy | No clear effect |
| prune the PAU ResNet101 subnetwork by 30% (ablation) | ResNet101 with PAU · CIFAR-10 | accuracy | Improved |
| replace fixed activations with learnable activations | VGG-8 and MobileNetV2 · CIFAR-10 | test accuracy | Improved |
Closest papers by meaning
- Deep Network Approximation: Beyond ReLU to Diverse Activation Functions (JMLR 2024 · similarity 0.87)
- Entropy-based Activation Function Optimization: A Method on Searching Better Activation Functions (ICLR 2025 · similarity 0.85)
- Smooth Maximum Unit: Smooth Activation Function for Deep Networks Using Smoothing Maximum Technique (CVPR 2022 · similarity 0.85)
- Better NTK Conditioning: A Free Lunch from (ReLU) Nonlinear Activation in Wide Neural Networks (NeurIPS 2025 · similarity 0.85)
- Trainable Highly-expressive Activation Functions (ECCV 2024 · similarity 0.85)
- Gompertz Linear Units: Leveraging Asymmetry for Enhanced Learning Dynamics (NeurIPS 2025 · 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.
