Research map / Detection and segmentation / CNN features
Smooth Maximum Unit: Smooth Activation Function for Deep Networks Using Smoothing Maximum Technique
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 |
|---|---|---|---|
| Replace ReLU with SMU | VGG-16 · MNIST, Fashion MNIST, and SVHN test | Top-1 accuracy (%): 99.53 ± 0.07 -> 99.69 ± 0.04; 93.79 ± 0.15 -> 94.48 ± 0.10; 95.97 ± 0.14 -> 96.59 ± 0.11 | Helped |
| Replace ReLU with SMU | ShuffleNet V2 2.0x · CIFAR100 test | Top-1 accuracy (%): 67.52 ± 0.25 -> 73.74 ± 0.20 | Helped |
| Replace ReLU with SMU-1 | ShuffleNet V2 2.0x · CIFAR100 test | Top-1 accuracy (%): 67.52 ± 0.25 -> 73.45 ± 0.23 | Helped |
| Replace ReLU with SMU | PreActResNet-50 · CIFAR100 test | Top-1 accuracy (%): 73.89 ± 0.23 -> 77.28 ± 0.17 | Helped |
| Replace ReLU with SMU-1 | PreActResNet-50 · CIFAR100 test | Top-1 accuracy (%): 73.89 ± 0.23 -> 76.85 ± 0.20 | Helped |
| Replace ReLU with SMU | SeNet-50 · CIFAR10 test | Top-1 accuracy (%): 94.55 ± 0.19 -> 95.92 ± 0.12 | Helped |
| Replace ReLU with SMU-1 | SeNet-50 · CIFAR10 test | Top-1 accuracy (%): 94.55 ± 0.19 -> 95.22 ± 0.17 | Helped |
| Replace ReLU with SMU | WideResNet 28-10 · Tiny ImageNet test | Top-1 accuracy (%): 63.74 ± 0.40 -> 66.52 ± 0.35 | Helped |
| Replace ReLU with SMU-1 | WideResNet 28-10 · Tiny ImageNet test | Top-1 accuracy (%): 63.74 ± 0.40 -> 65.25 ± 0.37 | Helped |
| Replace ReLU with SMU | SSD 300 with VGG-16 · Pascal VOC 2007 test | mAP: 77.2 ± 0.14 -> 78.2 ± 0.09 | Helped |
| Replace ReLU with SMU | U-Net · CityScapes test | Pixel accuracy: 79.49 ± 0.46 -> 81.79 ± 0.36 | Helped |
| Replace ReLU with SMU | U-Net · CityScapes test | mIOU: 69.31 ± 0.28 -> 71.11 ± 0.30 | Helped |
Closest papers by meaning
- Gompertz Linear Units: Leveraging Asymmetry for Enhanced Learning Dynamics (NeurIPS 2025 · similarity 0.86)
- Better NTK Conditioning: A Free Lunch from (ReLU) Nonlinear Activation in Wide Neural Networks (NeurIPS 2025 · similarity 0.85)
- ReLU Strikes Back: Exploiting Activation Sparsity in Large Language Models (ICLR 2024 · similarity 0.85)
- ReLUs Are Sufficient for Learning Implicit Neural Representations (ICML 2024 · similarity 0.85)
- Neural Characteristic Activation Analysis and Geometric Parameterization for ReLU Networks (NeurIPS 2024 · similarity 0.85)
- Compelling ReLU Networks to Exhibit Exponentially Many Linear Regions at Initialization and During Training (ICML 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.