Research map / Architectures and efficiency / RNNs and normalization
Adaptive Convolutional ReLUs
Read the paperCheck 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 ConvReLU | DenseNet-40 · Cifar10 | top-1 accuracy: 94.29% -> 94.72% | Helped |
| replace ReLU with ConvReLU | DenseNet-40 · Cifar100 | top-1 accuracy: 75.87% -> 76.41% | Helped |
| replace ReLU with ConvReLU | DenseNet-40 · Tiny ImageNet | top-1 accuracy: 56.31% -> 56.47% | Helped |
| replace ReLU with ConvReLU | VGG-like · MR | top-1 accuracy: 78.51% -> 80.39% | Helped |
| replace ReLU with ConvReLU | VGG-16 · Cifar100 | top-1 accuracy: 73.31% -> 74.04% | Helped |
| replace ReLU with ConvReLU | ResNet-18 · Cifar10 | top-1 accuracy: 94.37% -> 94.48% | Helped |
| use partial replacement instead of all-layer replacement (ablation) | PReLU · Cifar100 | top-1 accuracy: 74.87% -> 75.78% | Helped |
| use partial replacement instead of all-layer replacement (ablation) | LeakyReLU · Cifar100 | top-1 accuracy: 75.29% -> 76.06% | Helped |
| use partial replacement instead of all-layer replacement (ablation) | PReLU · AG's News | top-1 accuracy: 88.52% -> 88.77% | No clear effect |
| use partial replacement instead of all-layer replacement (ablation) | PReLU · MR | top-1 accuracy: 79.83% -> 77.19% | Hurt |
| use zero initialization instead of random initialization of thresholds (ablation) | ConvReLU · Cifar100 | top-1 accuracy: 75.81% -> 76.41% | Helped |
| use zero initialization instead of Glorot initialization (ablation) | ConvReLU · Cifar10 | top-1 accuracy: 94.45% -> 94.72% | Helped |
Closest papers by meaning
- ReLUs Are Sufficient for Learning Implicit Neural Representations (ICML 2024 · similarity 0.90)
- Empowering Convolutional Neural Nets with MetaSin Activation (NeurIPS 2023 · similarity 0.86)
- ReLU Strikes Back: Exploiting Activation Sparsity in Large Language Models (ICLR 2024 · similarity 0.85)
- Activation-Descent Regularization for Input Optimization of ReLU Networks (ICML 2024 · similarity 0.85)
- Compelling ReLU Networks to Exhibit Exponentially Many Linear Regions at Initialization and During Training (ICML 2025 · similarity 0.85)
- Better NTK Conditioning: A Free Lunch from (ReLU) Nonlinear Activation in Wide Neural Networks (NeurIPS 2025 · similarity 0.85)
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.