Research map / Architectures and efficiency / CNN design
Kervolutional Neural Networks
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 convolutional layers with polynomial kervolution | ResNet · CIFAR-10 and CIFAR-100 | validation error (%): 13.63 -> 10.85; 44.74 -> 37.12 | Helped |
| Replace convolution with kervolution | GoogLeNet · CIFAR-10+ and CIFAR-100+ | validation error (%): 13.37 -> 5.16; 26.65 -> 20.84 | Helped |
| Replace convolution with kervolution | ResNet · CIFAR-10+ and CIFAR-100+ | validation error (%): 6.43 -> 4.69; 27.22 -> 22.49 | Helped |
| Replace convolution with kervolution | DenseNet-40-12 · CIFAR-10+ and CIFAR-100+ | validation error (%): 5.24 -> 5.08; 24.42 -> 24.92 | Mixed |
| Replace ResNet convolutional layers with polynomial kervolution | ResNet-18, ResNet-34, ResNet-50, ResNet-101 · ImageNet validation | top-1 and top-5 error: top-1: 30.24 / 27.88 -> 29.74 / 27.43; 26.70 / 25.03 -> 26.29 / 24.28; 23.85 / 22.85 -> 23.56 / 22.05; 22.63 / 21.75 -> | Helped |
| Use sigmoid rather than polynomial or Gaussian kervolution (ablation) | LeNet-5 · MNIST validation | training time to 98% accuracy: 2× slower | Hurt |
| Use learnable polynomial kernel parameters (ablation) | LeNet-5 · MNIST | accuracy (%): 99.20% | Helped |
| Arrange polynomial kervolution as kerv-conv rather than conv-kerv (ablation) | LeNet-5 · MNIST validation | validation accuracy and convergence speed | Mixed |
| Use kervolution in both layers (kerv-kerv) (ablation) | LeNet-5 · MNIST validation | convergence speed and validation accuracy | Mixed |
| Remove activation layers and replace max pooling with average pooling (ablation) | CNN · MNIST | accuracy (%): 92.22% | Hurt |
| Remove activation layers and replace max pooling with average pooling (ablation) | Gaussian-polynomial KNN and polynomial-polynomial KNN · MNIST | accuracy (%): 99.11% | Helped |
| Use L2-norm rather than L1-norm kervolution (ablation) | LeNet-5 · MNIST | accuracy (%): 99.05% -> 99.19% | Helped |
Closest papers by meaning
- End-to-End Kernel Learning with Supervised Convolutional Kernel Networks (NeurIPS 2016 · similarity 0.90)
- Approximation and Learning with Deep Convolutional Models: a Kernel Perspective (ICLR 2022 · similarity 0.89)
- Are Large Kernels Better Teachers than Transformers for ConvNets? (ICML 2023 · similarity 0.88)
- More ConvNets in the 2020s: Scaling up Kernels Beyond 51x51 using Sparsity (ICLR 2023 · similarity 0.87)
- KernelWarehouse: Rethinking the Design of Dynamic Convolution (ICML 2024 · similarity 0.87)
- UniRepLKNet: A Universal Perception Large-Kernel ConvNet for Audio Video Point Cloud Time-Series and Image Recognition (CVPR 2024 · 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.