Research map / Architectures and efficiency / CNN design
Group Equivariant Convolutional Networks
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 planar convolutions with p4 group convolutions (with max-pooling over rotations after last layer) (ablation) | rotated MNIST CNN · Rotated MNIST | test error (%): 5.03 -> 2.28 | Helped |
| apply coset max-pooling over rotations in every intermediate layer (rotation-invariant features early) (ablation) | p4 CNN on rotated MNIST · Rotated MNIST | test error (%): 3.21 vs 2.28 | Hurt |
| replace planar convolutions with p4 convolutions (halved filters) (ablation) | All-CNN-C · CIFAR10 | test error (%): 9.44 -> 8.84 | Helped |
| replace planar convolutions with p4m convolutions (filters divided by ~sqrt(8)) (ablation) | All-CNN-C · CIFAR10 | test error (%): 9.44 -> 7.59 | Helped |
| replace planar convolutions with p4m convolutions (ablation) | ResNet44 · CIFAR10 | test error (%): 9.45 -> 6.46 | Helped |
| replace planar convolutions with p4 convolutions (ablation) | All-CNN-C · CIFAR10+ | test error (%): 8.86 -> 7.67 | Helped |
| replace planar convolutions with p4m convolutions (ablation) | All-CNN-C · CIFAR10+ | test error (%): 8.86 -> 7.04 | Helped |
| replace planar convolutions with p4m convolutions (ablation) | ResNet44 · CIFAR10+ | test error (%): 5.61 -> 4.94 | Helped |
| replace planar convolutions with p4m convolutions in a wide 26-layer ResNet (ablation) | wide ResNet26 · CIFAR10+ | test error (%): 5.27 -> 4.19 | Helped |
| G-convolution with p4m group in ResNet44 at roughly fixed parameter count | ResNet44 on CIFAR10 · CIFAR10 | parameter count: 2.64M -> 2.62M | No clear effect |
Closest papers by meaning
- Efficient Equivariant Network (NeurIPS 2021 · similarity 0.92)
- Variational Partial Group Convolutions for Input-Aware Partial Equivariance of Rotations and Color-Shifts (ICML 2024 · similarity 0.92)
- On the VC dimension of deep group convolutional neural networks (NeurIPS 2025 · similarity 0.89)
- Group Downsampling with Equivariant Anti-aliasing (ICLR 2025 · similarity 0.89)
- Learning Color Equivariant Representations (ICLR 2025 · similarity 0.89)
- Continuous Rotation Group Equivariant Network Inspired by Neural Population Coding (AAAI 2024 · similarity 0.87)
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.