Research map / Architectures and efficiency / CNN design
Exploiting Redundancy: Separable Group Convolutional Networks on Lie Groups
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 non-separable SE(2) convolutions with separable convolutions (ablation) | SE(2)-CNN · CIFAR10 | test accuracy (%): 78.50($\pm$0.99) -> 80.43($\pm$0.12) | Helped |
| Replace non-separable translation-dilation convolutions with separable convolutions (ablation) | Translation-dilation CNN · CIFAR100 | test accuracy (%): 44.80 ($\pm 0.56) -> 52.50 ($\pm 0.60) | Helped |
| Replace non-separable SE(2) convolutions with separable convolutions (ablation) | SE(2)-CNN · CIFAR100 | test accuracy (%): 51.67 ($\pm 0.65) -> 50.19 ($\pm 0.40) | Hurt |
| Use Sim(2) equivariance rather than translation equivariance | Sim(2)-CNN versus translation-equivariant baseline · CIFAR10 | test accuracy (%): 81.09($\pm0.11$) -> 89.38($\pm$0.25) | Helped |
| Add continuous-rotation augmentation (ablation) | Separable SE(2)-CNN · Rotated MNIST | test error (%): 0.89{ $\pm.008$} -> 0.66{ $\pm$.023} | Helped |
| Add scale equivariance to rotation equivariance (ablation) | Separable G-CNN · Rotated MNIST | test error (%): 0.89{ $\pm.008$} -> 0.66{ $\pm.009$} | Helped |
| Approximate continuous rotations with random sampling rather than discretisation (ablation) | SE(2)-CNN · Rotated MNIST | test error | Helped |
| Increase rotation sampling resolution beyond 4 elements (ablation) | Non-separable SE(2)-CNN · Rotated MNIST | test error | Hurt |
| Use H-separable rather than jointly separable scale-and-rotation convolutions (ablation) | Sim(2)-CNN · Scaled rotated MNIST | test error | Helped |
| Add scale and rotation equivariance (ablation) | H-separable Sim(2)-CNN · Galaxy10 | test accuracy (%): 86.61($\pm$0.38) -> 87.45 $(\pm 0.31)$ | Helped |
| Increase SIREN hidden size from 6 to 16 (ablation) | Sim(2)-CNN with data augmentation · CIFAR10 | test error (%): 6.47 -> 5.50 | Helped |
| Restrict translation-dilation group convolution support to two neighbouring scale elements (ablation) | Translation-dilation CNN · Scaled MNIST | performance | Helped |
Closest papers by meaning
- Group Equivariant Convolutional Networks (ICML 2016 · similarity 0.91)
- B-Spline CNNs on Lie groups (ICLR 2020 · similarity 0.90)
- Variational Partial Group Convolutions for Input-Aware Partial Equivariance of Rotations and Color-Shifts (ICML 2024 · similarity 0.89)
- Lie Group Decompositions for Equivariant Neural Networks (ICLR 2024 · similarity 0.88)
- Group Downsampling with Equivariant Anti-aliasing (ICLR 2025 · similarity 0.88)
- On the VC dimension of deep group convolutional neural networks (NeurIPS 2025 · 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.