Research map / Architectures and efficiency / CNN design
General E(2)-Equivariant Steerable CNNs
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 convolutions with D8 D4 D1 steerable convolutions (same params) | WRN-28-10 · CIFAR-10 | test error: 3.87 -> 3.13 ± 0.17 | Helped |
| add reflection equivariance (D1 D1 D1) | WRN-28-10 · CIFAR-100 | test error: 18.80 -> 17.97 ± 0.11 | Helped |
| use D8 D4 D1 instead of C8 C4 C1 (add reflections) | equivariant WRN-28-10 · CIFAR-100 | test error: 16.47 ± 0.22 -> 16.76 ± 0.40 | Hurt |
| keep rotation equivariance in last block (D8 D4 D4 vs D8 D4 D1) | equivariant WRN-28-10 · CIFAR-10 | test error: 3.13 ± 0.17 -> 2.91 ± 0.13 | Helped |
| add equivariance on top of AutoAugment | WRN-28-10 AA · CIFAR-10 | test error: 2.6 -> 2.05 ± 0.03 | Helped |
| replace convolutions with D8 D4 D1 steerable convolutions | WRN-16-8 · STL-10 | test error: 12.74±0.23 -> 9.80±0.40 | Helped |
| add group restriction (local equivariance, global level matching data) (ablation) | D16 steerable CNN · MNIST rot | test error: 1.68 ± 0.04 -> 0.68 ± 0.04 | Helped |
| restrict group at later depth instead of early (ablation) | D16 steerable CNN · MNIST rot | test error: 0.79 ± 0.03 -> 0.68 ± 0.04 | Helped |
| use quotient instead of regular representation (ablation) | C16 steerable CNN · MNIST rot | test error: 0.716 ± 0.028 -> 0.705 ± 0.025 | No clear effect |
| use equivariant vs vanilla CNN | CNN · MNIST | convergence speed | Helped |
Closest papers by meaning
- A Program to Build E(N)-Equivariant Steerable CNNs (ICLR 2022 · similarity 0.94)
- A Wigner-Eckart Theorem for Group Equivariant Convolution Kernels (ICLR 2021 · similarity 0.93)
- Implicit Convolutional Kernels for Steerable CNNs (NeurIPS 2023 · similarity 0.92)
- A Probabilistic Approach to Learning the Degree of Equivariance in Steerable CNNs (ICML 2024 · similarity 0.90)
- Variational Partial Group Convolutions for Input-Aware Partial Equivariance of Rotations and Color-Shifts (ICML 2024 · similarity 0.90)
- Clifford-Steerable Convolutional Neural Networks (ICML 2024 · similarity 0.89)
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.