Research map / Architectures and efficiency / CNN design
Attentive 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 |
|---|---|---|---|
| add full attentive group convolution | p4-CNN · rot-MNIST test | test error (%): 2.048 ± 0.045 -> 1.696 ± 0.021 | Helped |
| replace attention over rotations only with full group attention (ablation) | p4-CNN · rot-MNIST test | test error (%): 1.980 ± 0.032 -> 1.696 ± 0.021 | Helped |
| add full attentive group convolution instead of widening the baseline | p4-CNN · rot-MNIST test | test error (%): 1.796 ± 0.035 -> 1.696 ± 0.021 | Helped |
| use channel attention only rather than full attention (ablation) | p4-CNN · rot-MNIST test | test error (%): 1.696 ± 0.021 -> 1.825 ± 0.048 | Hurt |
| use spatial attention only rather than full attention (ablation) | p4-CNN · rot-MNIST test | test error (%): 1.696 ± 0.021 -> 1.761 ± 0.027 | Hurt |
| apply attention directly to the input rather than using full attentive convolution (ablation) | p4-CNN · rot-MNIST test | test error (%): 1.696 ± 0.021 -> 1.795 ± 0.028 | Hurt |
| add input attention | p4 All-CNN · CIFAR-10 | test error: 9.32 -> 8.8 | Helped |
| add input attention | p4 All-CNN · CIFAR-10+ | test error: 8.91 -> 7.05 | Helped |
| add input attention | p4m All-CNN · CIFAR-10 | test error: 7.61 -> 6.93 | Helped |
| add input attention | p4m All-CNN · CIFAR-10+ | test error: 7.48 -> 6.53 | Helped |
| add input attention | p4m ResNet44 · CIFAR-10 | test error: 15.72 -> 10.82 | Helped |
| add input attention | p4m ResNet44 · CIFAR-10+ | test error: 15.4 -> 10.12 | Helped |
Closest papers by meaning
- Probing Equivariance and Symmetry Breaking in Convolutional Networks (NeurIPS 2025 · similarity 0.91)
- Group Equivariant Convolutional Networks (ICML 2016 · similarity 0.91)
- Group Equivariant Stand-Alone Self-Attention For Vision (ICLR 2021 · similarity 0.90)
- Variational Partial Group Convolutions for Input-Aware Partial Equivariance of Rotations and Color-Shifts (ICML 2024 · similarity 0.90)
- Group Equivariant Generative Adversarial Networks (ICLR 2021 · similarity 0.89)
- Learning Color Equivariant Representations (ICLR 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.