Research map / Architectures and efficiency / CNN design
Group Equivariant Subsampling
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 MONet component VAEs with p1 V-GAEs | MONet · Multi-dSprites, training sizes 3200, 6400, 12800 | MSE (×10⁻³): 2.661(0.382) -> 0.659(0.103); 1.385(0.235) -> 0.359(0.025); 0.326(0.076) -> 0.264(0.042) | Helped |
| Replace MONet component VAEs with p4 V-GAEs | MONet · Multi-dSprites, training sizes 3200, 6400, 12800 | MSE (×10⁻³): 2.661(0.382) -> 0.563(0.195); 1.385(0.235) -> 0.317(0.060); 0.326(0.076) -> 0.231(0.067) | Helped |
| Replace MONet component VAEs with p1 V-GAEs | MONet · CLEVR6, training sizes 3200, 6400, 12800 | MSE (×10⁻³): 0.673(0.059) -> 0.473(0.064); 0.562(0.057) -> 0.432(0.052); 0.546(0.056) -> 0.388(0.016) | Helped |
| Replace MONet component VAEs with p4 V-GAEs | MONet · CLEVR6, training sizes 3200, 6400, 12800 | MSE (×10⁻³): 0.673(0.059) -> 0.461(0.025); 0.562(0.057) -> 0.414(0.022); 0.546(0.056) -> 0.413(0.018) | Helped |
| Replace MONet component VAEs with p1 V-GAEs | MONet · Multi-dSprites, training sizes 3200, 6400, 12800 | ARI: 0.597(0.022) -> 0.762(0.049); 0.747(0.049) -> 0.823(0.042); 0.891(0.009) -> 0.889(0.013) | Mixed |
| Replace MONet component VAEs with p4 V-GAEs | MONet · Multi-dSprites, training sizes 3200, 6400, 12800 | ARI: 0.597(0.022) -> 0.753(0.089); 0.747(0.049) -> 0.833(0.072); 0.891(0.009) -> 0.902(0.025) | Helped |
| Replace MONet component VAEs with p1 V-GAEs | MONet · CLEVR6, training sizes 3200, 6400, 12800 | ARI: 0.829(0.055) -> 0.921(0.015); 0.878(0.023) -> 0.917(0.032); 0.865(0.033) -> 0.920(0.025) | Helped |
| Replace MONet component VAEs with p4 V-GAEs | MONet · CLEVR6, training sizes 3200, 6400, 12800 | ARI: 0.829(0.055) -> 0.878(0.055); 0.878(0.023) -> 0.914(0.012); 0.865(0.033) -> 0.910(0.011) | Mixed |
| Use p4 rather than p1 equivariance | MONet-GAE · CLEVR6, 12800 | MSE (×10⁻³): 0.388(0.016) -> 0.413(0.018) | Hurt |
| Use p4 rather than p1 equivariance | MONet-GAE · CLEVR6, 3200 | ARI: 0.921(0.015) -> 0.878(0.055) | Hurt |
Closest papers by meaning
- Group Downsampling with Equivariant Anti-aliasing (ICLR 2025 · similarity 0.90)
- Group Equivariant Convolutional Networks (ICML 2016 · similarity 0.90)
- Variational Partial Group Convolutions for Input-Aware Partial Equivariance of Rotations and Color-Shifts (ICML 2024 · similarity 0.89)
- Probing Equivariance and Symmetry Breaking in Convolutional Networks (NeurIPS 2025 · similarity 0.88)
- Equivariant Neural Tangent Kernels (ICML 2025 · similarity 0.87)
- Affine Equivariant Networks Based on Differential Invariants (CVPR 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.