Research map / Architectures and efficiency / CNN design
Approximate Equivariance via Projection-Based Regularisation
Torben Berndt, Jan Stühmer
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 sample-based equivariance penalty with projection-based regulariser | ACDNet · AAPM CT-MAR | PSNR: 40.02 -> 42.68 | Improved |
| replace sample-based penalty with projection-based regulariser | ACDNet · AAPM CT-MAR | training throughput: 2.54 -> 4.25 | Improved |
| add projection-based regulariser to unregularised network | ACDNet · AAPM CT-MAR | PSNR: 42.08 -> 42.68 | Improved |
| add projection-based regulariser to unregularised network | OSCNet · AAPM CT-MAR | PSNR: 42.36 -> 41.88 | Hurt |
| add projection-based regulariser to unregularised network | DICDNet · AAPM CT-MAR | SSIM: 0.9468 -> 0.9605 | Improved |
| project linear layers onto equivariant subspace at test time instead of training with the regulariser (ablation) | ACDNet · AAPM CT-MAR | PSNR: 42.68 -> 23.63 | Hurt |
| use projection regulariser instead of residual pathway priors (RPP) | ACDNet · AAPM CT-MAR | PSNR: 37.12 -> 42.68 | Improved |
| add projection regulariser to relaxed steerable CNN | RSteer · Smoke plume rotation Domain | MSE: 0.67 -> 0.58 | Improved |
| add projection regulariser to relaxed group conv | RGroup · Smoke plume translation Future | MSE: 0.71 -> 0.72 | No clear effect |
| add projection regulariser | RGroup · Smoke plume scaling Domain | MSE: 0.88 -> 0.88 | No clear effect |
| increase non-equivariance penalty weight lambda_perp (ablation) | approximately SO(2)-invariant MLP · toy SO(2) classification | empirical equivariance defect: 1.10e+01 -> 1.53e-02 | Improved |
| enforce strict SO(3) equivariance instead of partial equivariance | steerable CNN · MedMNIST Organ | test accuracy: 0.920 (CNN) -> 0.607 | Hurt |
Closest papers by meaning
- Approximately Equivariant Neural Processes (NeurIPS 2024 · similarity 0.92)
- Learning (Approximately) Equivariant Networks via Constrained Optimization (NeurIPS 2025 · similarity 0.91)
- Probing Equivariance and Symmetry Breaking in Convolutional Networks (NeurIPS 2025 · similarity 0.90)
- A Tale of Two Symmetries: Exploring the Loss Landscape of Equivariant Models (NeurIPS 2025 · similarity 0.90)
- Improving Equivariant Model Training via Constraint Relaxation (NeurIPS 2024 · similarity 0.89)
- A Canonicalization Perspective on Invariant and Equivariant Learning (NeurIPS 2024 · similarity 0.88)
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.
