Research map / Architectures and efficiency / CNN design
Meta-learning Symmetries by Reparameterization
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 |
|---|---|---|---|
| Learn equivariance-inducing parameter sharing with MSR | standard convolution layer · Synthetic rotation equivariance | MSE: .504 -> .004 | Helped |
| Learn rotation and reflection equivariance with MSR | standard convolution layer · Synthetic rotation and flip equivariance | MSE: .507 -> .001 | Helped |
| Use learned parameter sharing rather than the strongest reported baseline | few-shot classification · Aug-Omniglot 5-way 1-shot | accuracy: 94.6\pm 0.4 -> 95.3 \pm 0.3 | Helped |
| Use learned parameter sharing rather than the strongest reported baseline | few-shot classification · Aug-Omniglot 5-way 5-shot | accuracy: \mathbf{97.4\pm 0.2} -> 97.7 \pm 0.2 | Helped |
| Use learned parameter sharing rather than the strongest reported baseline | few-shot classification · Aug-Omniglot 20-way 1-shot | accuracy: \mathbf{85.1\pm 0.3} -> 84.3\pm 0.2 | Hurt |
| Use learned parameter sharing rather than the strongest reported baseline | few-shot classification · Aug-Omniglot 20-way 5-shot | accuracy: \mathbf{94.3\pm 0.2} -> 92.6\pm 0.2 | Hurt |
| Use learned parameter sharing rather than the strongest reported baseline | few-shot classification · Aug-MiniImagenet 5-way 1-shot | accuracy: 44.9\pm 1.2 -> 45.5\pm 1.1 | Helped |
| Use learned parameter sharing rather than the strongest reported baseline | few-shot classification · Aug-MiniImagenet 5-way 5-shot | accuracy: 63.2 \pm 1.0 -> 65.2 \pm 1.0 | Helped |
| Allow symmetry matrix U to update in the inner loop jointly with filter parameters (ablation) | MSR-FC · Synthetic translation symmetry tasks | MSE | Hurt |
| Train the shared symmetry matrix by multi-task learning instead of meta-learning (ablation) | MSR-FC · Synthetic translation symmetry tasks with small training datasets | MSE | Hurt |
| Train the shared symmetry matrix by multi-task learning instead of meta-learning (ablation) | MSR-FC · Synthetic translation symmetry tasks with large training datasets | MSE | No clear effect |
Closest papers by meaning
- A Probabilistic Approach to Learning the Degree of Equivariance in Steerable CNNs (ICML 2024 · similarity 0.91)
- Probing Equivariance and Symmetry Breaking in Convolutional Networks (NeurIPS 2025 · similarity 0.90)
- Learning symmetries via weight-sharing with doubly stochastic tensors (NeurIPS 2024 · similarity 0.90)
- Approximately Equivariant Neural Processes (NeurIPS 2024 · similarity 0.89)
- On Making $SE(2)$-Invariant Networks Optimal (NeurIPS 2026 · similarity 0.89)
- A Tale of Two Symmetries: Exploring the Loss Landscape of Equivariant Models (NeurIPS 2025 · 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.