Research map / Learning theory and algorithms / Kernels and regression theory
A General Theory of Correct, Incorrect, and Extrinsic Equivariance
Dian Wang, Xupeng Zhu, Jung Yeon Park, Mingxi Jia, Guanang Su, Robert Platt, Robin Walters
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 |
|---|---|---|---|
| Impose D4 invariance instead of using an unconstrained CNN | digit classifier · Printed digits | overall accuracy: 96.62 -> 92.5 | Hurt |
| Impose D4 invariance instead of using an unconstrained CNN | digit classifier · MNIST | overall accuracy: 98.21 -> 96.15 | Hurt |
| Impose D4 invariance | digit classifier · Printed digits, class 6 | accuracy: 94.93 -> 63.08 | Hurt |
| Impose D4 invariance | digit classifier · Printed digits, class 9 | accuracy: 96.13 -> 76.17 | Hurt |
| Impose D4 invariance | digit classifier · MNIST, class 2 | accuracy: 98.62 -> 91.84 | Hurt |
| Impose D4 invariance | digit classifier · MNIST, class 6 | accuracy: 97.99 -> 93.71 | Hurt |
| Use a D1-equivariant policy under extrinsic symmetry | robot sorting policy · Sorting, extrinsic equivariance | test success rate: 99.5± 0.5 % -> 71.5 ± 1.6 % | Hurt |
| Constrain linear classifiers with extrinsic C2 invariance | linear classifier · Extrinsic equivariance example | empirical Rademacher complexity: 1 -> \frac{31}{32} | Hurt |
| Mix in more data with incorrect equivariance | z-invariant network · Swiss Roll | test success rate: 1-0.5i | Hurt |
| Increase the extrinsic data proportion | z-invariant network · Swiss Roll | test success rate | Hurt |
| Replace the unconstrained CNN with D4 invariance | digit classifier · Printed digits, overall | accuracy | Improved |
Closest papers by meaning
- Improving Equivariant Networks with Probabilistic Symmetry Breaking (ICLR 2025 · similarity 0.91)
- On the hardness of learning under symmetries (ICLR 2024 · similarity 0.90)
- Learning (Approximately) Equivariant Networks via Constrained Optimization (NeurIPS 2025 · similarity 0.90)
- On Universality Classes of Equivariant Networks (NeurIPS 2025 · similarity 0.90)
- A Tale of Two Symmetries: Exploring the Loss Landscape of Equivariant Models (NeurIPS 2025 · similarity 0.90)
- A Characterization Theorem for Equivariant Networks with Point-wise Activations (ICLR 2024 · similarity 0.90)
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.
