Research map / Learning theory and algorithms / Kernels and regression theory
Understanding the Generalization Benefit of Model Invariance from a Data Perspective
Sicheng Zhu, Bang An, Furong Huang
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 |
|---|---|---|---|
| add cropping invariance via augmentation | ResNet18 · CIFAR-10 n=100 | test accuracy (%): 41.05 -> 50.47 | Improved |
| add flip invariance via augmentation | ResNet18 · CIFAR-10 n=100 | test accuracy (%): 41.05 -> 44.19 | Improved |
| add rotation invariance via augmentation | ResNet18 · CIFAR-10 n=100 | test accuracy (%): 41.05 -> 47.02 | Improved |
| add cropping invariance via augmentation, full data | ResNet18 · CIFAR-10 n=all | generalization gap: 14.57 -> 7.48 | Improved |
| add 3D-view invariance via augmentation | ResNet18 · ShapeNet n=100 | test accuracy (%): 67.75 -> 78.13 | Improved |
| add 3D-view invariance via augmentation | ResNet18 · ShapeNet n=100 | generalization gap: 32.25 -> 14.94 | Improved |
| add cropping invariance via augmentation | ResNet18 · ShapeNet n=100 | test accuracy (%): 67.75 -> 74.88 | Improved |
| add 3D-view invariance via augmentation, full data | ResNet18 · ShapeNet n=all | test accuracy (%): 91.81 -> 94.38 | Improved |
| add flip invariance via augmentation, full data | ResNet18 · ShapeNet n=all | test accuracy (%): 91.81 -> 92.07 | No clear effect |
| add rotation invariance via augmentation, full data | ResNet18 · ShapeNet n=all | test accuracy (%): 91.81 -> 91.85 | No clear effect |
| increase invariance penalty lambda from 0 to 1 (ablation) | ResNet18, 3D-view regularization · ShapeNet | test accuracy (%): 91.81 -> 94.68 | Improved |
| increase invariance penalty lambda from 0 to 100 (ablation) | ResNet18, 3D-view regularization · ShapeNet | generalization gap: 8.19 -> 1.03 | Improved |
Closest papers by meaning
- On the Strong Correlation Between Model Invariance and Generalization (NeurIPS 2022 · similarity 0.90)
- Invariance Principle Meets Information Bottleneck for Out-of-Distribution Generalization (NeurIPS 2021 · similarity 0.86)
- Automatic Data Augmentation via Invariance-Constrained Learning (ICML 2023 · similarity 0.86)
- Provably Strict Generalisation Benefit for Invariance in Kernel Methods (NeurIPS 2021 · similarity 0.86)
- Learning Invariances using the Marginal Likelihood (NeurIPS 2018 · similarity 0.86)
- Show Me What You Don’t Know: Efficient Sampling from Invariant Sets for Model Validation (NeurIPS 2026 · similarity 0.85)
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.
