Research map / Label-efficient and robust learning / Recognition and long-tail
On the Strong Correlation Between Model Invariance and Generalization
Weijian Deng, Stephen Gould, Liang Zheng
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 |
|---|---|---|---|
| measure invariance with EI instead of JS | invariance measure · ImageNet-R | Pearson r with accuracy: -0.173 -> 0.928 | Improved |
| measure invariance with EI instead of JS | invariance measure · ImageNet-S | Pearson r with accuracy: 0.131 -> 0.915 | Improved |
| measure invariance with EI instead of JS | invariance measure · ImageNet-A | Pearson r with accuracy: -0.297 -> 0.910 | Improved |
| measure invariance with EI instead of JS | invariance measure · ImageNet-Val | Pearson r with accuracy: -0.861 -> 0.929 | Improved |
| use (pre)training with more data instead of heavy augmentation | ResNet-50 · ImageNet-A | accuracy-invariance trend | Improved |
| use l2 distance or accuracy difference as invariance measure instead of EI | invariance measure · OOD ImageNet sets | correlation with accuracy | Hurt |
| use average confidence as invariance score instead of EI | invariance measure · ImageNet | correlation with accuracy | Hurt |
| use prediction consistency only instead of EI | invariance measure · OOD ImageNet sets | correlation with accuracy | Hurt |
| pretrain or train semi-supervised on more data | standard models · ImageNet | rotation and grayscale invariance and accuracy | Improved |
| evaluate on style-shift ImageNet-R vs ImageNet-Val vs ImageNet-V2-A | grayscale EI · ImageNet | correlation: 0.940 vs 0.900 vs 0.868 | Improved |
Closest papers by meaning
- Understanding the Generalization Benefit of Model Invariance from a Data Perspective (NeurIPS 2021 · similarity 0.90)
- Invariance Principle Meets Information Bottleneck for Out-of-Distribution Generalization (NeurIPS 2021 · similarity 0.89)
- Provably Strict Generalisation Benefit for Invariance in Kernel Methods (NeurIPS 2021 · similarity 0.88)
- From Bricks to Bridges: Product of Invariances to Enhance Latent Space Communication (ICLR 2024 · similarity 0.86)
- Diagnosing and Rectifying Fake OOD Invariance: A Restructured Causal Approach (AAAI 2024 · similarity 0.85)
- Aggregation Hides Out-of-Distribution Generalization Failures from Spurious Correlations (NeurIPS 2025 · 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.
