Research map / Learning theory and algorithms / Kernels and regression theory
Fantastic Generalization Measures and Where to Find Them
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 |
|---|---|---|---|
| Use inverse magnitude-aware sharpness scale instead of the original sharpness bound (ablation) | Sharpness-based generalization prediction · CIFAR-10 | overall Kendall τ: 0.399 -> 0.484 | Helped |
| Use inverse magnitude-aware sharpness scale instead of the original sharpness bound (ablation) | Sharpness-based generalization prediction · CIFAR-10 | Ψ: 0.406 -> 0.516 | Helped |
| Include parameter magnitude in the PAC-Bayes perturbation bound (ablation) | PAC-Bayes generalization prediction · CIFAR-10 | overall Kendall τ | No clear effect |
| Use inverse magnitude-aware sharpness scale as the predictor | Generalization prediction · CIFAR-10 | overall Kendall τ: 0.484 | Helped |
| Use spectral complexity to predict the generalization gap | Generalization prediction · CIFAR-10 | overall Kendall τ: -0.537 | Hurt |
| Use path norm rather than Fisher-Rao metric | Norm-based generalization measures · CIFAR-10 | overall Kendall τ: 0.078 -> 0.373 | Helped |
| Measure final-training gradient variance rather than gradient variance after one epoch (ablation) | Optimization-based generalization prediction · CIFAR-10 | overall Kendall τ: 0.070 -> 0.311 | Helped |
| Measure optimization speed from cross-entropy 0.1 to 0.01 | Generalization prediction · CIFAR-10 | overall Kendall τ: -0.088 | No clear effect |
Closest papers by meaning
- In search of robust measures of generalization (NeurIPS 2020 · similarity 0.92)
- Fantastic Generalization Measures are Nowhere to be Found (ICLR 2024 · similarity 0.88)
- Input Margins Can Predict Generalization Too (AAAI 2024 · similarity 0.87)
- Generalizability of Neural Networks Minimizing Empirical Risk Based on Expressive Power (ICLR 2025 · similarity 0.87)
- Generalization of Scaled Deep ResNets in the Mean-Field Regime (ICLR 2024 · similarity 0.87)
- A Manifold Perspective on the Statistical Generalization of Graph Neural Networks (ICML 2025 · similarity 0.87)
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.