Research map / Optimization / Implicit bias and generalization
Rényi Sharpness: A Novel Sharpness that Strongly Correlates with Generalization
Qiaozhe Zhang, Jun Sun, Yingzhuang Liu
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 |
|---|---|---|---|
| use Renyi sharpness regularizer (RSAM) instead of plain SGD | ResNet20 · CIFAR100 (from scratch) | test accuracy (%): 69.12 -> 70.91 | Improved |
| use RSAM instead of ASAM | ResNet50 · TinyImageNet (from scratch) | test accuracy (%): 62.56 -> 63.33 | Improved |
| use RSAM instead of ASAM | ResNet20 · CIFAR10 (from scratch) | test accuracy (%): 93.62 -> 93.69 | No clear effect |
| use RSAM instead of SAM | WideResNet-28-10 · CIFAR100 (from scratch) | test accuracy (%): 83.55 -> 83.67 | No clear effect |
| use RSAM instead of SGD when fine-tuning pretrained ViT | ViT-B-16 · CIFAR100 (fine-tuning) | test accuracy (%): 88.27 -> 89.58 | Improved |
| use RSAM instead of SAM when fine-tuning ViT | ViT-B-16 · CIFAR10 (fine-tuning) | test accuracy (%): 98.50 -> 98.59 | Improved |
| use order alpha=0.5 for multi-cluster Hessian spectra (ablation) | Renyi sharpness · ResNet18 CIFAR10 correlation | Kendall correlation with generalization gap | Improved |
| use alpha=1.5 (alpha>1) for uniform spectra (ablation) | Renyi sharpness · ResNet18 CIFAR10 correlation | Kendall correlation with generalization gap | Improved |
Closest papers by meaning
- A Universal Class of Sharpness-Aware Minimization Algorithms (ICML 2024 · similarity 0.90)
- Simplicity Bias via Global Convergence of Sharpness Minimization (ICML 2024 · similarity 0.88)
- Sharpness-Aware Minimization: General Analysis and Improved Rates (ICLR 2025 · similarity 0.87)
- Neglected Hessian component explains mysteries in sharpness regularization (NeurIPS 2024 · similarity 0.87)
- Explicit Eigenvalue Regularization Improves Sharpness-Aware Minimization (NeurIPS 2024 · similarity 0.87)
- Adam Reduces a Unique Form of Sharpness: Theoretical Insights Near the Minimizer Manifold (NeurIPS 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.
