Research map / Optimization / Implicit bias and generalization
Towards Understanding The Calibration Benefits of Sharpness-Aware Minimization
Chengli Tan, Yubo Zhou, Haishan Ye, Guang Dai, Junmin Liu, Zengjie Song, Jiangshe Zhang, Zixiang Zhao, Yunda Hao, Yong Xu
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 |
|---|---|---|---|
| replace SGD with SAM | ResNet-56 · CIFAR-10 | ECE: six times larger | Improved |
| replace SGD with SAM | ResNet-18 CIFAR-10 · CIFAR-10 | ECE: 5.76 -> 3.24 | Improved |
| add ensembling (size 5) to SAM (ablation) | ResNet-18 · CIFAR-10 | ECE: 3.24 -> 1.09 | Improved |
| add MC-Dropout to SAM (ablation) | ResNet-18 · SVHN OOD AUROC | AUROC: 86.38 -> 83.02 | Hurt |
| add MC-Dropout to SGD (ablation) | ResNet-18 · CIFAR-10 ECE | ECE: 5.76 -> 4.39 | Improved |
| replace SAM with CSAM | ResNet-50 · ImageNet-1K | ECE: 1.52 -> 1.18 | Improved |
| replace AdamW with SAM | ViT-S/32 · ImageNet-1K | test accuracy: 65.03 -> 69.21 | Improved |
| replace AdamW with SAM | ViT-S/16 · ImageNet-1K | ECE: 9.72 -> 1.76 | Improved |
| replace SAM with CSAM | WideResNet-20-10 · CIFAR-10 | ECE: 0.86 -> 0.50 | Improved |
| replace SAM with bSAM | WideResNet-20-10 · CIFAR-10 | ECE: 0.86 -> 1.82 | Hurt |
| add CSAM entropy term gamma=2.0 (vs 0) at rho=0.05 (ablation) | ResNet-56 · CIFAR-100 | ECE: 8.83 -> 3.67 | Improved |
| add CSAM gamma=2.0 at rho=0.1 (ablation) | ResNet-56 · CIFAR-10 | ECE: 0.82 -> 2.04 | Hurt |
Closest papers by meaning
- Sharpness-Aware Minimization: General Analysis and Improved Rates (ICLR 2025 · similarity 0.90)
- Sharpness-Aware Minimization Enhances Feature Quality via Balanced Learning (ICLR 2024 · similarity 0.89)
- A Universal Class of Sharpness-Aware Minimization Algorithms (ICML 2024 · similarity 0.89)
- Sharpness-Aware Minimization Efficiently Selects Flatter Minima Late In Training (ICLR 2025 · similarity 0.89)
- Friendly Sharpness-Aware Minimization (CVPR 2024 · similarity 0.88)
- Self-Calibrating Vicinal Risk Minimisation for Model Calibration (CVPR 2024 · 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.
