Research map / Optimization / Implicit bias and generalization
Random Sharpness-Aware Minimization
Yong Liu, Siqi Mai, Minhao Cheng, Xiangning Chen, Cho-Jui Hsieh, Yang You
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 random Gaussian initialization to SAM inner ascent (R-SAM) | SAM / WRN-28-10 · CIFAR-10 | accuracy: 97.3 -> 97.5 | Improved |
| add random initialization to SAM (R-SAM) | SAM / ResNet-18 · CIFAR-10 | accuracy: 96.4±0.2 -> 96.5±0.1 | Improved |
| add random initialization to SAM (R-SAM) | SAM / ResNet-50 · CIFAR-100 | accuracy: 83.3% to 84.0% | Improved |
| add random initialization to SAM (R-SAM) | SAM / WRN-28-10 · CIFAR-100 | accuracy: 84.6% to 85.2% | Improved |
| use SAM instead of SGD+M | SGD+M / WRN-28-10 · CIFAR-10 | accuracy: 96.5% to 97.3% | Improved |
| use SAM instead of AdamW | AdamW / ViT-B-16 · ImageNet-1k | accuracy: 74.7% (AdamW) to 79.8% (SAM) | Improved |
| add random initialization to SAM (R-SAM) | SAM / ViT-B-16 · ImageNet-1k | accuracy: 79.8% (vanilla SAM) to 80.7% | Improved |
| add random initialization to SAM (R-SAM) | SAM / ViT-S-16 · ImageNet-1k | accuracy: 77.9% (vanilla SAM) to 78.7% | Improved |
| add mixup augmentation | AdamW / ViT-B-16 · ImageNet-1k | accuracy: 74.7% to 75.7% | Improved |
| add mixup augmentation | SAM / ViT-B-16 · ImageNet-1k | accuracy: 79.8% to 80.4% | Improved |
| add R-SAM on top of SAM with mixup | SAM / ViT-B-16 · ImageNet-1k | accuracy: 80.4% to 81.1% | Improved |
| add RandAug on top of mixup (ablation) | R-SAM / ViT-B-16 · ImageNet-1k | accuracy: 81.1 -> 81.6 | Improved |
Closest papers by meaning
- Sharpness-Aware Minimization: General Analysis and Improved Rates (ICLR 2025 · similarity 0.92)
- Friendly Sharpness-Aware Minimization (CVPR 2024 · similarity 0.90)
- CR-SAM: Curvature Regularized Sharpness-Aware Minimization (AAAI 2024 · similarity 0.90)
- A Universal Class of Sharpness-Aware Minimization Algorithms (ICML 2024 · similarity 0.90)
- Lookbehind-SAM: k steps back, 1 step forward (ICML 2024 · similarity 0.90)
- Sharpness-Aware Minimization Efficiently Selects Flatter Minima Late In Training (ICLR 2025 · similarity 0.89)
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.
