Research map / Learning theory and algorithms / Prediction and decision losses
Robust Generalization Against Photon-Limited Corruptions via Worst-Case Sharpness Minimization
Zhuo Huang, Miaoxi Zhu, Xiaobo Xia, Li Shen, Jun Yu, Chen Gong, Bo Han, Bo Du, Tongliang 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 worst-case sharpness minimization with distribution-aware selection | SharpDRO · CIFAR10, CIFAR100, and ImageNet30; Gaussian and JPEG, severity 5 | accuracy (%): 84.7±0.01 vs 80.5±0.02; 84.5±0.01 vs 81.9±0.02; 53.8±0.03 vs 51.5±0.01; 55.3±0.03 vs 53.1±0.02; 68.7±0.01 vs 66.3±0.01; | Improved |
| Use SharpDRO with OOD-score worst-case selection without distribution labels | SharpDRO · CIFAR10, CIFAR100, and ImageNet30; Gaussian and JPEG, severity 5 | accuracy (%): 84.3±0.01 vs 79.8±0.02; 82.3±0.03 vs 79.1±0.01; 54.1±0.03 vs 52.1±0.02; 54.8±0.02 vs 52.2±0.04; 68.4±0.03 vs 66.5±0.02; | Improved |
| Add worst-case data selection (ablation) | SharpDRO on CIFAR10 Gaussian corruption, severity 5 · CIFAR10 Gaussian | accuracy (%): 75.4 -> 84.7 | Improved |
| Add worst-case sharpness minimization (ablation) | SharpDRO on CIFAR10 Gaussian corruption, severity 5 · CIFAR10 Gaussian | accuracy (%): 78.2 -> 84.7 | Improved |
| Increase perturbation scale rho (ablation) | SharpDRO · CIFAR10 Gaussian corruption; severe versus slight corruption | test accuracy | Mixed |
| Use OOD score based on prediction change under weight perturbation to select hard examples | Distribution-agnostic SharpDRO · CIFAR10 Gaussian corruption, severity 0 to 5 | OOD score | Improved |
Closest papers by meaning
- Towards Robust Out-of-Distribution Generalization Bounds via Sharpness (ICLR 2024 · similarity 0.89)
- Sharpness-Aware Minimization: General Analysis and Improved Rates (ICLR 2025 · similarity 0.86)
- Unveiling m-Sharpness Through the Structure of Stochastic Gradient Noise (NeurIPS 2025 · similarity 0.86)
- Friendly Sharpness-Aware Minimization (CVPR 2024 · similarity 0.86)
- Focal-SAM: Focal Sharpness-Aware Minimization for Long-Tailed Classification (ICML 2025 · similarity 0.86)
- Changing the Training Data Distribution to Reduce Simplicity Bias Improves In-distribution Generalization (NeurIPS 2024 · 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.
