Research map / Learning theory and algorithms / Prediction and decision losses
Fisher SAM: Information Geometry and Sharpness Aware Minimisation
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 |
|---|---|---|---|
| Replace the Euclidean SAM neighborhood with a Fisher-information ellipsoid | SAM · CIFAR-10 image classification | test accuracy: 97.63^{\pm 0.04} -> 97.89^{\pm 0.07} | Helped |
| Replace the Euclidean SAM neighborhood with a Fisher-information ellipsoid | SAM · CIFAR-100 image classification | test accuracy: 85.60^{\pm 0.05} -> 85.60^{\pm 0.11} | No clear effect |
| Use a Fisher-information ellipsoid for the sharpness-aware neighborhood | DeiT-B extra-training · ImageNet Top-1 | accuracy: 82.03^{\pm 0.04} -> 82.17^{\pm 0.01} | Helped |
| Use a Fisher-information ellipsoid for the sharpness-aware neighborhood | DeiT-B extra-training · ImageNet Top-5 | accuracy: 95.83^{\pm 0.04} -> 95.90^{\pm 0.01} | Helped |
| Use a Fisher-information neighborhood instead of the Euclidean SAM neighborhood | ResNet-32 · CIFAR-10 with label noise 0.2 | test accuracy: 93.12^{\pm 0.24} -> 93.03^{\pm 0.11} | Hurt |
| Use a Fisher-information neighborhood instead of the Euclidean SAM neighborhood | ResNet-32 · CIFAR-10 with label noise 0.4 | test accuracy: 90.54^{\pm 0.19} -> 90.95^{\pm 0.17} | Helped |
| Use a Fisher-information neighborhood instead of the Euclidean SAM neighborhood | ResNet-32 · CIFAR-10 with label noise 0.6 | test accuracy: 85.39^{\pm 0.52} -> 85.76^{\pm 0.21} | Helped |
| Use a Fisher-information neighborhood instead of the Euclidean SAM neighborhood | ResNet-32 · CIFAR-10 with label noise 0.8 | test accuracy: 74.31^{\pm 1.02} -> 74.66^{\pm 0.67} | Helped |
| Increase the Fisher neighborhood size to an excessively large value (gamma=1.0) (ablation) | FSAM on WRN-28-10 · CIFAR-100 hyperparameter sensitivity | test accuracy | Hurt |
Closest papers by meaning
- Sharpness-Aware Minimization: General Analysis and Improved Rates (ICLR 2025 · similarity 0.87)
- Bayesian Optimization with Fisher Information Geometry: Gradient Bounds and Trust-Region Methods (NeurIPS 2026 · similarity 0.87)
- A Universal Class of Sharpness-Aware Minimization Algorithms (ICML 2024 · similarity 0.86)
- Avoiding spurious sharpness minimization broadens applicability of SAM (ICML 2025 · similarity 0.86)
- Fisher information and the geometry of memorization in neural networks (NeurIPS 2026 · similarity 0.86)
- Sharpness-Aware Minimization Enhances Feature Quality via Balanced Learning (ICLR 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.