Research map / Optimization / Implicit bias and generalization
Sharpness-aware minimization theory
Analyzes why and how sharpness-aware minimization works and proposes improved variants.
37 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2022–2026) · an approach in Implicit bias and generalization, keyword cluster “sgd · sharpness · noise”
State History Tried Checklist Preprints Who Export
Where it stands
Not enough benchmark data Fewer than four papers report the same benchmark, so there is no record to follow.
12 papers in 2023–24 and 19 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic grew, from 5.1% to 7.3%.
How this is decided
A benchmark is tracked when at least four papers of this approach report their own method on the same dataset and metric in at least two different years. Its record is the best number reported up to 2024 compared with the best reported in 2025–26: still improving means at least 1% better, flat means at most 0.2% better with at least two papers from 2025–26 trying. Numbers on a different scale from the rest (a fraction among percentages) and far outliers are left out. Results come from different papers with different setups, so they are signals, not head-to-head comparisons. The share of papers is counted at ICML, NeurIPS, ICLR, CVPR and AAAI.
How it got here
Most cited paper of each year
- 2022 · Efficient Sharpness-aware Minimization for Improved Training of Neural Networks (ICLR 2022 · 25 citations)
- 2023 · Gradient Norm Aware Minimization Seeks First-Order Flatness and Improves Generalization (CVPR 2023 · 45 citations)
- 2024 · Friendly Sharpness-Aware Minimization (CVPR 2024 · 22 citations)
- 2025 · Sharpness-Aware Minimization: General Analysis and Improved Rates (ICLR 2025 · 1 citation)
- 2026 · Fix the Loss, Not the Radius: Rethinking the Adversarial Perturbation of Sharpness-Aware Minimization (ICML 2026 · 0 citations)
Earliest papers here
- Efficient Sharpness-aware Minimization for Improved Training of Neural Networks (ICLR 2022 · 25 citations)
- Make Sharpness-Aware Minimization Stronger: A Sparsified Perturbation Approach (NeurIPS 2022 · 17 citations)
- Surrogate Gap Minimization Improves Sharpness-Aware Training (ICLR 2022 · 13 citations)
What has been tried
20 papers here report 208 measured changes (a component added, removed or swapped, compared with a baseline): helped 69%, no clear effect 8%, hurt 16%, mixed 6%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| replace SAM perturbation direction with the full-gradient component (SAM-full) | CIFAR-100, ResNet-18 · test accuracy (%) | 80.88 -> 78.02 | Hurt | Friendly Sharpness-Aware Minimization (CVPR 2024 · 22 citations) |
| Use λ=1−1/t instead of λ=0.0 | CIFAR-10 test, ρ=0.2 · test accuracy (%) | 95.8 -> 95.98 | Helped | Sharpness-Aware Minimization: General Analysis and Improved Rates (ICLR 2025 · 1 citation) |
| Normalize estimated gradient norms before the exponential weighting | Hyperparameter sensitivity · performance sensitivity to λ | Helped | Unveiling m-Sharpness Through the Structure of Stochastic Gradient Noise (NeurIPS 2025 · 0 citations) | |
| switch from SGD to SAM for the final 1–5 epochs | CIFAR-10 · test error | Helped | Sharpness-Aware Minimization Efficiently Selects Flatter Minima Late In Training (ICLR 2025 · 0 citations) | |
| Add the importance-weighting effect | Toy spiral feature · hard feature probe error | Helped | Sharpness-Aware Minimization Enhances Feature Quality via Balanced Learning (ICLR 2024 · 0 citations) | |
| use stochastic gradients for the outer perturbation instead of full-batch gradients | Training loss · trace of Hessian decrease | d times larger gradient | Helped | How to Escape Sharp Minima with Random Perturbations (ICML 2024 · 0 citations) |
What reviewers will expect
Datasets and baselines named in the 7 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- CIFAR-10 · 3 of 7 papers
- CIFAR-100 · 3 of 7 papers
Baselines
- SGD · 5 of 7 papers
- SAM · 4 of 7 papers
- USAM · 2 of 7 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 31 papers; from OpenAlex and the papers' affiliations).
- Google · 3 papers
- Massachusetts Institute of Technology · 3 papers
- Tsinghua University · 2 papers
Use it
Download BibTeX of 31 key papersOpen the topic in the interactive map
Get alerts. Describe your idea in scime atlas and save it (free scime account): you get a weekly email when new preprints land close to it.
Related approaches
Same keyword cluster
- SGD noise and flat minima (112 papers)
- Edge-of-stability dynamics (116 papers)
- Stability-based generalization analysis (88 papers)
Elsewhere in the topic
- Shallow ReLU network convergence (132 papers)
- Neural tangent kernel analysis (114 papers)
- Gradient flow and mean-field dynamics (77 papers)
- Low-rank and single-index recovery (70 papers)
- Explicit and implicit regularization interplay (63 papers)
- Implicit regularization of SGD (55 papers)
- Implicit bias in linear networks (53 papers)
- In-context learning as gradient descent (41 papers)
Benchmark numbers, ablations and baseline numbers are extracted by AI from each paper's full text and can be wrong; the paper is the authority. Data: results index of 2026-11-02, citation counts from OpenAlex (2026-10-09).