Research map / Optimization / Implicit bias and generalization
How SGD Selects the Global Minima in Over-parameterized Learning: A Dynamical Stability Perspective
Lei Wu, Chao Ma, Weinan E
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 |
|---|---|---|---|
| decrease batch size from GD (1000) to small SGD batch (ablation) | SGD on CIFAR10 VGG · CIFAR10 non-uniformity, eta=0.01 | non-uniformity: about 350 -> about 100 | Improved |
| increase learning rate (ablation) | GD · FashionMNIST and CIFAR10 | sharpness of found minimum: 53.5 -> 1.9 (FashionMNIST, eta 0.01 -> 1) | Improved |
| use SGD with small batch instead of GD (ablation) | SGD · FashionMNIST and CIFAR10 | sharpness | Improved |
| switch from GD eta=0.1 to SGD eta=0.1 B=4 | corrupted FashionMNIST FNN · Corrupted FashionMNIST | test accuracy: 71.44 -> 72.42 | Improved |
| switch from GD eta=0.1 to SGD eta=0.1 B=4 | FashionMNIST FNN · FashionMNIST | test accuracy: 80.04 -> 80.40 | Improved |
| use batch size 100 instead of 4 (SGD eta=0.1) (ablation) | SGD · Corrupted FashionMNIST | escape from minimum: 70.90 vs 72.42 test acc | Hurt |
| run GD/SGD from an L-BFGS minimum | L-BFGS solution · Corrupted FashionMNIST | test accuracy: 69.5 -> 73.25 | Improved |
| flatter minima vs sharper | minima selected · Clean FashionMNIST | test accuracy correlation with sharpness | Mixed |
Closest papers by meaning
- Characterizing Dynamical Stability of Stochastic Gradient Descent in Overparameterized Learning (JMLR 2025 · similarity 0.93)
- Stability and Generalization of Learning Algorithms that Converge to Global Optima (ICML 2018 · similarity 0.91)
- Global Convergence and Stability of Stochastic Gradient Descent (NeurIPS 2022 · similarity 0.91)
- A Unified Stability Analysis of SAM vs SGD: Role of Data Coherence and Emergence of Simplicity Bias (NeurIPS 2025 · similarity 0.89)
- Towards Stability and Generalization Bounds in Decentralized Minibatch Stochastic Gradient Descent (AAAI 2024 · similarity 0.89)
- A Precise Characterization of SGD Stability Using Loss Surface Geometry (ICLR 2024 · similarity 0.88)
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.
