Research map / Optimization / Implicit bias and generalization
Fast Mixing of Stochastic Gradient Descent with Normalization and Weight Decay
Zhiyuan Li, Tianhao Wang, Dingli Yu
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 weight decay to SGD on a scale-invariant loss | SGD on scale-invariant loss · limiting diffusion mixing time (theory, Theorem 3.2) | time scale to reach limiting diffusion | Improved |
| use the rescaled time scale ln(1+(e^{4 eta lambda t}-1) eta/(2 lambda))/2 instead of plain T/(eta lambda) | SGD+WD on normalized linear model · Synthetic normalized linear regression, D=10, lambda=0.05 | fit of dynamics to predicted time scale | Improved |
| train longer with SGD+WD after parameter norm has converged, then decay LR | 32-layer PreResNet · CIFAR-10 | test accuracy after LR decay | Improved |
| train longer once train accuracy and parameter norm have plateaued | 32-layer PreResNet · CIFAR-10 | train accuracy | No clear effect |
Closest papers by meaning
- Fast Equilibrium of SGD in Generic Situations (ICLR 2024 · similarity 0.89)
- Fast Last-Iterate Convergence of SGD in the Smooth Interpolation Regime (NeurIPS 2025 · similarity 0.87)
- The Global Convergence Time of Stochastic Gradient Descent in Non-Convex Landscapes: Sharp Estimates via Large Deviations (ICML 2025 · similarity 0.87)
- Emergence of heavy tails in homogenized stochastic gradient descent (NeurIPS 2024 · similarity 0.86)
- Incremental Gradient Descent with Small Epoch Counts is Surprisingly Slow on Ill-Conditioned Problems (ICML 2025 · similarity 0.86)
- High-dimensional Limit of SGD for Diagonal Linear Networks (NeurIPS 2026 · similarity 0.86)
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.
