Research map / Optimization / Nonconvex and smooth optimization
Momentum Improves Normalized SGD
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 |
|---|---|---|---|
| Add momentum to normalized SGD | Normalized SGD · Non-convex objectives under assumptions A1–A3 | average expected gradient norm: O\left(\frac{1}{\sqrt{T}} + \frac{\sqrt{\sigma}}{T^{1/4}}\right) | Helped |
| Use implicit gradient transport with an extrapolated gradient point | Normalized SGD with momentum · Second-order smooth non-convex objectives | average expected gradient norm: O(1/T^{2/7}) | Helped |
| Adapt learning rate and momentum to observed gradients | Normalized SGD with momentum · Second-order smooth non-convex objectives | average expected gradient norm: \tilde O\left(\frac{1}{\sqrt{T}} + \frac{\sigma^{4/7}}{T^{2/7}}\right) | Helped |
| Use NIGHT instead of Adam | BERT-Base · BERT pretraining, masked language modeling validation | accuracy: 70.76 vs 70.91 | Helped |
| Use NIGHT instead of SGD with momentum | ResNet-50 · ImageNet, top-1 validation | accuracy: 76.2% and 76.37% | Helped |
| Use practical learning-rate heuristics, including warm-up, polynomial decay, and weight-norm scaling, instead of the adaptive algorithm | NIGHT implementation · BERT pretraining and ImageNet ResNet-50 | performance | Helped |
Closest papers by meaning
- Normalized SGD in the Convex Regime: First High-Probability Guarantees and Momentum Extension (NeurIPS 2026 · similarity 0.91)
- The Marginal Value of Momentum for Small Learning Rate SGD (ICLR 2024 · similarity 0.90)
- Random Scaling and Momentum for Non-smooth Non-convex Optimization (ICML 2024 · similarity 0.87)
- Gradient correlation is a key ingredient to accelerate SGD with momentum (ICLR 2025 · similarity 0.87)
- Ordered Momentum for Asynchronous SGD (NeurIPS 2024 · similarity 0.87)
- Perfect Parallelization in Mini-Batch SGD with Classical Momentum Acceleration (NeurIPS 2026 · similarity 0.87)
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.