Research map / Optimization / Nonconvex and smooth optimization
On the Provable Suboptimality of Momentum SGD in Nonstationary Stochastic Optimization
Sharan Sahu, Cameron Hogan, Martin T. Wells
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 |
|---|---|---|---|
| increase momentum beta from 0.9 to 0.99 in Heavy-Ball (ablation) | Heavy-Ball · Drifting quadratic, gamma=0.10, sigma2=0.8 | mean tracking error: 40.67 -> 401.37 | Hurt |
| compare SGD at beta 0.9 vs 0.99 (SGD ignores beta) (ablation) | SGD · Drifting quadratic, gamma=0.10, sigma2=0.8 | mean tracking error: 4.29 -> 4.11 | No clear effect |
| use Nesterov instead of Heavy-Ball at large beta (ablation) | Momentum SGD · Drifting quadratic, gamma=0.10, beta=0.99, sigma2=0.8 | mean tracking error: 401.37 -> 38.32 | Improved |
| use momentum (HB) instead of SGD at low beta and small step size (ablation) | Heavy-Ball · Drifting quadratic, gamma=0.01, beta=0.5, sigma2=0.1 | mean tracking error: 1.04 -> 0.34 | Improved |
| use momentum (HB) instead of SGD at high beta, large step size (ablation) | Heavy-Ball · Drifting quadratic, gamma=0.10, beta=0.9, sigma2=0.8 | mean tracking error: 4.29 -> 40.67 | Hurt |
| increase step size from 0.01 to 0.10 for Heavy-Ball at beta=0.99 (ablation) | Heavy-Ball · Drifting quadratic, beta=0.99, sigma2=0.8 | mean tracking error: 38.80 -> 401.37 | Hurt |
| increase gradient noise sigma2 from 0.1 to 0.8 for Heavy-Ball (ablation) | Heavy-Ball · Drifting quadratic, gamma=0.10, beta=0.99 | mean tracking error: 54.25 -> 401.37 | Hurt |
Closest papers by meaning
- On the Hyperparameters in Stochastic Gradient Descent with Momentum (JMLR 2024 · similarity 0.91)
- An Improved Analysis of Stochastic Gradient Descent with Momentum (NeurIPS 2020 · similarity 0.90)
- Gradient correlation is a key ingredient to accelerate SGD with momentum (ICLR 2025 · similarity 0.89)
- Random Scaling and Momentum for Non-smooth Non-convex Optimization (ICML 2024 · similarity 0.89)
- The Marginal Value of Momentum for Small Learning Rate SGD (ICLR 2024 · similarity 0.89)
- Derivatives of Stochastic Gradient Descent in parametric optimization (NeurIPS 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.
