Research map / Optimization / Nonconvex and smooth optimization
On the Influence of Momentum Acceleration on Online Learning
Kun Yuan, Bicheng Ying, Ali H. Sayed
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 |
|---|---|---|---|
| use momentum (beta=0.9) with the same step-size 0.003 instead of plain LMS | LMS · Synthetic LMS, 10-dim | MSD: mu=mu_m=0.003, beta=0.9 | Mixed |
| scale momentum step-size down to mu_m=mu(1-beta)=0.0003 | Momentum LMS · Synthetic LMS, 10-dim | MSD curve: mu_m=0.0003 vs mu=0.003 | Improved |
| add decaying momentum parameter beta(i) | Momentum LMS · Synthetic LMS, 10-dim | convergence and steady-state MSD | Improved |
| scale momentum step-size to mu_m=mu(1-beta)=0.0005 | Momentum SGD · Synthetic logistic regression | MSD curve: mu_m=0.0005 vs mu=0.005 | Improved |
| add momentum (beta=0.9) at equal step-size 0.005 | SGD · Synthetic logistic regression | MSD: mu=mu_m=0.005, beta=0.9 | Mixed |
| add momentum beta=0.9 with mu_m=0.01 vs SGD mu=0.1 | SGD · Adult dataset, logistic regression | test accuracy | No clear effect |
| add momentum beta=0.9 with mu_m=0.01 vs SGD mu=0.1 | SGD · CIFAR-10 (6-layer FC net) | training accuracy: 90% -> 90% | No clear effect |
| add momentum beta=0.9 with mu_m=0.001 vs SGD mu=0.01 | SGD · CIFAR-10 (4-layer CNN) | training accuracy: 94% | No clear effect |
| vary beta over 0.5, 0.6, 0.7, 0.8 instead of 0.9 (ablation) | Momentum vs standard LMS · Synthetic LMS, 10-dim | equivalence behaviour | No clear effect |
Closest papers by meaning
- Random Scaling and Momentum for Non-smooth Non-convex Optimization (ICML 2024 · similarity 0.89)
- Provable and Practical Online Learning Rate Adaptation with Hypergradient Descent (ICML 2025 · similarity 0.89)
- The Marginal Value of Momentum for Small Learning Rate SGD (ICLR 2024 · similarity 0.89)
- On the Performance Analysis of Momentum Method: A Frequency Domain Perspective (ICLR 2025 · similarity 0.88)
- Gradient correlation is a key ingredient to accelerate SGD with momentum (ICLR 2025 · similarity 0.87)
- Gradient-Variation Online Adaptivity for Accelerated Optimization with Hölder Smoothness (NeurIPS 2025 · 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.
