Research map / Optimization / Optimizers
Accelerating SGD with momentum for over-parameterized learning
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 |
|---|---|---|---|
| Use MaSS instead of the strongest listed baseline at η=0.01 | CNN · CIFAR-10 test | classification accuracy: 83.06% -> 83.97% | Helped |
| Use MaSS instead of the strongest listed baseline at η=0.3 | CNN · CIFAR-10 test | classification accuracy: 82.41% -> 84.48% | Helped |
| Use MaSS instead of the strongest listed baseline at η=0.1 | ResNet-32 · CIFAR-10 test | classification accuracy: 92.60% -> 92.77% | Helped |
| Use MaSS instead of the strongest listed baseline at η=0.3 | ResNet-32 · CIFAR-10 test | classification accuracy: 92.51% -> 92.71% | Helped |
| Increase initial learning rate from 0.01 to 0.3 | SGD+Nesterov on CNN · CIFAR-10 test | classification accuracy: 83.06% -> 75.79% | Hurt |
| Increase initial learning rate from 0.1 to 0.3 | SGD+Nesterov on ResNet-32 · CIFAR-10 test | classification accuracy: 92.60% -> 91.13% | Hurt |
| Use the SGD-optimal step size with Nesterov momentum | SGD+Nesterov · Interpolated linear regression | convergence | Hurt |
| Add a gradient compensation term to Nesterov SGD | MaSS · Interpolated quadratic loss | convergence rate: O(e^{-t/κ_m}) -> O(e^{-t/√(κ_m\tilde{κ}_m)}) | Helped |
| Increase mini-batch size in the linear-scaling regime | MaSS · Interpolated quadratic loss, m < m*_1 | convergence speed: O(e^{-m·t/√(κ₁\tilde{κ})}) | Helped |
| Increase mini-batch size in the diminishing-returns regime | MaSS · Interpolated quadratic loss, m∈[m*_1,m*_2) | convergence speed | Helped |
| Increase mini-batch size beyond m*_2 | MaSS · Interpolated quadratic loss, m≥m*_2 | convergence speed | No clear effect |
Closest papers by meaning
- Gradient correlation is a key ingredient to accelerate SGD with momentum (ICLR 2025 · similarity 0.90)
- EMA-Nesterov: Stabilizing Nesterov's Lookahead for Accelerated Deep Learning Optimization (NeurIPS 2026 · similarity 0.88)
- Nesterov acceleration in benignly non-convex landscapes (ICLR 2025 · similarity 0.87)
- Dimension-adapted Momentum Outscales SGD (NeurIPS 2025 · similarity 0.87)
- Nesterov acceleration despite very noisy gradients (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.