Research map / Optimization / Distributed and compressed
On the Linear Speedup Analysis of Communication Efficient Momentum SGD for Distributed Non-Convex Optimization
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 periodically restarted distributed SGD | PR-SGD-Momentum · Non-convex stochastic optimization | average expected squared gradient norm: O(1/\sqrt{NT}) | Helped |
| Skip communication by synchronizing every I iterations | PR-SGD-Momentum · Identical objectives (κ=0) | communication rounds: O(N^{3/2}T^{1/2}) | Helped |
| Skip communication by synchronizing every I iterations | PR-SGD-Momentum · Non-identical objectives (κ≠0) | communication rounds: O(N^{3/4}T^{3/4}) | Helped |
| Use Nesterov momentum instead of Polyak momentum (ablation) | PR-SGD-Momentum · Non-convex stochastic optimization | average expected squared gradient norm: O(\frac{1}{\gamma T}) + O(\frac{\gamma}{N} \sigma^2) + O(\gamma^2 I \sigma^2)+ O(\gamma^2 I^2 \kappa^2) | No clear effect |
| Set the momentum coefficient to β=0 (ablation) | PR-SGD-Momentum · Non-convex stochastic optimization | average expected squared gradient norm: O(1/\sqrt{NT}) | No clear effect |
| Replace global periodic averaging with decentralized neighborhood aggregation (ablation) | Distributed Momentum SGD · Non-convex stochastic optimization | average expected squared gradient norm: O(\frac{1}{\sqrt{NT}}) + O(\frac{N}{T}) | No clear effect |
| Use bounded variance rather than a bounded second-moment assumption | PR-SGD-Momentum analysis · Non-convex stochastic optimization | assumption strength | Helped |
Closest papers by meaning
- STL-SGD: Speeding Up Local SGD with Stagewise Communication Period (AAAI 2021 · similarity 0.91)
- Communication-Efficient Distributed Blockwise Momentum SGD with Error-Feedback (NeurIPS 2019 · similarity 0.91)
- Communication-efficient SGD: From Local SGD to One-Shot Averaging (NeurIPS 2021 · similarity 0.91)
- 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 0.88)
- SLowcalSGD : Slow Query Points Improve Local-SGD for Stochastic Convex 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.