Research map / Optimization / Distributed and compressed
Is Local SGD Better than Minibatch 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 |
|---|---|---|---|
| Use local SGD instead of minibatch SGD on quadratic objectives | SGD · Quadratic objectives | worst-case error upper bound: \frac{HB^2}{R} -> \frac{HB^2}{KR} | Helped |
| Use local AC-SA instead of local SGD on quadratic objectives | Local-SGD · Quadratic objectives | worst-case error upper bound: \frac{HB^2}{KR} -> \frac{HB^2}{K^2R^2} | Helped |
| Use local SGD when K ≳ R in the noise-independent regime | Local-SGD versus Minibatch SGD · General convex objectives (H = B = σ² = 1) | worst-case error: K^{-\frac{1}{3}}R^{-\frac{2}{3}} vs. R^{-1} | Helped |
| Use local SGD when K ≲ R in the noise-independent regime | Local-SGD versus Minibatch SGD · General convex objectives (H = B = σ² = 1) | worst-case error: K^{-\frac{1}{3}}R^{-\frac{2}{3}} vs. R^{-1} | Hurt |
| Use local SGD with small K on the constructed non-quadratic hard instance | Local-SGD versus Minibatch SGD · General convex objectives (H = B = σ² = 1) | worst-case error: K^{-\frac{2}{3}}R^{-\frac{2}{3}} + \prn{MKR}^{-\frac{1}{2}} vs. \Theta\prn*{\frac{1}{R} + \frac{1}{\sqrt{MKR}}} | Hurt |
| Use local SGD with few iterations per round, then increase K | Local-SGD versus Minibatch SGD · Logistic regression task | suboptimality | Mixed |
| Increase M while holding K fixed | Local-SGD · Logistic regression task | suboptimality | No clear effect |
| Increase M while holding K fixed | Thumb-twiddling SGD versus Minibatch SGD · Logistic regression task | suboptimality | Helped |
Closest papers by meaning
- SLowcalSGD : Slow Query Points Improve Local-SGD for Stochastic Convex Optimization (NeurIPS 2024 · similarity 0.89)
- Towards Stability and Generalization Bounds in Decentralized Minibatch Stochastic Gradient Descent (AAAI 2024 · similarity 0.89)
- Improved Stability and Generalization Guarantees of the Decentralized SGD Algorithm (ICML 2024 · similarity 0.88)
- What’s in a Smoothness Constant? Tight Rates for Local SGD with Bounded Second-order Heterogeneity (NeurIPS 2026 · similarity 0.88)
- Do We Need Asynchronous SGD? On the Near-Optimality of Synchronous Methods (NeurIPS 2026 · similarity 0.87)
- Tighter Convergence Bounds for Shuffled SGD via Primal-Dual Perspective (NeurIPS 2024 · 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.