Research map / Optimization / Distributed and compressed
Optimal Distributed Learning with Multi-pass Stochastic Gradient Methods
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 |
|---|---|---|---|
| Increase partition level m from 2 to 64 (distributed SGM, step-size 1/(8n)) (ablation) | distributed SGM · Simulated nonparametric regression | generalization error and number of passes to minimal error | Hurt |
| Increase partition level m (more local subsets) while keeping total passes fixed | distributed SGM · Simulated nonparametric regression | total computational cost to reach minimal error | Helped |
| Distributed SGM with partition level m ≤ N^β (β < (2ζ+γ-1)/(2ζ+γ)) and tuned step-size/mini-batch/passes | distributed SGM · Theory, capacity/regularity assumptions (Corollary 2) | convergence rate of generalization error | No clear effect |
| Distributed partitioning with saturation for regularity ζ > 1 | distributed SGM · Theory (Corollary 2) | convergence rate | Hurt |
| Single-machine multi-pass SGM (m=1) with improved analysis vs. prior Lin & Rosasco 2017b | multi-pass SGM · Theory (Theorem 2) | convergence rate exponent/log factor | Helped |
Closest papers by meaning
- Distributed Nyström Kernel Learning with Communications (ICML 2021 · similarity 0.90)
- Effective Distributed Learning with Random Features: Improved Bounds and Algorithms (ICLR 2021 · similarity 0.89)
- Towards Stability and Generalization Bounds in Decentralized Minibatch Stochastic Gradient Descent (AAAI 2024 · similarity 0.89)
- Decentralised Learning with Random Features and Distributed Gradient Descent (ICML 2020 · similarity 0.88)
- Generalization Properties and Implicit Regularization for Multiple Passes SGM (ICML 2016 · similarity 0.88)
- Approximate Gradient Coding for Distributed Learning with Heterogeneous Stragglers (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.