Research map / Optimization / Nonconvex and smooth optimization
ProxSkip: Yes! Local Gradient Steps Provably Lead to Communication Acceleration! Finally!
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 randomized prox skipping with control variates, choosing p=1/sqrt(kappa) | ProxSkip · Smooth strongly convex objective with expensive prox | prox evaluations: $\mathcal O(\kappa\log\nicefrac{1}{\varepsilon})$ to $\mathcal O(\sqrt{\kappa}\log\nicefrac{1}{\varepsilon})$ | Helped |
| Apply ProxSkip to consensus optimization for federated learning | Scaffnew · Strongly convex federated learning | communication rounds: $\tilde{\mathcal O}(\kappa)$ to $\tilde{\mathcal O}(\sqrt{\kappa})$ | Helped |
| Use local gradient steps without a control variate | LocalGD · Heterogeneous federated learning | convergence | Hurt |
| Set p=1, so prox is not skipped (ablation) | ProxSkip · Smooth strongly convex objective with expensive prox | iterations: $\mathcal O(\kappa\log\nicefrac{1}{\varepsilon})$ | No clear effect |
| Set the expected local steps per communication round to 100 rather than the experimental optimum of 300 (ablation) | Scaffnew · w8a | convergence: $\nicefrac{1}{p} = 100$; $\nicefrac{1}{p} = 300$ | Hurt |
| Set the expected local steps per communication round to 1000 rather than the experimental optimum of 300 (ablation) | Scaffnew · w8a | convergence: $\nicefrac{1}{p} = 1000$; $\nicefrac{1}{p} = 300$ | Hurt |
Closest papers by meaning
- Communication Acceleration of Local Gradient Methods via an Accelerated Primal-Dual Algorithm with an Inexact Prox (NeurIPS 2022 · similarity 0.94)
- Variance Reduced ProxSkip: Algorithm, Theory and Application to Federated Learning (NeurIPS 2022 · similarity 0.91)
- Proximal and Federated Random Reshuffling (ICML 2022 · similarity 0.89)
- On Convergence of FedProx: Local Dissimilarity Invariant Bounds, Non-smoothness and Beyond (NeurIPS 2022 · similarity 0.88)
- Stabilized Proximal-Point Methods for Federated Optimization (NeurIPS 2024 · similarity 0.86)
- Adaptive Proximal Gradient Method for Convex Optimization (NeurIPS 2024 · similarity 0.86)
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.