Research map / Optimization / Distributed and compressed
EF21-P and Friends: Improved Theoretical Communication Complexity for Distributed Optimization with Bidirectional Compression
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 |
|---|---|---|---|
| Combine EF21-P downlink compression with DIANA uplink compression | Distributed optimization · Strongly convex | communication rounds: $(1 + \omega_{\textnormal{\textnormal{s}}})\frac{L}{\mu} + \frac{\omega_{\textnormal{w}}}{n}\frac{L_{\max}}{\mu} + \omeg | Helped |
| Use EF21-P + DCGD instead of EF21-P + DIANA | Distributed optimization · Strongly convex interpolation regime | communication rounds: $(1 + \omega_{\textnormal{\textnormal{s}}})\frac{L}{\mu} + \frac{\omega_{\textnormal{w}}}{n}\frac{L_{\max}}{\mu}$ | Helped |
| Combine EF21-P with DCGD for bidirectional compression | Distributed optimization · General nonconvex | communication rounds: $\frac{\Delta_0^2 \omega_{\textnormal{\textnormal{w}}} L L_{\max}}{n \varepsilon^2} + \frac{\Delta_0 \omega_{\textnormal | Helped |
| Apply EF21-P + DCGD under strong-growth | EF21-P + DCGD · General nonconvex, strong-growth assumption | communication rounds: $\frac{\Delta_0 D \omega_{\textnormal{w}} L}{n \varepsilon} + \frac{\Delta_0 \omega_{\textnormal{s}} L}{\varepsilon}$ | Helped |
| Use EF21-P + DCGD with stochastic gradients in the homogeneous regime | EF21-P + DCGD · Homogeneous nonconvex | communication terms | Helped |
| Compress server-to-worker model broadcasts with TopK using EF21-P | Logistic regression on real-sim · real-sim | coordinates sent server-to-workers: ×400 less | Helped |
| Use EF21-P + DCGD instead of its competitors | Logistic regression on real-sim · real-sim | convergence | Helped |
| Use EF21-P + DCGD rather than EF21-BC or DASHA | Nonconvex logistic regression with real-sim · real-sim | convergence | Helped |
| Compress server-to-worker updates with EF21-P | EF21-P · Strongly convex | iterations to ε-solution: \nicefrac{1}{\alpha} | Hurt |
| Set β = 0, reducing DIANA to DCGD (ablation) | EF21-P + DIANA · Distributed optimization | convergence rate | Helped |
Closest papers by meaning
- Improving the Worst-Case Bidirectional Communication Complexity for Nonconvex Distributed Optimization under Function Similarity (NeurIPS 2024 · similarity 0.94)
- Scalable Distributed Stochastic Optimization via Bidirectional Compression: Beyond Pessimistic Limits (NeurIPS 2026 · similarity 0.92)
- Towards Faster Decentralized Stochastic Optimization with Communication Compression (ICLR 2025 · similarity 0.91)
- Accelerated Methods with Compressed Communications for Distributed Optimization Problems Under Data Similarity (AAAI 2025 · similarity 0.91)
- Distributed Online Convex Optimization with Compressed Communication: Optimal Regret and Applications (NeurIPS 2026 · similarity 0.90)
- Near-Optimal Distributed Minimax Optimization under the Second-Order Similarity (NeurIPS 2024 · similarity 0.90)
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.