Research map / Optimization / Distributed and compressed
EF21: A New, Simpler, Theoretically Better, and Practically Faster Error Feedback
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 |
|---|---|---|---|
| Replace standard error feedback with EF21's recursively updated Markov compressor | Distributed optimization with biased compression · Smooth nonconvex problems | convergence rate: O(1/T^{2/3}) -> O(1/T) | Helped |
| Use EF21 with the PL condition | Distributed optimization with biased compression · Polyak-Lojasiewicz functions | convergence rate: (1- γ μ )^T | Helped |
| Replace EF with EF21 | Distributed nonconvex logistic regression · a9a | stepsize tolerance | Helped |
| Replace EF with EF21 | CIFAR-10 image classification · ResNet18 and VGG11 | test accuracy | Helped |
| Use the EF21+ hybrid that selects the lower-distortion compressor | Distributed optimization with biased compression · Nonconvex logistic regression | stepsize tolerance | Helped |
| Increase the stepsize substantially | EF · Nonconvex logistic regression | convergence: 16×--128× | Hurt |
| Increase the stepsize substantially | EF · Least squares, PL setting | convergence: 512×--4096× | Hurt |
| Choose smaller k in the Markov compressor | EF21 · CIFAR-10, ResNet18 | test accuracy | Helped |
| Use EF21 or EF21+ instead of EF with individually tuned k and stepsizes | Distributed nonconvex logistic regression · phishing, mushrooms, a9a, and w8a | communication efficiency | Helped |
Closest papers by meaning
- Error Feedback Reloaded: From Quadratic to Arithmetic Mean of Smoothness Constants (ICLR 2024 · similarity 0.95)
- Tight analyses of first-order methods with error feedback (NeurIPS 2025 · similarity 0.90)
- EControl: Fast Distributed Optimization with Compression and Error Control (ICLR 2024 · similarity 0.89)
- Safe-EF: Error Feedback for Non-smooth Constrained Optimization (ICML 2025 · similarity 0.89)
- Error Feedback under $(L_0,L_1)$-Smoothness: Normalization and Momentum (NeurIPS 2025 · similarity 0.89)
- EFSkip: A New Error Feedback with Linear Speedup for Compressed Federated Learning with Arbitrary Data Heterogeneity (AAAI 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.