Research map / Optimization / Distributed and compressed
Sharper Convergence Guarantees for Asynchronous SGD for Distributed and Federated Learning
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 |
|---|---|---|---|
| Analyze standard asynchronous SGD using concurrency rather than only maximum delay | Asynchronous SGD with constant stepsize · Homogeneous non-convex L-smooth objectives | iterations to ε-stationary point: \cO\big(\frac{\sigma^2}{\varepsilon^2} + \frac{\tau_{\max}}{\varepsilon} \big) -> \cO\!\left(\frac{\sigma^2}{\varepsilon | Helped |
| Use a delay-adaptive stepsize, reducing it for gradients with large delays | Asynchronous SGD · Homogeneous non-convex L-smooth objectives with constant concurrency | iterations to ε-stationary point: \cO\!\left(\frac{\sigma^2}{\varepsilon^2} + \frac{\tau_{avg}}{\varepsilon}\right) | Helped |
| Assume bounded gradient norm | Asynchronous SGD with constant stepsize · Homogeneous non-convex L-smooth objectives | iterations to ε-stationary point: \cO\!\left(\frac{\sigma^2}{\varepsilon^2} + \frac{\tau_{avg} G}{\varepsilon^{3/2}} + \frac{\tau_{avg}}{\varepsilon}\righ | Helped |
| Use uniform random client sampling with constant concurrency | Asynchronous SGD · Heterogeneous objectives | iterations to ε-stationary point: \cO\!\left(\frac{\sigma^2}{\varepsilon^2} + \frac{\zeta^2}{\varepsilon^2} + \frac{\sqrt{\tau_{avg} \frac{1}{n} \sum_{i = | Helped |
| Use delay-adaptive learning rates in the heterogeneous-client setting | Asynchronous SGD · Heterogeneous objectives | convergence to the intended objective | Hurt |
| Run asynchronous rather than mini-batch SGD with the same concurrency | SGD · Homogeneous objectives, any delay pattern | expected per-iteration time | Helped |
Closest papers by meaning
- Asynchronous Distributed Semi-Stochastic Gradient Optimization (AAAI 2016 · similarity 0.93)
- Rescaled Asynchronous SGD: Optimal Distributed Optimization under Data and System Heterogeneity (NeurIPS 2026 · similarity 0.93)
- Sharper Generalization Guarantees for Asynchronous SGD: Beyond Lipschitzness, Smoothness and Data Homogeneity (ICML 2026 · similarity 0.93)
- Asynchronous Decentralized Parallel Stochastic Gradient Descent (ICML 2018 · similarity 0.93)
- Ringmaster ASGD: The First Asynchronous SGD with Optimal Time Complexity (ICML 2025 · similarity 0.92)
- Stability and Generalization of Asynchronous SGD: Sharper Bounds Beyond Lipschitz and Smoothness (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.