Research map / Optimization / Distributed and compressed
Communication-Computation Efficient Gradient Coding
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 |
|---|---|---|---|
| Code across gradient-vector components as well as data subsets | Gradient coding on Amazon Employee Access · Amazon Employee Access | running time: 32% | Helped |
| Add communication reduction to straggler-focused gradient coding | Gradient coding on Amazon Employee Access · Amazon Employee Access | running time: 23% | Helped |
| Use communication-efficient coded training instead of uncoded training | Logistic regression · Amazon Employee Access | generalization AUC | No clear effect |
| Choose d=4 and m=3 rather than the uncoded setting d=m=1 | Runtime model with n=k=8 · Runtime model, n=k=8 | expected total runtime: 36.1138 -> 21.3697 | Helped |
| Choose d=4 and m=3 rather than the best m=1 coded setting | Runtime model with n=k=8 · Runtime model, n=k=8 | expected total runtime: 24.1063 -> 21.3697 | Helped |
| Use the Vandermonde-based scheme at n=23 | Gradient reconstruction · Gradient reconstruction, n=23 | relative error: up to 80% | Hurt |
| Replace the Vandermonde matrix with a Gaussian random matrix | Proposed gradient coding · Gradient reconstruction, n≤30 | numerical stability | Helped |
| Use a communication reduction factor m>1 rather than m=1 | Gradient coding on Amazon Employee Access · Amazon Employee Access | time to target generalization error | Helped |
| Increase the communication reduction factor m | Runtime-model parameter selection · n=k=10, λ1=0.6, t1=1.5 | optimal communication reduction factor | Mixed |
Closest papers by meaning
- Gradient Coding: Avoiding Stragglers in Distributed Learning (ICML 2017 · similarity 0.92)
- Approximate Gradient Coding for Distributed Learning with Heterogeneous Stragglers (NeurIPS 2025 · similarity 0.92)
- Gradient Coding from Cyclic MDS Codes and Expander Graphs (ICML 2018 · similarity 0.91)
- Sequential Gradient Coding For Straggler Mitigation (ICLR 2023 · similarity 0.91)
- Leveraging partial stragglers within gradient coding (NeurIPS 2024 · similarity 0.90)
- Accelerated Methods with Compressed Communications for Distributed Optimization Problems Under Data Similarity (AAAI 2025 · 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.