Research map / Optimization / Distributed and compressed
Coordinating Distributed Example Orders for Provably Accelerated Training
A. Feder Cooper, Wentao Guo, Khiem Pham, Tiancheng Yuan, Charlie F. Ruan, Yucheng Lu, Christopher De Sa
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 |
|---|---|---|---|
| coordinate per-worker permutations with CD-GraB instead of distributed random reshuffling (D-RR) | CD-GraB · Mortgage logistic regression, WikiText-2 LSTM, M4 Weekly MLP | train loss and test metric (accuracy, perplexity, SMAPE) | Improved |
| add server-side coordination of worker orderings versus independent per-worker GraB with RandomizedBalance (ID-GraB Bal) (ablation) | CD-GraB · LeNet on CIFAR-10, m in {4,8,16,32,64} | train loss convergence | Improved |
| add server-side coordination versus independent per-worker GraB with PairBalance (ID-GraB PairBal) (ablation) | CD-GraB · LeNet on CIFAR-10, m in {4,8,16,32,64} | train loss convergence | Improved |
| increase number of workers m without coordination (ID-GraB) (ablation) | ID-GraB (Bal) and ID-GraB (PairBal) · LeNet on CIFAR-10 | train loss convergence versus D-RR | Hurt |
| replace Balance with online PairBalance in centralized GraB (ablation) | GraB · LeNet on CIFAR-10, centralized, B=64 | train loss convergence | No clear effect |
| use CD-GraB (with PairBalance) instead of RR ordering, allowing larger learning rates | CD-GraB · distributed training tasks | training speed and final performance | Improved |
| reduce learning rate for D-RR | D-RR · distributed training tasks | smoothness of curves | Improved |
Closest papers by meaning
- Rescaled Asynchronous SGD: Optimal Distributed Optimization under Data and System Heterogeneity (NeurIPS 2026 · similarity 0.84)
- A Unified Analysis of Stochastic Gradient Descent with Arbitrary Data Permutations and Beyond (NeurIPS 2025 · similarity 0.84)
- Don't Compress Gradients in Random Reshuffling: Compress Gradient Differences (NeurIPS 2024 · similarity 0.83)
- Fault Tolerant ML: Efficient Meta-Aggregation and Synchronous Training (ICML 2024 · similarity 0.83)
- Leveraging partial stragglers within gradient coding (NeurIPS 2024 · similarity 0.83)
- Unveiling the Power of Multiple Gossip Steps: A Stability-Based Generalization Analysis in Decentralized Training (NeurIPS 2025 · similarity 0.83)
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.
