Research map / Optimization / Optimizers
Cyclades: Conflict-free Asynchronous Machine Learning
Xinghao Pan, Maximilian Lam, Stephen Tu, Dimitris Papailiopoulos, Ce Zhang, Michael I Jordan, Kannan Ramchandran, Christopher Ré
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 |
|---|---|---|---|
| Use Cyclades conflict-free scheduling instead of Hogwild! | SGD matrix completion/word embedding · 18 threads | speedup: 7-9 -> 9-10 | Improved |
| Use Cyclades instead of Hogwild! with diminishing stepsizes | SGD on MovieLens · 18 threads | speedup: 3 -> 6.5 | Improved |
| Use Cyclades instead of Hogwild! for SAGA least squares | SAGA on DBLP · 16 threads | convergence | Improved |
| Use Cyclades instead of Hogwild! for SVRG graph eigenvector | SVRG · 16 threads | convergence/speedup | Improved |
| Cyclades vs Hogwild! stochastic update time per epoch on MovieLens SGD | SGD · MovieLens, 18 threads | time per epoch (s): 0.36 -> 0.32 | Improved |
| Cyclades vs Hogwild! per-epoch time on EN-Wiki word embedding | SGD · EN-Wiki, 1 thread | time per epoch (s): 10.25 -> 9.85 | Improved |
| Cyclades vs Hogwild! per-epoch time on SAGA DBLP | SAGA · DBLP, 18 threads | time per epoch (s): 0.51 -> 0.52 | No clear effect |
| Cyclades vs Hogwild! per-epoch time on SVRG NH2010 | SVRG · NH2010, 18 threads | time per epoch (s): 0.02 -> 0.02 | No clear effect |
| Increase threads from 1 to 18 for Cyclades partitioning cost | Cyclades · Least squares SAGA DBLP | partition cost in Hogwild! epochs: 2.2245 -> 14.1792 | Hurt |
| Run more epochs to amortize partitioning | Cyclades · all tasks | relative overhead | Improved |
| Filter densest 0.016% to 0.048% of features in URL data | Cyclades · URL logistic regression, 16 threads | speedup | Improved |
| Use nearly all features (no filtering) | Cyclades · URL logistic regression | speedup | Hurt |
Closest papers by meaning
- Improved Asynchronous Parallel Optimization Analysis for Stochastic Incremental Methods (JMLR 2018 · similarity 0.91)
- Faster Derivative-Free Stochastic Algorithm for Shared Memory Machines (ICML 2018 · similarity 0.89)
- Rescaled Asynchronous SGD: Optimal Distributed Optimization under Data and System Heterogeneity (NeurIPS 2026 · similarity 0.87)
- Ringmaster ASGD: The First Asynchronous SGD with Optimal Time Complexity (ICML 2025 · similarity 0.87)
- Faster Stochastic Optimization with Arbitrary Delays via Adaptive Asynchronous Mini-Batching (ICML 2025 · similarity 0.86)
- Tight Time Complexities in Parallel Stochastic Optimization with Arbitrary Computation Dynamics (ICLR 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.
