Research map / Optimization / Optimizers
SySCD: A System-Aware Parallel Coordinate Descent Algorithm
Nikolas Ioannou, Celestine Mendler-Dünner, Thomas Parnell
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 |
|---|---|---|---|
| add cache-line-aligned bucketed coordinate access (ablation) | SySCD on Criteo-Kaggle · Criteo-Kaggle | training time: 63% | Improved |
| increase bucket size to eight (ablation) | SySCD · Criteo-Kaggle | training time: eight | Improved |
| add dynamic re-partitioning instead of static partitioning (ablation) | SySCD · Criteo-Kaggle | epochs to convergence: 54% | Improved |
| add dynamic re-partitioning instead of static partitioning (ablation) | SySCD · HIGGS | epochs to convergence | No clear effect |
| add hierarchical NUMA-aware optimization (ablation) | SySCD on a 4-node system · Criteo-Kaggle | training time: 33% | Improved |
| replace the NUMA-oblivious implementation with SySCD's full system-aware method | parallel coordinate descent · Criteo-Kaggle, HIGGS, epsilon | training time: ×5.4 (P9); ×4.8 (X86_64) | Improved |
| use SySCD multi-threading rather than the best alternative GLM solver | GLM solvers · Criteo-Kaggle and HIGGS | training time: over ×10 | Improved |
| use SySCD multi-threading rather than alternative solvers | GLM solvers · epsilon, 4-node and 2-node systems | training time: ×7.3 faster on the 4 node system; ×41.9 faster on the 2-node system | Improved |
Closest papers by meaning
- Fast Asynchronous Parallel Stochastic Gradient Descent: A Lock-Free Approach with Convergence Guarantee (AAAI 2016 · similarity 0.89)
- Asynchronous Decentralized Parallel Stochastic Gradient Descent (ICML 2018 · similarity 0.88)
- Asynchronous Parallel Greedy Coordinate Descent (NeurIPS 2016 · similarity 0.88)
- STL-SGD: Speeding Up Local SGD with Stagewise Communication Period (AAAI 2021 · similarity 0.87)
- Ringmaster ASGD: The First Asynchronous SGD with Optimal Time Complexity (ICML 2025 · similarity 0.86)
- Rescaled Asynchronous SGD: Optimal Distributed Optimization under Data and System Heterogeneity (NeurIPS 2026 · similarity 0.85)
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.
