Research map / Optimization / Nonconvex and smooth optimization
Derandomized Online-to-Non-convex Conversion for Stochastic Weakly Convex Optimization
Fanfan Ji, Xiaotong Yuan
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 periodic momentum restart (T = 50) instead of plain SGDM | SGDM · CIFAR-10 ResNet-101 | test accuracy (%): 93.91 -> 94.81 | Improved |
| use periodic momentum restart (T = 50) instead of plain SGDM | SGDM · CIFAR-10 ViT | test accuracy (%): 84.26 -> 86.22 | Improved |
| use periodic momentum restart (T = 20) instead of plain SGDM | SGDM · CIFAR-10 ResNet-101 | train loss (x10^-3): 2.55 -> 0.55 | Improved |
| use periodic momentum restart (T = 20) instead of plain SGDM | SGDM · CIFAR-10 ViT | train loss (x10^-1): 2.78 -> 2.12 | Improved |
| use restart interval T = 20 versus T = 50 (ablation) | D-O2NC · CIFAR-10 ViT | test accuracy (%): 84.92 vs 86.22 | Mixed |
| use extreme restart frequency T = 2 instead of moderate T (ablation) | D-O2NC · CIFAR-10 | test performance | Hurt |
| increase restart interval T (ablation) | D-O2NC · CIFAR-10 | performance | Mixed |
| use D-O2NC across varied learning rate, momentum and restart settings | SGDM · CIFAR-10 | accuracy | Improved |
Closest papers by meaning
- Random Scaling and Momentum for Non-smooth Non-convex Optimization (ICML 2024 · similarity 0.90)
- General framework for online-to-nonconvex conversion: Schedule-free SGD is also effective for nonconvex optimization (ICML 2025 · similarity 0.88)
- Optimistic Online-to-Batch Conversions for Accelerated Convergence and Universality (NeurIPS 2025 · similarity 0.87)
- Second-Order Convergence in Private Stochastic Non-Convex Optimization (NeurIPS 2025 · similarity 0.86)
- High-Probability Bound for Non-Smooth Non-Convex Stochastic Optimization with Heavy Tails (ICML 2024 · similarity 0.86)
- Single-Loop Stochastic Algorithms for Difference of Max-Structured Weakly Convex Functions (NeurIPS 2024 · 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.
