Research map / Bandits and online learning / Online convex optimization and games
Matrix-Free Preconditioning in Online Learning
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 |
|---|---|---|---|
| Use RecursiveOptimizer instead of diagonal Adagrad | Synthetic data with the optimum along the minimum-eigenvalue direction · Synthetic data, poorly conditioned | test error | Helped |
| Use RecursiveOptimizer instead of diagonal Adagrad | Synthetic data with the optimum along the maximum-eigenvalue direction · Synthetic data, well conditioned | test error | No clear effect |
| Use RecursiveOptimizer instead of Adam and Adagrad | Transformer models · Transformer tasks | test accuracy | Helped |
| Use RecursiveOptimizer instead of Adam and Adagrad | ResNet-32 · CIFAR-10 | test accuracy | Hurt |
| Use RecursiveOptimizer instead of Adam and Adagrad | Logistic regression · MNIST | train loss | No clear effect |
| Replace DiagOptimizer with ScInOL as the inner optimizer (ablation) | RecursiveOptimizer · Synthetic data | performance | No clear effect |
| Replace DiagOptimizer with ScInOL as the inner optimizer (ablation) | RecursiveOptimizer · Deep learning tasks | performance | Helped |
| Add the momentum-like offset (ablation) | RecursiveOptimizer · CIFAR-10 | performance | Helped |
| Add the momentum-like offset (ablation) | RecursiveOptimizer · Transformer tasks | performance | Hurt |
| Clip the inner betting fraction to at most 0.1 until the squared-gradient sum reaches 1 (ablation) | RecursiveOptimizer with ScInOL · Deep learning experiments | performance | Helped |
Closest papers by meaning
- Structured Preconditioners in Adaptive Optimization: A Unified Analysis (ICML 2025 · similarity 0.91)
- Better Full-Matrix Regret via Parameter-Free Online Learning (NeurIPS 2020 · similarity 0.90)
- Universal Sequence Preconditioning (NeurIPS 2025 · similarity 0.88)
- Unconstrained Robust Online Convex Optimization (ICML 2025 · similarity 0.88)
- Dynamic Regret Reduces to Kernelized Static Regret (NeurIPS 2025 · similarity 0.87)
- Online Inverse Linear Optimization: Efficient Logarithmic-Regret Algorithm, Robustness to Suboptimality, and Lower Bound (NeurIPS 2025 · similarity 0.87)
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.