Research map / Optimization / Optimizers
Romberg-Extrapolated Zeroth-Order Gradient Estimator: Higher-Order Bias Reduction with Preserved Leading Directional Variance
Hongcheng Dong, Wenqiang Pu, Licheng Zhao, Rui Zhou, Feng Yin
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 |
|---|---|---|---|
| sample independent directions across radii instead of sharing one direction (ablation) | Romberg-ZOGE (R=2) · Poly8/Rosenbrock directional variance | leading variance coefficient: approximately 2.22 | Hurt |
| use Romberg extrapolation (R=2) instead of two-point, spherical directions | ZO prompt tuning OPT-1.3B · SST-2 | test accuracy: 0.787 -> 0.912 | Improved |
| use Romberg extrapolation (R=2) instead of two-point, Gaussian directions | ZO prompt tuning OPT-1.3B · SST-2 | test accuracy: 0.570 -> 0.873 | Improved |
| use Romberg-ZOGE instead of P4 unbiased estimator | ZO prompt tuning OPT-1.3B · SST-2 | test accuracy: 0.877 -> 0.912 | Improved |
| use kernel-based estimator instead of two-point | ZO prompt tuning OPT-1.3B · SST-2 | test accuracy: 0.787 -> 0.552 | Hurt |
| use spherical instead of Gaussian directions for Romberg-ZOGE (ablation) | ZO prompt tuning OPT-1.3B · SST-2 | test accuracy: 0.873 -> 0.912 | Improved |
| use Romberg-ZOGE (R=2) instead of two-point in ZO-SGD with equal evaluations per iteration | ZO-SGD · Poly8 and Rosenbrock (n=100) | objective value | Improved |
| use Romberg-ZOGE (R=2) instead of two-point under equal evaluations per iteration | ZO optimization · wireless network smooth-quantile SE | objective value | Improved |
| increase extrapolation order R under a stochastic oracle (ablation) | Romberg-ZOGE · stochastic oracle theory | bias and noise amplification | Mixed |
Closest papers by meaning
- Revisiting Zeroth-Order Optimization: Minimum-Variance Two-Point Estimators and Directionally Aligned Perturbations (ICLR 2025 · similarity 0.90)
- On the Optimal Construction of Unbiased Gradient Estimators for Zeroth-Order Optimization (NeurIPS 2025 · similarity 0.90)
- Refining Adaptive Zeroth-Order Optimization at Ease (ICML 2025 · similarity 0.89)
- Double Variance Reduction: A Smoothing Trick for Composite Optimization Problems without First-Order Gradient (ICML 2024 · similarity 0.88)
- Nonasymptotic Analysis of Stochastic Gradient Descent with the Richardson–Romberg Extrapolation (ICLR 2025 · similarity 0.88)
- Adaptively Incorporating Directional Hints into Zeroth-Order Optimization (NeurIPS 2026 · 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.
