Research map / Architectures and efficiency / RNNs and normalization
Depth over Fidelity in Fixed-Budget Noisy Evolution Strategies
Sichen Wang, Zhipeng Lu
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 Probe-and-Switch instead of 10-fold resampling | CMA-ES · All COCO bbob-noisy + 6 external tasks | win rate: 85.6% | Improved |
| use Probe-and-Switch instead of 5-fold resampling | CMA-ES · All COCO bbob-noisy + 6 external tasks | win rate: 79.3% | Improved |
| use Probe-and-Switch instead of UH-CMA-ES | CMA-ES · All COCO bbob-noisy + 6 external tasks | win rate: 77.4% | Improved |
| add Probe-and-Switch over vanilla CMA-ES | CMA-ES · All COCO bbob-noisy + 6 external tasks | win rate: 54.3% | Improved |
| add probe-and-switch to RB-PEM (revert to CMA-ES in low-misranking regimes) (ablation) | RB-PEM · All COCO bbob-noisy + 6 external tasks | win rate: 58.1% | Improved |
| add Probe-and-Switch vs CMA-ES, COCO d=10 | CMA-ES · COCO bbob-noisy d=10 | wins: 230/210 | No clear effect |
| add Probe-and-Switch vs CMA-ES, COCO d=40 | CMA-ES · COCO bbob-noisy d=40 | wins: 257/170 | Improved |
| add Probe-and-Switch vs CMA-ES on Digits | CMA-ES · Digits | wins: 17/33 | Hurt |
| add Probe-and-Switch vs CMA-ES on LQR | CMA-ES · LQR | wins: 31/19 | Improved |
| use conservative threshold 0.22 instead of 0.12 (ablation) | Probe-and-Switch · external tasks (HPO, Pendulum) | regret reduction: tau 0.12 -> 0.22 | Mixed |
Closest papers by meaning
- Variance-Reduced Gradient Estimation via Noise-Reuse in Online Evolution Strategies (NeurIPS 2023 · similarity 0.86)
- Evolution Strategies at the Hyperscale (ICML 2026 · similarity 0.86)
- Diffusion Models are Evolutionary Algorithms (ICLR 2025 · similarity 0.86)
- Evolution Fine-Tuning: Learning to Discover Across 371 Optimization Tasks (NeurIPS 2026 · similarity 0.84)
- EvoOptiGraph: Weakness-Driven Coevolution via Graph-Based Structural Generation for Optimization Modeling (NeurIPS 2026 · similarity 0.83)
- EVA-0: Test-Time Model Evolution with Only Two Forward Passes per Sample (NeurIPS 2026 · similarity 0.83)
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.
