Research map / Optimization / Optimizers
Local Bayesian optimization via maximizing probability of descent
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 the MPD acquisition and most-probable-descent update together instead of GIBO | Local Bayesian optimization · Swimmer-v1, cosmological constant, rover trajectory | terminal objective value: 348.88 -> 360.50; -55.25 -> -23.97; 152.77 -> 89.89 | Helped |
| Use trace-based gradient learning instead of the MPD acquisition while retaining MPD movement (ablation) | MPD · Swimmer-v1 | terminal objective value: 360.50 -> 350.58 | Hurt |
| Use trace-based gradient learning instead of the MPD acquisition while retaining MPD movement (ablation) | MPD · Cosmological constant | terminal objective value: -23.97 -> -27.72 | Hurt |
| Use trace-based gradient learning instead of the MPD acquisition while retaining MPD movement (ablation) | MPD · Rover trajectory | terminal objective value: 89.89 -> 84.17 | Helped |
| Move along the negative expected gradient instead of the most probable descent direction (ablation) | MPD · Swimmer-v1 | terminal objective value: 360.50 -> 340.12 | Hurt |
| Move along the negative expected gradient instead of the most probable descent direction (ablation) | MPD · Cosmological constant | terminal objective value: -23.97 -> -21.24 | Helped |
| Move along the negative expected gradient instead of the most probable descent direction (ablation) | MPD · Rover trajectory | terminal objective value: 89.89 -> 293.08 | Hurt |
| Lower the descent probability threshold from 65% to 50% (ablation) | MPD · Rover trajectory | terminal objective value: 89.89 -> 51.48 | Helped |
| Lower the descent probability threshold from 65% to 50% (ablation) | MPD · Swimmer-v1 | terminal objective value: 360.50 -> 342.36 | Hurt |
| Raise the descent probability threshold to 85% or 95% (ablation) | MPD · Swimmer-v1 | terminal objective value: 360.50 -> 294.67; 360.50 -> 15.97 | Hurt |
| Increase step size to 10⁻² (ablation) | MPD · Swimmer-v1 | terminal objective value: 360.50 -> 350.15 | Hurt |
| Disable state-normalization updates during inner learning (ablation) | MPD and GIBO · Hopper-v1 | terminal optimized reward: GIBO: 2827.96 -> 1032.97; MPD: 2199.48 -> 1398.35 | Mixed |
Closest papers by meaning
- Minimizing UCB: a Better Local Search Strategy in Local Bayesian Optimization (NeurIPS 2024 · similarity 0.93)
- Understanding High-Dimensional Bayesian Optimization (ICML 2025 · similarity 0.90)
- A General Framework for User-Guided Bayesian Optimization (ICLR 2024 · similarity 0.89)
- Optimizing the Unknown: Black Box Bayesian Optimization with Energy-Based Model and Reinforcement Learning (NeurIPS 2025 · similarity 0.89)
- Informed Initialization for Bayesian Optimization and Active Learning (NeurIPS 2025 · similarity 0.89)
- BayeSQP: Bayesian Optimization through Sequential Quadratic Programming (NeurIPS 2025 · similarity 0.89)
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.