Research map / Probabilistic and scientific modeling / Bayesian optimization
Local Entropy Search over Descent Sequences for Bayesian Optimization
David Stenger, Armin Lindicke, Alexander von Rohr, Sebastian Trimpe
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 |
|---|---|---|---|
| target descent sequence of local optimizer (LES) instead of trust region (TuRBO) | TuRBO · GP samples within-model, high complexity, d=50 | median final objective: -7.1 -> -10.8 | Improved |
| use LES instead of global logEI | logEI · GP samples within-model, high complexity, d=50 | median final objective: -2.9 -> -10.8 | Improved |
| use LES instead of logEI at low dimension | logEI · GP samples within-model, high complexity, d=5 | median final objective: -3.9 vs -2.8 | Hurt |
| use LES instead of logEI | logEI · GP samples within-model, medium complexity, d=50 | median final objective: -9.6 -> -10.4 | Improved |
| use LES instead of logEI | logEI · GP samples within-model, low complexity, d=50 | median final objective: -8.5 -> -8.8 | No clear effect |
| use LES instead of logEI | logEI · GP samples within-model, extremely low complexity, d=50 | median final objective: -3.0 -> -3.0 | No clear effect |
| use LES instead of TuRBO under MAP hyperparameters | TuRBO · GP samples out-of-model, high complexity, d=50 | median final objective: -7.1 -> -7.8 | Improved |
| use LES instead of logEI-DSP under MAP hyperparameters | logEI-DSP · GP samples out-of-model, low complexity, d=50 | median final objective: -8.1 vs -8.5 for LES: LES better | Hurt |
| use wrong, less complex hyperprior (logEI-DSP vs logEI) | logEI · GP samples out-of-model, high complexity, d=50 | median final objective: -2.9 -> -4.1 | Improved |
| run LES on rover and Mopta08 tasks | LES · Mopta08 | regret | Mixed |
| use LES with higher L and P (more accurate approximation) | LES · GP samples | performance | Improved |
| use local Thompson sampling or condition only on final point instead of full descent sequences | LES · GP samples | performance | Hurt |
Closest papers by meaning
- Local Bayesian optimization via maximizing probability of descent (NeurIPS 2022 · similarity 0.91)
- Optimization, fast and slow: optimally switching between local and Bayesian optimization (ICML 2018 · similarity 0.90)
- A Unified Framework for Entropy Search and Expected Improvement in Bayesian Optimization (ICML 2025 · similarity 0.88)
- Minimizing UCB: a Better Local Search Strategy in Local Bayesian Optimization (NeurIPS 2024 · similarity 0.88)
- Optimizing Posterior Samples for Bayesian Optimization via Rootfinding (ICLR 2025 · similarity 0.87)
- Batched Energy-Entropy acquisition for Bayesian Optimization (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.
