Research map / Probabilistic and scientific modeling / Bayesian optimization
Bayesian Optimization for Policy Search via Online-Offline Experimentation
Benjamin Letham, Eytan Bakshy
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 |
|---|---|---|---|
| add simulator observations via multi-task GP to online-only GP | GP for online value model outcomes · 99 outcomes from 11 News Feed experiments (20 online, 100 simulator observations) | leave-one-out MSE | Improved |
| use shared kernel fixed from simulator data instead of kernel inferred from online data only (ablation) | single-task GP · empirical learning curves | prediction error | Improved |
| add multi-task modeling beyond offline kernel when inter-task correlation is medium or high (ablation) | single-task GP with offline kernel · empirical learning curves | prediction error | Improved |
| add simulator observations when inter-task correlation is low | MTGP · learning curves, rho^2=0.09 | prediction error | Mixed |
| add simulator observations when inter-task correlation is high | MTGP · learning curves, rho^2=0.8 | prediction error: 2 online observations combined with 90 simulator observations produce a better model than 18 online observations | Improved |
| use offline observations as pre-test before online batches (interleaved simulator batches) rather than only in initialization (ablation) | MTGP Bayesian optimization · synthetic online-offline problem, 30 runs | best feasible objective | Improved |
| add MTGP with simulator to single-task BO | Bayesian optimization · synthetic online-offline problem, 30 runs | best feasible objective | Improved |
Closest papers by meaning
- Bayesian Optimization from Human Feedback: Near-Optimal Regret Bounds (ICML 2025 · similarity 0.87)
- The Importance of Online Data: Understanding Preference Fine-tuning via Coverage (NeurIPS 2024 · similarity 0.86)
- Information-Directed Offline-to-Online Reinforcement Learning (NeurIPS 2026 · similarity 0.86)
- Probabilistic Offline Policy Ranking with Approximate Bayesian Computation (AAAI 2024 · similarity 0.86)
- Data Mixture Optimization: A Multi-fidelity Multi-scale Bayesian Framework (NeurIPS 2025 · similarity 0.86)
- Online Learning with Sublinear Best-Action Queries (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.
