Research map / Reinforcement learning / Visual and representation RL
Provably Efficient Online Hyperparameter Optimization with Population-Based Bandits
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 |
|---|---|---|---|
| replace PBT random/greedy exploration with GP-bandit selection (PB2) | PBT · BipedalWalker B=4 | median best reward: 223 -> 276 (+24%) | Helped |
| replace PBT exploration with PB2 | PBT · LunarLanderContinuous B=4 | median best reward: 159 -> 235 (+48%) | Helped |
| replace PBT exploration with PB2 | PBT · Hopper B=4 | median best reward: 1492 -> 2346 (+57%) | Helped |
| replace PBT exploration with PB2 | PBT · InvertedDoublePendulum B=4 | median best reward: 8893 -> 8179 (-8%) | Hurt |
| replace PBT exploration with PB2 at larger population | PBT · BipedalWalker B=8 | median best reward: 277 -> 291 (+5%) | Helped |
| replace PBT exploration with PB2 at larger population | PBT · LunarLanderContinuous B=8 | median best reward: 247 -> 275 (+11%) | Helped |
| use PB2 when hyperparameter range is mis-specified (batch size 5000-200000) | PBT · BipedalWalker | median best reward: -12 -> 203 | Helped |
| use PB2 instead of PBT for supervised CNN tuning | PBT · CIFAR-10 | test accuracy: 87.20 -> 89.10 | Helped |
| use PB2 instead of ASHA | ASHA · CIFAR-10 | test accuracy: 88.85 -> 89.10 | Helped |
| small population (B=4) PBT vs random search | PBT · PPO control tasks | median best reward: e.g. BipedalWalker 223 vs 234 RS | Hurt |
| increase population size from 4 to 8 (ablation) | PB2 · BipedalWalker | median best reward: 276 -> 291 | Helped |
Closest papers by meaning
- Hyper: Hyperparameter Robust Efficient Exploration in Reinforcement Learning (ICML 2025 · similarity 0.86)
- Offline-to-Online Hyperparameter Transfer for Stochastic Bandits (AAAI 2025 · similarity 0.86)
- Adaptive Learn-then-Test: Statistically Valid and Efficient Hyperparameter Selection (ICML 2025 · similarity 0.86)
- A Method for Evaluating Hyperparameter Sensitivity in Reinforcement Learning (NeurIPS 2024 · similarity 0.85)
- Parallelizing Model-based Reinforcement Learning Over the Sequence Length (NeurIPS 2024 · similarity 0.85)
- The Power of Resets in Online Reinforcement Learning (NeurIPS 2024 · similarity 0.85)
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.