Research map / Reinforcement learning / Offline RL
Efficient Offline Policy Optimization with a Learned Model
Zichen Liu, Siyi Li, Wee Sun Lee, Shuicheng YAN, Zhongwen Xu
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 |
|---|---|---|---|
| replace MCTS policy improvement with regularized one-step look-ahead | MuZero Unplugged · RL Unplugged Atari (12 games) | IQM normalized score: 151% -> 194% | Improved |
| replace MCTS with one-step look-ahead | MuZero Unplugged · RL Unplugged Atari (12 games), reaching 150% IQM normalized score | wall-clock time: 5.6% | Improved |
| use ROSMO instead of MZU | offline RL agents · Catch, MountainCar, and Cartpole | episode return | Mixed |
| use one-step look-ahead rather than multi-step MCTS planning | MZU · BSuite with stochastic transitions | episode return: 0.980 ± 0.060 -> 1.0 ± 0.0; 0.984 ± 0.054 -> 1.0 ± 0.0; 0.772 ± 0.253 -> 0.900 ± 0.237; 0.404 ± 0.422 -> 0.916 ± 0.139 | Improved |
| use ROSMO instead of MZU in the low-data regime | offline RL agents · BSuite, sub-sampled datasets | IQM normalized score | Improved |
| use ROSMO instead of MZU across dynamics model capacities | offline RL agents · BSuite, 10% data | IQM normalized score | Improved |
| limit MZU search depth | MZU · BSuite | IQM normalized score | Improved |
| reduce MZU simulation budget | MZU · BSuite | IQM normalized score | Hurt |
| use deep MZU search with high simulation budget | MZU · BSuite | IQM normalized score | Hurt |
| replace standard MZU policy targets with MPO-style Q-based policy updates (ablation) | MZU · MsPacman with 4 simulations | IQM normalized score | Improved |
| sample actions for policy improvement under a limited sampling budget (ablation) | ROSMO · MsPacman with 4 samples | IQM normalized score | Improved |
| reduce model unroll length from 5 to 1 (ablation) | ROSMO · MsPacman | IQM normalized score | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- COMBO — COMBO: Conservative Offline Model-Based Policy Optimization (NeurIPS 2021) · better on 1 of 1 shared result
- MOReL — MOReL: Model-Based Offline Reinforcement Learning (NeurIPS 2020) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Online and Offline Reinforcement Learning by Planning with a Learned Model (NeurIPS 2021 · similarity 0.91)
- MOPO: Model-based Offline Policy Optimization (NeurIPS 2020 · similarity 0.89)
- SUMO: Search-Based Uncertainty Estimation for Model-Based Offline Reinforcement Learning (AAAI 2025 · similarity 0.87)
- OCEAN-MBRL: Offline Conservative Exploration for Model-Based Offline Reinforcement Learning (AAAI 2024 · similarity 0.87)
- Model Selection for Off-policy Evaluation: New Algorithms and Experimental Protocol (NeurIPS 2025 · similarity 0.87)
- Efficient Policy Evaluation with Offline Data Informed Behavior Policy Design (ICML 2024 · similarity 0.87)
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.
