Research map / Reinforcement learning / Policy gradient and actor-critic
Model-based Policy Optimization with Unsupervised Model Adaptation
Jian Shen, Han Zhao, Weinan Zhang, Yong Yu
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 unsupervised feature-distribution adaptation with Wasserstein-1 distance to MBPO | MBPO / AMPO · Six MuJoCo continuous-control tasks | sample efficiency | Improved |
| Add unsupervised feature-distribution adaptation with Wasserstein-1 distance to MBPO | MBPO / AMPO · Six MuJoCo continuous-control tasks | asymptotic return | Improved |
| Add unsupervised model adaptation | MBPO dynamics model · One-step model prediction | model loss | Improved |
| Add unsupervised model adaptation | MBPO dynamics model · Hopper | multi-step compounding error | Improved |
| Use MMD rather than Wasserstein-1 distance for feature alignment (ablation) | AMPO · Three continuous-control environments | return | Improved |
| Increase the number of model-adaptation updates (ablation) | AMPO · Hopper | return | Mixed |
| Use longer rollout lengths earlier in training (ablation) | AMPO and MBPO · Hopper | return | Mixed |
| Increase the number of policy updates per environment step (ablation) | AMPO and MBPO · Continuous-control environments | return | Mixed |
| Continue model adaptation throughout training instead of early stopping (ablation) | AMPO · Hopper and other continuous-control environments | return | No clear effect |
| Share feature-extractor weights across real and simulated data (ablation) | AMPO · Continuous-control environments | return | No clear effect |
| Fix the real-data feature extractor rather than learning separate real and simulated extractors (ablation) | AMPO · Continuous-control environments | return | No clear effect |
Closest papers by meaning
- Adaptation Augmented Model-based Policy Optimization (JMLR 2023 · similarity 0.94)
- When to Trust Your Model: Model-Based Policy Optimization (NeurIPS 2019 · similarity 0.89)
- Bidirectional Model-based Policy Optimization (ICML 2020 · similarity 0.88)
- Control-Oriented Model-Based Reinforcement Learning with Implicit Differentiation (AAAI 2022 · similarity 0.88)
- PWM: Policy Learning with Multi-Task World Models (ICLR 2025 · similarity 0.87)
- The Edge-of-Reach Problem in Offline Model-Based Reinforcement Learning (NeurIPS 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.
