Research map / Reinforcement learning / Offline RL
A Unified Framework for Alternating Offline Model Training and Policy Learning
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 |
|---|---|---|---|
| remove MIW-weighted model retraining (NW) (ablation) | AMPL · D4RL average (16 datasets) | average normalized return: 69.1 -> 83.8 | Helped |
| remove MIW weighting on Maze2D (ablation) | AMPL · D4RL Maze2D average | average normalized return: 82.7 -> 114.3 | Helped |
| replace MIW estimator with VPM (ablation) | AMPL · D4RL average (16 datasets) | average normalized return: 62.8 -> 83.8 | Helped |
| replace MIW estimator with GenDICE (ablation) | AMPL · D4RL average (16 datasets) | average normalized return: 67.8 -> 83.8 | Helped |
| replace MIW estimator with DualDICE (ablation) | AMPL · D4RL average (16 datasets) | average normalized return: 63.4 -> 83.8 | Helped |
| weight the policy regularizer by MIW (WPR) instead of unweighted (ablation) | AMPL · D4RL average (16 datasets) | average normalized return: 83.8 -> 46.6 | Hurt |
| use KL-dual regularizer instead of JSD (ablation) | AMPL · D4RL average (16 datasets) | average normalized return: 83.8 -> 9.7 | Hurt |
| use weighted KL (KL-Dual+WPR) instead of JSD (ablation) | AMPL · D4RL average (16 datasets) | average normalized return: 83.8 -> 10.2 | Hurt |
| use Gaussian policy instead of implicit policy (ablation) | AMPL · D4RL average (16 datasets) | average normalized return: 83.8 -> 59.5 | Hurt |
| remove the proposed policy regularizer (ablation) | AMPL · D4RL average (16 datasets) | average normalized return: 83.8 -> 16.6 | Hurt |
| remove model-rollout data from policy learning (ablation) | AMPL · D4RL average (16 datasets) | average normalized return: 83.8 -> 63.6 | Hurt |
| use estimated reward function as MIW test function instead of action-value (ablation) | AMPL · D4RL average (16 datasets) | average normalized return: 83.8 -> 73.0 | Hurt |
Closest papers by meaning
- MOPO: Model-based Offline Policy Optimization (NeurIPS 2020 · similarity 0.93)
- Policy-Driven World Model Adaptation for Robust Offline Model-based Reinforcement Learning (ICML 2026 · similarity 0.92)
- Model-Based Offline Planning (ICLR 2021 · similarity 0.91)
- Pareto Policy Pool for Model-based Offline Reinforcement Learning (ICLR 2022 · similarity 0.91)
- A Clean Slate for Offline Reinforcement Learning (NeurIPS 2025 · similarity 0.91)
- OCEAN-MBRL: Offline Conservative Exploration for Model-Based Offline Reinforcement Learning (AAAI 2024 · similarity 0.90)
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.