Research map / Reinforcement learning / Offline RL
OptiDICE: Offline Policy Optimization via Stationary Distribution Correction Estimation
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 |
|---|---|---|---|
| Use OptiDICE instead of the strongest listed baseline | D4RL offline policy optimization · maze2d-umaze | normalized performance: 88.2 -> 111.0 | Helped |
| Use OptiDICE instead of the strongest listed baseline | D4RL offline policy optimization · maze2d-medium | normalized performance: 33.8 -> 145.2 | Helped |
| Use OptiDICE instead of the strongest listed baseline | D4RL offline policy optimization · maze2d-large | normalized performance: 40.6 -> 155.7 | Helped |
| Use OptiDICE instead of the strongest listed baseline | D4RL offline policy optimization · hopper-medium | normalized performance: 58.0 -> 94.1 | Helped |
| Use OptiDICE instead of the strongest listed baseline | D4RL offline policy optimization · hopper-random | normalized performance: 12.2 -> 11.2 | Hurt |
| Use OptiDICE instead of the strongest listed baseline | D4RL offline policy optimization · hopper-medium-replay | normalized performance: 48.6 -> 36.4 | Hurt |
| Use OptiDICE instead of the strongest listed baseline | D4RL offline policy optimization · walker2d-random | normalized performance: 7.3 -> 9.9 | Helped |
| Use OptiDICE instead of the strongest listed baseline | D4RL offline policy optimization · walker2d-medium | normalized performance: 81.1 -> 21.8 | Hurt |
| Use OptiDICE instead of the strongest listed baseline | D4RL offline policy optimization · halfcheetah-medium-expert | normalized performance: 64.7 -> 91.1 | Helped |
| Use the MSE objective rather than the minimax objective to train the correction model (ablation) | OptiDICE · D4RL continuous-control tasks | normalized performance | No clear effect |
| Use importance-weighted behavioral cloning rather than I-projection for policy extraction (ablation) | OptiDICE continuous-domain policy extraction · D4RL continuous-control tasks | policy performance | Hurt |
| Add the normalization constraint to stationary-distribution correction estimation (ablation) | OptiDICE · offline RL, including γ<1 | learning stability | Helped |
Closest papers by meaning
- A Minimalist Approach to Offline Reinforcement Learning (NeurIPS 2021 · similarity 0.94)
- COptiDICE: Offline Constrained Reinforcement Learning via Stationary Distribution Correction Estimation (ICLR 2022 · similarity 0.93)
- Relaxed Stationary Distribution Correction Estimation for Improved Offline Policy Optimization (AAAI 2024 · similarity 0.93)
- Regularizing a Model-based Policy Stationary Distribution to Stabilize Offline Reinforcement Learning (ICML 2022 · similarity 0.93)
- MOPO: Model-based Offline Policy Optimization (NeurIPS 2020 · similarity 0.92)
- A Clean Slate for Offline Reinforcement Learning (NeurIPS 2025 · similarity 0.91)
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.