Research map / Reinforcement learning / Offline RL
Adversarial Counterfactual Environment Model 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 |
|---|---|---|---|
| Use GALILEO instead of standard supervised model learning | Environment dynamics models · Counterfactual queries under selection bias | MISE | Helped |
| Use GALILEO instead of IPW | Environment dynamics models · GNFC with e=0.05, p<=0.2 | MISE | Helped |
| Use GALILEO instead of SCIGAN | Environment dynamics models · Synthetic counterfactual tasks | MISE | Mixed |
| Use GALILEO instead of OFF-SL | MuJoCo environment models · Expert and medium-replay datasets | Prediction error | Helped |
| Use GALILEO instead of Best DICE | Off-policy evaluation · DOPE benchmark (three-task average) | Norm. value gap: 0.48 $\pm$ 0.19 -> 0.37 $\pm$ 0.24 | Helped |
| Use GALILEO instead of Best DICE | Off-policy evaluation · DOPE benchmark (three-task average) | Rank corr.: 0.15 $\pm$ 0.04 -> 0.44 $\pm$ 0.10 | Helped |
| Use GALILEO instead of VPM | Off-policy evaluation · DOPE benchmark (three-task average) | Regret@1: 0.17 $\pm$ 0.11 -> 0.09 $\pm$ 0.02 | Helped |
| Use GALILEO instead of OFF-SL | Model-based policy improvement · MuJoCo (nine tasks) | avg. norm.: 21.1 -> 51.1 | Helped |
| Use GALILEO instead of SCIGAN | Model-based policy improvement · MuJoCo (nine tasks) | avg. norm.: 41.8 -> 51.1 | Helped |
| Use GALILEO instead of IPW | Model-based policy improvement · HalfCheetah (H=40) | Return: -11.3 $\pm$ 0.9 -> -14.2 $\pm$ 1.4 | Hurt |
| Use GALILEO-learned model for MPC instead of the behavior policy | Online policy in the food-delivery platform · Six cities | Five-minute order-taken rate: 0.14 to 1.63 percentage points | Helped |
| Use standard SL instead of GALILEO | Food-delivery supply model · Real-world food-delivery platform | AUUC | Hurt |
Closest papers by meaning
- Constrained Latent Action Policies for Model-Based Offline Reinforcement Learning (NeurIPS 2024 · similarity 0.86)
- Off-Policy Actor-Critic for Adversarial Observation Robustness: Virtual Alternative Training via Symmetric Policy Evaluation (ICML 2025 · similarity 0.86)
- Learning from Reward-Free Offline Data: A Case for Planning with Latent Dynamics Models (NeurIPS 2025 · similarity 0.86)
- Policy-conditioned Environment Models are More Generalizable (ICML 2024 · similarity 0.86)
- Adversarial Diffusion for Robust Reinforcement Learning (NeurIPS 2025 · similarity 0.86)
- Adversarial Environment Design via Regret-Guided Diffusion Models (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.